RE

VIE

W

JOURNAL OF NANOSCIENCE AND NANOTECHNOLOGY

Metal Oxide Nanoarchitecturesfor Environmental Sensing

Oomman K Varghese and Craig A Grimescurren

Department of Electrical Engineering and Materials Research Institute217 Materials Research Laboratory Pennsylvania State University

University Park Pennsylvania USA

Metal oxide materials are widely used for gas sensing Capable of operating at elevated temper-atures and in harsh environments they are mechanically robust and relatively inexpensive andoffer exquisite sensing capabilities the performance of which is dependent upon the nanoscalemorphology In this paper we rst review different routes for the fabrication of metal oxide nanoarchi-tectures useful to sensing applications including mesoporous thin lms nanowires and nanotubesTwo sensor test cases are then presented The rst case examines the use of highly uniformnanoporous Al2O3 for humidity sensing we nd that such materials can be successfully used asa wide-range humidity sensor The second test case examines the use of TiO2 nanotubes forhydrogen sensing Going from a nitrogen atmosphere to one containing 1000 ppm of hydrogen at290 C 22-nm-diameter titania nanotubes demonstrate a 104 change in measured resistance withno measurement hysteresis

Keywords Nanotubes Nanowires Metal Oxide Gas Sensing Hydrogen Humidity Sensor

1 INTRODUCTION

As illustrated by Figure 1 it is necessary to simultane-ously solve several design variables to achieve a usefulsensing device Often these design variables which cer-tainly include but are not limited to cost size and dura-bility are contradictory in nature Perhaps a starting pointin considering a sensor platform is the transduction mech-anism do we seek for example to detect changes inelectrical impedance electrical phase magnetic proper-ties frequency elasticity or mass Is the material beingused as a sensor selective for the target species or does itrequire an e-nose approach Once a target sensor materialis identi ed operational issues that must be determinedinclude sensitivity dynamic range resolution hysteresisfatigue and drift Materials investigated for sensing utilitymay possess one or more favorable attributes but be unus-able because of a severe limitation in another Further-more since there appears to be no universally applicableperfect sensor each sensing application generally requiresthe bottom-up design of a suitable sensor Hence consid-erable effort has historically been spent in both broad andspeci c development of sensor technologies and it is still

currenAuthor to whom correspondence should be addressed

an area of vital importance and interest to the medicalmanufacturing environmental and defensesecurity com-munities

Chemical or biological sensors outside of a controlledclean environment immediately discover that it is a dirtyworld out there with dust dirt diesel fumes smoke sootbiologicals and what have you ready to foul and degradeoperational capabilities Not only do sensors get dirtythe more sensitive they are the more susceptible they areto insult and fouling Although we seek to develop self-cleaning sensors our immediate interest lies in develop-ing exquisitely sensitive gas-sensing materials inexpensiveenough to be used on a disposable basis Metal oxidematerials such as SnO2 Al2O3 and TiO2 can be hadat relatively low cost and have long been recognized fortheir outstanding gas-sensing properties Furthermore it isgenerally now recognized that nanoscale control of metaloxide surface morphologies permits signi cant enhance-ment of gas-sensing properties Hence we have focusedour efforts as described here at investigating and con-trolling the gas-sensing properties of metal oxide nanoar-chitectures to successfully create an exquisitely sensitivegas sensor that when needed can readily and practicallybe used on a disposable basis

J Nanosci Nanotech 2003 Vol 3 No 4 copy 2003 by American Scienti c Publishers 1533-4880200304277293$1700+25 doi101166jnn2003158 277

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Stimulus (s)

Electrical Signal S(s)

Circuit Design Signal Processing

Materials Design -useful transduction mechanism

-chemistry -nano to microstructure

Device Design

Fig 1 Schematic drawing of design variables inherent in the fabrica-tion of a useful sensor

2 FABRICATION OF METAL OXIDENANOARCHITECTURES

21 Mesoporous Thin Films

Nanoarchitectured thin lms are of considerable interestfor applications including photocatalysis and sensing andas templates for cell growth1ndash6 Apart from the conven-tional top-down engineering approach of translating a con-structed pattern onto a substrate by lithography7 or softlithography8 self-assembly and self-organization of mate-rials offer a rapid fabrication route at low cost Consider-able effort has focused on the use of a template aroundwhich the material of interest is assembled Depend-ing on the required pore size block copolymers91 10

latex spheres11ndash13 water-in-oil emulsions14 polystyreneparticles151 16 colloidal crystals17ndash22 and bioskeletons23ndash25

have been used as templates Problems associated withtemplate-assisted fabrication of porous structures includepreparation of a high-quality template complete lling ofthe voids in the template and minimization of shrinkageupon template removal Since any of these factors can

Oomman K Varghese received his PhD degree in physics from the Indian Institute of Technology Delhi India in2001 Since graduation he has worked as a postdoctoral fellow in the research group of Professor Grimes Departmentof Electrical Engineering and the Materials Science and Engineering Department the Pennsylvania State University His elds of interest include gas sensors based on metal oxide nanodimensional structures and carbon nanotubes impedancespectroscopy and transmission electron microscopy for materials characterization grain boundary space charge segregationeffects in binary oxides and sol-gel synthesis of metal oxide thin lms

Craig A Grimes received BS degrees in electrical engineering and physics from the Pennsylvania State University in1984 and the MS and PhD degrees in electrical engineering from the University of Texas at Austin in 1986 and 1990respectively He was employed by Lockheed Research Laboratories from 1990 to 1994 From 1994 to 2001 Dr Grimeswas a member of the Electrical and Computer Engineering Department at the University of Kentucky where he wasthe Frank J Derbyshire Professor He is currently an associate professor at the Pennsylvania State University in theDepartment of Electrical Engineering and Materials Science and Engineering His research interests include remote querychemical and environmental sensors nanodimensional metal-oxide thin- lm architectures for sensing and biotemplatingpropagation and control of electromagnetic energy and carbon nanotube-based electronic devices He has contributed over120 archival journal publications and seven book chapters and is co-author of the book The Electromagnetic Origin ofQuantum Theory and Light published by the World Scienti c Publishing Company (2002) He is North American editorof Sensors and an editorial board member of IEEE Transactions on Magnetics

in uence the nal quality of the porous structure all threerequirements must be ful lled at the same time

A nontemplate method for the synthesis of orderedmicrometer-sized honeycomb structures by self-assemblyof block copolymers was reported by Francois et al26ndash28

and Jenekhe and co-workers29 Shimomura et al301 31

helped pioneer efforts to fabricate patterned thin lmsby self-assembly An organic chloroform solution withamphiphiles containing metal acetylacetonates or alkox-ides cast at high atmospheric humidity was found toform a closely packed layer of water droplets on topof the organic solvent with the water droplets acting asa template301 31 after evaporation of the chloroform andwater a honeycomb structure remains Finally the pyrol-ysis of the metal alkoxide lm leads to the formationof microporous metal oxide such as anatase with poresize and wall thickness of the resulting lm controlledby solution concentration and ambient humidity level Ina similar vein Nishikawa and co-workers32 reported thefabrication of porous lms by casting of a dilute solu-tion of amphiphillic polymers onto solid substrates at highhumidity levels

Several sol-gel methods have been employed to fab-ricate patterned TiO2 lms typically by the incorpora-tion of organic polymers with the precursor solution toobtain precisely controlled macroscopic structures Tat-suma et al15 for example reported the fabrication ofmicroporous TiO2 lms prepared with the use of a two-dimensional array of polystyrene microspheres as a tem-plate A TiO2 aqueous sol was introduced into the gapbetween the polystyrene spheres upon the evaporationof water the remaining lms were calcined to incineratethe polystyrene particles leaving a porous TiO2 lmIn an analogous approach Kajihara et al33 reported the

278

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

fabrication of porous TiO2 lms with the use of analkoxide-based sol-gel containing polyethylene glycolwhich served as the template into which the substrateswere dip coated

A disadvantage of using a template is that the dom-inant length scale of the resulting porous structure is xed by the template size therefore dynamic control ofthe length scale becomes almost impossible Our inter-est lies in the fabrication of nanoporous lms with-out the use of a template for example we recentlyreported the fabrication of metal oxide lms with magnet-ically modulated nanodimensionalporosity34 The absenceof a template makes the ambient humidity level andthe sol pH key process variables for the fabrication ofcontrolled structures35 We describe here the fabricationof TiO2 mesoporous structures via sol-gel breath gureformation36ndash48 as well as thermocapillary and surfacetension-driven Benard-Marangoni convection49ndash56

The overall hydrolysis and condensation reaction fortitanium isopropoxide the sol-gel precursor for TiO2 canbe represented as Ti(OC3H7)4 C 2H2O TiO2 C4HOC3H7 The condensation reaction leads to the for-mation of colloidal particles which can be polymeric orparticulate depending on the type of precursors and pHof the sol Colloidal particle aggregation may be inhib-ited by the formation of surface charge developed byeither preferential dissociation of one of the lattice ionsof the sol particle or preferential adsorption of chargedspecies from solution The surface charge is formed byeither protonation (Ti-OH C HC Ti-OHC

2 5 or depro-tonation (Ti-OH C OHƒ Ti-Oƒ C H2O) of the Ti-OHbonds The pH at which the surface is electrically neutralis called the point of zero charge (PZC) The surface isnegatively charged at a pH gt PZC and positively chargedat pH lt PZC For TiO2 the PZC varies at values at leastbetween 52 (Ref 57) and 55 (Ref 58)

In a nitrogen environment the precursor titaniumtetraisopropoxide (TTIP) was dissolved in isopropanolto which deionized water and then nitric acid wereadded The reagents used in the experiment namely TTIP(99999) isopropanol (995) and nitric acid (70redistilled) were procured from Aldrich The solution wascontinuously stirred for 2 h and then stored in a nitrogenenvironment In a typical preparation of 01 M TiO2 sol1 ml of TTIP 005 ml of HNO3 (70 distilled) 01 ml ofdeionized water and 327 ml of isopropanol were usedThe lms were deposited by either dip coating of theglass or silicon substrate or simply by placing a dropletof solution on a clean substrate The ambient humidityin which the TiO2 lms dried was controlled by passingnitrogen through a room-temperature bubbler The humid-ity and temperature of the chamber were monitored with adigital hydrometer Upon drying all lms were annealedat 100 C for 1 h in a nitrogen environment

A 001 M sol was modi ed by a 12 nitric aciddeionizedwater solution with 15 Œl of acidwater solution added to

the original 3385 ml sol the pH of the resulting solutionwas approximately 05 Topology formation is found to bea function of (1) ambient humidity (2) atmosphere owvelocity during drying (3) sol concentration and (4) solpH Figure 2andashc illustrates the effect of drying rate on thestructure Figure 2a shows the fastest drying rate and Fig-ure 2c the slowest (the dark areas are pores ie absenceof lm) In the initial stages of the lm growth process thecondensed droplets grow as isolated objects without inter-action between neighbors as seen in Figure 2a Longerdrying times permit the drops to begin to coalesce as seenin Figure 2b with further coalescence seen in Figure 2cBreath gures36ndash48 are formed when a liquid surface is

Fig 2 Drying-time-dependent FE-SEM images of a thin lmdeposited from a 01 M TiO2 sol 1 ml of TTIP 005 ml of HNO3 (70distilled) 01 ml of deionized water and 327 ml of isopropanol towhich 15 Œl of a 12 nitric aciddeionized water solution was added(a) the fastest drying region and (c) the slowest drying region wherethe liquid has coalesced into drops

279

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

brought into contact with moist air the solvent vapor pres-sure and air velocity across the surface drive solvent evap-oration rapidly cooling the surface This in turn facili-tates the nucleation and growth of water droplets from theatmosphere36ndash41 The temperature difference between theliquid surface and air results in thermocapillary convec-tion within the liquid that acts to stabilize the condensingwater droplets on or at the solution surface43ndash48 Air owacross the surface coupled with surface convection cur-rents drives the ordering of water droplets into hexagonalarrays44ndash46 Once the surface is completely covered withwater droplets the temperature difference between the sur-face and the droplets diminishes and the droplets beingdenser than the solvent sink into the solution de ning theresidual structure Figure 3 shows the effect of sol pHshowing the pH-dependent variation in the resulting struc-ture achieved

Benard-Marangoni convection typically results from avertical temperature gradient due to solvent evaporationfrom a thin liquid lm46ndash54 Preferential solvent evapo-ration removes heat from the lm surface resulting inBenard-Marangoni convection50ndash56 across the uid layerwith the suspension welling up in the center of a convec-tion cell and then owing back down the cell boundary Asevaporation proceeds the colloidal suspension becomesmore concentrated changing the convection character-istics and in turn de ning the ultimate lm structureWell-de ned hexagonal pentagonal or square patternsare expected from an ideal time-independent Benard-Marangoni convection50 in a homogeneous liquid whereasthe cell patterns of our lms seen in Figure 2 are typicallyirregular This deviation of structure from the theoreticalideal may be the result of time-dependent convection owthrough changes in sol viscosity with preferential evapora-tion of the propanol solvent leading to coupled thermoso-lutal Benard-Marangoni convection51 in the nonhomoge-neous uid

22 Nanotube Nanopore and NanowireFabrication via Anodization of Al and Ti

Highly ordered nanoporous alumina lms (see Fig 4aandb)aremadethrougha two-stepanodizationprocess59ndash61

The aluminum substrate is rst anodized in an oxalic orsulfuric acid solution The anodization is stopped after afew microns of aluminum are consumed and the porousalumina lm is removed through etching The etchant amixture of chromic acid and phosphoric acid is highlyselective attacking alumina much faster than aluminumThe remaining aluminum is dimpled with the dimplesserving as a uniform seed layer upon which a highly uni-form porous layer can then be achieved through a sec-ond anodization step at the same voltage Alumina tem-plates are widely used as templates for the fabrication ofnanowires Figure 4c shows a TiO2 nanowire mat fab-ricated by lling pores of an alumina membrane via sol

Fig 3 Images taken of a dried 01 M TiO2 sol 1 ml of TTIP 005 mlof HNO3 (70 distilled) 01 ml of deionized water and 327 ml ofisopropanol to which varying amounts of a 12 nitric aciddeionizedwater solution were added to vary pH values (a) pH 017 (b) pH 029(c) pH 037

gel letting the lm dry and then subsequently removingthe alumina template by a sodium hydroxide etch

Anodization has also been used to fabricate TiO2 nano-tube arrays62 with pore size linearly proportional toanodization voltage (see Fig 5) The nanotube array inFigure 5 was fabricated by anodization of titanium at 20 Vin 05 wt HF solution for 20 min resulting in a well-aligned titanium oxide nanotube array with an approxi-mate average tube diameter of 60 nm and a tube lengthof 400 nm Diameters of fabricated tubes have ranged insize from 25 nm to 65 nm

280

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

Fig 4 (a) The surface of a two-step anodized alumina membrane (the akes seen in the image are the ldquodustrdquo associated with tissue paper)(b) An image of a cleaved sample showing the pore channels (c) TiO2

nanowires made by lling an alumina template via sol-gel then subse-quently removing the template by a sodium hydroxide etch resulting ina TiO2 nanowire mat

3 EXPERIMENTAL RESULTSAPPLICATION TO SENSING

31 The Water Vapor Sensing Performanceof Highly Ordered NanoporousAlumina Films

Humidity sensors have attracted considerable attentionover many years because of their great importancein applications ranging from monitoring food qual-ity to meteorological studies63ndash65 Ceramic humiditysensors651 71ndash82 are commercially available and offer majoradvantages with high resistance to chemical attack

Fig 5 TiO2 nanotubes made by anodization of titanium

thermal stability mechanical strength and quick responseHowever ceramic humidity sensors still suffer from insuf- cient sensitivity over wide humidity ranges as well aslack of reversibility and drift in base resistance with timebecause of water molecule chemisorption

The humidity-sensing properties of alumina discoveredalmost 50 years ago83ndash94 are based upon ionic conduc-tion the presence of an adsorbed layer of water at thesurface reduces the total sensor impedance because of theincrease in the ionic conductivity as well as capacitancedue to the high dielectric constant of water The humidity-sensing characteristics of a given sensing element dependupon the material used the method of preparation andthe resulting surface topology Porous alumina is usuallypreferred for sensing applications because of the largesurface area available for water adsorption

An illustration of water adsorption by a surfaceredrawn from Ref 95 is shown in Figure 6 Figure6a shows the adsorption of a water molecule on aclean alumina surface Physisorption of water initiallyoccurs on an active surface site and forms an adsorption

Fig 6 Schematic illustration of interaction of a water molecule withan alumina surface redrawn from Ref 95

281

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Fig 7 Schematic illustration of buildup of adsorbed water layersredrawn from Ref 96 building of a water layer upon an alumina surface

complex (Fig 6a and b) The negatively charged oxygenof the water molecule becomes attached to the positivelycharged cation site and one of the positively chargedhydrogen atoms becomes attached to the anionic site byelectrostatic attraction Eventually they are converted totwo chemisorbed hydroxyl (OH) groups (Fig 6c) Anincrease in humidity makes the water molecules physisorbto this chemisorbed hydroxyl layer The effectiveness ofphysisorption depends upon the cation charge complexes(from alumina or the surface impurities) and the hydroxylions on the surface of the alumina The physisorptionprocess is facilitated by higher surface charge densitiesAfter hydroxyl formation the next water molecule willbe physisorbed through hydrogen double bonds on thetwo neighbouring hydroxyl groups (Fig 6d) and a protonmay be transferred from a hydroxyl group to the watermolecule to form a H3OC ion At higher humidity levelseach water molecule will be singly bonded to a hydroxylgroup as illustrated in Figure 7 (redrawn from [96]) andproton hopping between adjacent water molecules in thecontinuous water layer takes place The conduction pro-cess in this case occurs by attachment of a proton to awater molecule thereby forming a hydronium ion thehydronium ion releases another proton to a second watermolecule that accepts this proton while releasing a thirdproton and so on through the liquid Hence the dominantcharge carrier in a high-moisture atmosphere is the HC

(proton) The concentration of HC increases with increas-ing humidity and it moves freely through the water-likelayer

In the case of porous alumina capillary condensationcan take place in pores with a radius up to rK at a partic-ular relative humidity and temperature which is given byKelvinrsquos relation

rKD 2ƒM

RT ln4Ps=P5

Here P is the water vapor pressure Ps is the water vaporpressure at saturation ƒ is the surface tension R is theuniversal gas constant T is the temperature in Kelvin

H O

H O

O

H

H +

H

H

O

H +H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H O

H O

O

H

H +

H

H

O

H +H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H O

H O

O

H

H +

H

H

O

H+H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H

OH

O

O

HH

O

H+H

H+

H

OH

OH

O

O

HH

O

H+

HH

+H

O

H

O

H

O

O

HH

O

H+ HH+ H

O

H

O

H

O

O

HH

O

H+ HH+ H

O

Aluminium

AluminaCapillary Condensation

Capillary Condensation Adsorption

Adsorbed Water Layer

Fig 8 Schematic representation of water vapor interaction withnanoporous alumina illustrating some but not all97 of the multiple fac-tors inherent in determining the response

and and M are respectively the density and molecularweight of water

According this relation the condensation occurs insidesmall pores at lower humidity levels as illustrated inFigure 8 which is an illustrative drawing presentingthe different water-surface interaction mechanisms (NoteFigure 8 is not meant to be an exhaustive considera-tion of all possible effects for example the drawingdoes not address effects associated with capillary surfacecurvature97)

This condensed water reduces the impedance of alu-mina considerably No condensation occurs inside poresof radius greater than rK at a particular humidity level (seethe largest pore shown in Fig 8) In such pores as in thecase of a plane surface it is the number of physisorbedlayers that decides the impedance

Since capillary condensation enhances the sensingcapabilities of a material pore size distribution has beenwidely considered to be an important parameter in deter-mining the sensitivity in a particular humidity range941 98

However the results of our study on uniform nanoporousalumina lms presented below show that an easily builtand highly reproducible wide-range humidity sensor canbe achieved with nanodimensional pores of a narrow sizedistribution made by anodization of aluminum59ndash611 99

Starting from adhesive-backed aluminum tape (99pure) purchased from Tesa Tape Inc100 and subsequentlyanodized sensors SO50 SO30 and DO15 were preparedunder the conditions given in Table I Per sample nota-tion the rst letter denotes a single S or double D stepanodization process the second letter denotes the use of

Table I Anodization conditions used for making nanoporous alumina lms

Anodizing Duration of No ofvoltage anodization anodization

Sample (v) (min) stages

SO50 50 120 1SO30 30 120 1DO15 15 120 2

The electrolyte used in all cases was 2 wt oxalic acid

282

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

oxalic acid (O) in the anodization bath the two-digit num-ber denotes the anodization voltage It was observed thatsingle-step anodization results in disordered pore struc-tures at low voltages Hence a double-step process wasused to make sample DO15 These samples were initiallyanodized under the conditions given in Table I The sam-ples were then immersed in an aqueous solution contain-ing 18 wt chromic acid and 4 wt phosphoric acid for12 h to remove the initial nanoporous layer while keep-ing the nonporous bottom layer of alumina intact Thesamples were rinsed in distilled water and again anodizedunder the same conditions used for the rst anodizationstep Estimates of pore size distribution were determinedfrom FE-SEM images and are shown in Figure 9andashcas a size distribution histograph Samples SO50 SO30and DO15 have respectively average pore diameters of

0

10

20

30

375 40 425 45 475 50

o

f po

res

Pore diameter (nm)

0

10

20

30

25 325 375 425 475 525

o

f po

res

0

5

10

15

20

25

175 20 225 25 275 30 325 375

DO15

o

f po

res

(a)

Pore diameter (nm)

(c)

Pore diameter (nm)

(b)

Fig 9 Pore size distribution of the tested sensors (a) SO50 (b) SO30(c) DO15

Fig 10 Digital image of an impedance-based interdigital nanoporousalumina humidity sensor

452 nm 384 nm and 136 nm with a correspondingstandard deviation of 34 nm 78 nm and 49 nm Allsamples have a predominantly ordered pore structure

Immediately after anodization all samples were thor-oughly washed in distilled water and then dried in a nitro-gen atmosphere at 100 C An interdigital capacitor pat-tern with dimensions of 12 cm pound 12 cm was formed onthe surface of the alumina lms by the evaporation of gold(thickness ordm 250 nm) through a mask A digital cameraimage of a typical sensor is shown in Figure 10 To createthe necessary ambient humidity argon passed through amass ow controller (MFC) was bubbled through a bot-tle containing deionized water and then mixed with dryargon coming from another MFC in appropriate ratiosbefore being passed to the test chamber A constant total ow was maintained throughout the experiment Cham-ber humidity was monitored with the humidity probeof a Keithley 6517A electrometer and all measurementswere carried out at room temperature (23 C) Sensorimpedance was measured over the frequency range 5 Hzto 13 MHz with a computer-controlled Hewlett Packardimpedance analyzer (4192A) tted with an impedance test xture (Agilent 16034E) A signal amplitude of 90 mVwas used for all measurements Electrical contact wasmade between the sensor under test and the impedanceanalyzer test xture with 42-gauge 5-cm-long jumpersattached to the interdigital electrodes with silver pasteThe contacts were subsequently annealed at 100 C for2 h to cure the silver paste

Figure 11 shows the measured 5-kHz sensor impedanceof the different sensors as a function of humidity Thehumidity-sensitive region of DO15 is sup145 to 95 RHSensors SO30 and SO50 become sensitive to humidity atapproximately 65 and 75 RH respectively A smallerpore size increases the range of humidity values overwhich the sensor is responsive and increases the sensitiv-ity at lower humidity levels Therefore an optimal poresize can be selected for the operating region of interest

283

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

103

104

105

106

0 20 40 60 80 100

Z (SO30)

Z(SO50)

Z(DO15)

| Z| (

W)

RH ()

Fig 11 Measured 5-kHz impedance of sensors over different humiditylevels

Figure 12 shows the response of the different sensorsto humidity at different measurement frequencies It canbe seen from the gure that as the frequency decreasesthe width of the sensitive region increases and the point atwhich the sensor becomes responsive shifts to lower rela-tive humidity values For a given pore size the humidity-sensitive region of operation can be selected by frequencytuning

The response time de ned as the time needed to reach90 of the nal signal for a given relative humidityand the recovery time de ned as the time taken forthe signal to come to within 10 of the initial valuewere determined by alternately exposing the sensors toa 2 and a 45 RH ambient with impedance measuredat 5 kHz A typical responserecovery graph of sensorSO50 is shown in Figure 13 All sensors were com-pletely reversible regaining their original impedance val-ues even after repeated exposure to high humidity levelsThe response time and recovery times corresponding todifferent pore diameters are shown in Figure 14 The min-imum responserecovery times were observed for SO50the sample with the largest pore size

The impedance spectra of the sensors were taken overa range of humidity levels The total impedance wasresolved into real (Z0) and imaginary (Z00) parts and Cole-Cole impedance plots were constructed Figure 15andashcshows respectively the Cole-Cole plot of sensors SO50SO30 and DO15 The equivalent circuit models shown inFigure 16 were used to t the experimental data The t-ting was accomplished with the complex nonlinear least-square- tting program101 The dots in the plots repre-sent experimental data and lines represent the equiva-lent circuit model t The equivalent circuit denoted (a)or (b) with reference to Figure 16 used for tting isdenoted near the corresponding curves in Figure 15 Theequivalent circuit models are different from the empiri-cal model suggested by Falk et al102 for porous aluminasensors fabricated by anodization of aluminum thin lmsIn the equivalent circuit models R1 and R2 represent twofrequency-independent resistors in parallel with disper-sive frequency-dependent capacitors Cn14mdash5 and Cn24mdash5respectively103 These non-Debye capacitances can be

103

104

105

106

107

SO50

5kHz10kHz100kHz1MHz

0 20 40 60 80 100

0 20 40 60 80 100

0 20 40 60 80 100

| Z| (

W)

| Z| (

W)

| Z| (

W)

RH ()(a)

(c)

(b)

103

104

105

106

107

SO30

5kHz10kHz100kHz1MHz

RH ()

102

103

104

105

106

DO15

5kHz10kHz100kHz1MHz

RH ()

Fig 12 Variation of sensor impedance with humidity and measure-ment frequency (a) SO50 (b) SO30 (c) DO15

68

7

72

74

76

78

0 200 400 600 800

Time (Sec)

45 RH

2 RH

45 RH 45 RH

2 RH

| Z| (

x105

W)

Fig 13 Response of sensor SO50 upon repeated cycling between rel-ative humidity environments of 2 and 45

284

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

10

15

20

25

30

35

40

45

50

20 25 30 35 40 45 50

Res

pons

eR

ecov

ery

Tim

e (S

ec)

Pore diameter (nm)

DO15

SO30

SO50

Fig 14 Dependence of response time and recovery time on pore size

expressed in a general form as Cn4mdash5 D Bn4imdash5nƒ1 whereBn is a constant for a given set of experimental conditionsand n takes a value between 0 and 1 according to the localmicroscopic environment through which charge transporttakes place751 1021 103 Bn behaves as an ideal frequency-independent capacitor as n approaches 1 and as an idealconductor when the value of n approaches 0101

By examination of Figure 15a SO50 it can be seenthat at low humidity levels the Cole-Cole plot describesan arc with a very large radius of curvature As the humid-ity increases the radius decreases and at an RH of about80 the complete semicircle comes inside the measur-able impedance range However at the same humiditylevel a spur appears in the low-frequency region that dis-torts the semicircle The spur which can be viewed aspart of a second semicircle dominates at higher humid-ity levels while the semicircle described by the high-frequency impedance values diminishes in size Thisbehavior is similar for all sensors For sensors SO50SO30 and DO15 respectively the RH at which thehigh-frequency semicircle comes completely within themeasurable impedance range is ordm75 65 and 35these values roughly match the in ection points seen inFigure 11

The fundamental mechanism that enables ceramic sen-sors to sense humidity is the physisorption of watermolecules on an initially chemisorbed layer of hydroxylions651 96 The chemisorbed hydroxyl ions enhance theelectrical conductivity of the sensor either by donatingelectrons to the conduction band of the base materialor through proton hopping between adjacent hydroxylgroups upon the application of an electric eld Theprocess of chemisorption occurs at very low humiditylevels and is unaffected by further changes in humid-ity However an increase in humidity makes the watermolecules physisorb to this hydroxyl layer The effec-tiveness of physisorption depends upon the cation chargecomplexes (from the material or the impurities) and thehydroxyl ions (water molecules present at the surface ofthe base material)96 The physisorption process is facili-tated by higher surface charge densities During formationof the rst physisorbed layer a water molecule becomesattached to two neighboring hydroxyl groups through

0

02

04

06

08

1

12

0 02 04 06 08 1 12

Zrsquo (MW)

35 (a) 55 (a)

65 (a)

75 (b)

81 (b)

93 (b) SO50

0

02

04

06

08

1

12

14

0 02 04 06 08 1 12 14

Zrsquo (W)

Z

(W)

Z

(MW

)Z

(M

W)

35 (a)

55 (a)

65 (b)

75 (b)

81 (b)

93 (b)

SO30

0

2 105

4 105

6 105

8 105

1 106

12 106

0 2 105 4 105 6 105 8 105 1 106 12 106

17 (a)27 (a)

35 (b)45 (b)55 (b)

93 (b)

DO15

(a)

Zrsquo (MW)

(b)

(c)

Fig 15 Cole-Cole plot of the sensors at different humidity levels(a) SO50 (b) SO30 (c) DO15 The points represent the experimentaldata and lines represent the t with the equivalent circuit model givenin Figure 9 The RH values at which the data were collected and theequivalent circuit used for tting each curve (given as (a) or (b) denot-ing the circuit of Figure 17a and b respectively) are shown adjacent toeach curve

hydrogen double bonds and a proton may be transferredfrom a hydroxyl group to the water molecule to form aH3OC ion

When the physisorption occurs within less than a monolayer that is when clusters of physisorbed molecules

285

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

(b)

(a)

Cn1 (w)

R1

Cn2 (w) Cn1 (w)

R1 R2

Fig 16 Equivalent circuit used for tting the experimental data

are present at the surface it is thought that H3OC diffu-sion within hydroxyl groups and proton transfer betweenadjacent water molecules clusters take place96 HoweverKhanna and Nahar104ndash106 refute this theory They arguethat the proton diffusion is less probable in a surfacethat contains hydroxyl ions and clusters of physisorbedmolecules and hence cannot account for the conductiv-ity of porous alumina at an RH less than 40 that isbefore the rst monolayer is formed Instead they sug-gested a phonon-assisted electron tunneling mechanism atlow humidity levels in which electron tunneling occursfrom one water molecule to another as enhanced by thesurface anions of the alumina sensors

At higher humidity levels the number of physisorbedlayers increases allowing each water molecule to besingly bonded to a hydroxyl group and proton hop-ping between adjacent water molecules in the continu-ous water layer takes place The conduction process isthe same as that of pure water and is called the Grot-thuss chain reaction96 The dominant charge carrier ina high-humidity environment is therefore HC ions (pro-tons) The concentration of HC increases with increas-ing humidity and HC move freely through the water-likelayer According to the theory of Khanna and Nahar104

in porous alumina the protons for conduction are donatednot only by H2O but also by the Al(OH)3 formed asa result of water adsorption and the impurity anionsincorporated into the lm from the electrolyte duringanodization107ndash109 The water-like network consists ofsingly bonded water molecules which possess a highdielectric constant as they form dipoles and reorient freelyunder an externally applied electric eld The impedanceof the nanoporous alumina sensors includes this capaci-tive contribution as well

From the analysis of Figure 15 the high-frequencysemicircle that diminishes in size with increasing RHarises because of the effect of a physisorbed water layerat the surface Therefore R1 in the equivalent circuitof Figure 16 represents the charge transfer resistancethrough the adsorbed water layer The effect of electronic

102

104

106

108

SO50SO30

DO15

R1

(W)

0 20 40 60 80 100

RH ()

Fig 17 Humidity dependence of resistance R1 of different sensorsobtained by tting the equivalent circuit model with experimental data

conduction through the alumina bulk is parallel to thatof the surface conduction affected by the adsorbed waterlayer This bulk conduction process is dominant only atextremely low humidity values when there are only afew adsorbed molecules Hence the incomplete semicir-cle or arc in the low-humidity region is partially deter-mined from the base alumina Bn1 represents the capac-itance contribution from the adsorbed water layer Thelarge spurs seen in the low-frequency region are a con-sequence of ion migration in the adsorbed layer towardelectrodes1011 103 This migration leads to accumulation ofions at the electrodes Since the relaxation time of ionmigration is larger than the charge transfer process itis seen only at low frequencies The accumulated ionsin the adsorbed water layer give rise to R2 and Bn2From the equivalent circuits given in Figure 16 it canbe seen that the presence of ion accumulation repre-sented by the combination of R2 and Cn24mdash5 increasesthe overall impedance of the sensor Hence the reductionin impedance caused at high humidity due to the chargetransfer process is countered by ion accumulation

To better understand the fundamental sensing mecha-nisms R1 and Bn1 obtained through the numerical ttingprogram for the different sensors are plotted in Figures 17and 18 respectively It can be seen from Figure 17 thatR1 which is the resistance to charge transport in a staticelectric eld by the material containing an adsorbed waterlayer in the absence of ion migration is linear with RH

-11

10-10

10-9

10-8

10-7

50V30VDO15

Bn1

0 20 40 60 80 100

RH ()

Fig 18 Humidity dependence of Bn1 of different sensors obtained by tting the equivalent circuit model (Fig 17) with experimental data

286

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

in a semilog plot For sensor DO15 R1 decreases to afew hundred ohms at high RH levels indicating that theobserved conductivity is not due merely to the protonsprovided by the adsorbed water layer Since deionizedwater and argon gas are used to create the required humid-ity environment it is reasonable to assume that the sur-faces of the alumina lms contain anions presumablyfrom the electrolyte used for anodization which controlsconduction through the adsorbed layer108

With reference to Figure 18 Bn1 of the larger pore sam-ples SO50 and SO30 increases signi cantly only at thehigher humidity levels This result is in contrast to thatof the smaller pore size sensors which shows that Bn1

increases at lower humidity levels The relative humid-ity values where a signi cant increase in Bn1 takes placegenerally match the point seen in Figure 11 where theimpedance drops signi cantly which in turn is a func-tion of pore size These same relative humidity valuescorrelate with the point at which the spur appears inthe impedance diagrams given in Figure 15 The resultslead to the following conclusions Since hydrogen-bondedwater molecules possess a high dielectric constant theincrease in Bn1 indicates the presence of these bondson the surface which means that a liquid-like networkmore than one monolayer of physisorbed molecules hasformed at that humidity level This liquid-like network isresponsible for the sharp reduction in impedance seen inFigure 11 and the ion migration that creates the spur inthe low-frequency portion of the Cole-Cole plots as seenin Figure 15 According to Eq (1) capillary condensationdoes not take place in the alumina test sensors

In light of the above observations we propose thefollowing model to explain the behavior of nanoporousalumina in response to humidity The anodization pro-cess results in a certain amount of electrolyte anionstrapped within the pores1041 1071 108 Even in a low-humidityregime the presence of these anions on the sensor sur-face provides a high charge density for easy physisorp-tion of water molecules The result is the formationof a liquid-like network within the pores The impu-rity anions also act as proton donors by a dissociativemechanism104 joining the protons donated by the watermolecules The smaller the pore size the lower the RHlevel at which the liquid-like network is formed Hencefor the small pore sensors there is variation in Bn1 aswell as impedance magnitude in low-humidity regimesthe exponential dependence of R1 on RH also suggeststhe same phenomenon

In the case of DO15 the rst semicircle is highly dis-torted by the low-frequency spur originating with ionmigration A relatively large number of mobile ions mightbe present in these sensors leading to an enhancedaccumulation of protons at the electrode-sample contactAs mentioned earlier the dipoles in the singly bondedadsorbed water layer are free to rotate in an electric eld

Hence in the region close to the electrode-alumina inter-face the adsorbed water layer exhibits a net dipolar ori-entation and therefore it does not display its normaldielectric constant1011 103 Depending on ion size up toa monolayer of charge can exist at the electrode101 thediffuse (Gouy) layer of counter-ions spreads away fromthis inner (Stern) layer The number of ions available toform the Stern and Gouy layers at a particular frequencyincreases with increasing number of physisorbed layersHence as the number of physisorbed layers increases theStern and Gouy layers start forming at higher frequencies

With reference to Figure 12 the relative humiditylevel at which the sharp decrease in impedance occursshifts higher with increasing frequency This result can beexplained with the equivalent circuit shown in Figure 16athat is the circuit in the absence of ion accumulation Athigh frequencies the reactive impedance Xc of Cn14mdash5equal to Xc

D 1=jmdashCn14mdash5 is low and hence impedanceis largely determined by R1 As the humidity increasesR1 decreases however this translates into a reduction inimpedance only at relatively low frequencies where thevalue of R1 is comparable to or less than the capacitivereactance of Cn1(mdash) As the measurement frequency isreduced the reactance of Cn1(mdash) decreases hence the effectof R1 becomes more signi cant at lower humidity levels

32 Hydrogen Sensing with Titania Nanotubes

We recently reported62 the fabrication of self-organizedtitania nanotube arrays with an anodization techniqueAlthough the as-prepared nanotubes are amorphous theycrystallize on annealing at elevated temperatures and arestructurally stable to at least 600 C This stability ofstructure which is an essential criterion of a gas-sensingmaterial prompted us to study the gas-sensing behavior ofthese nanotubes for technologically important gases suchas oxygen carbon monoxide ammonia carbon dioxideand hydrogen

Titania has earned much attention for its oxygen-sensing capabilities110ndash115 Furthermore with propermanipulation of the microstructure or crystalline phaseandor the addition of the proper impurities or surfacefunctionalization titania can also be used as a reducinggas sensor116ndash127 The interaction of a gas with a metaloxide semiconductor is primarily a surface phenomenonand nanoporous metal oxides1221 1231 1281 129 offer the advan-tage of providing large sensing surface areas

Room-temperature metal oxide hydrogen sensors aregenerally based on Schottky barrier modulation of deviceslike PdTiO2 or PtTiO2 by hydrogen130ndash132 Elevated-temperature hydrogen sensors examine electrical resis-tance with hydrogen concentration for example Birke-feld et al133 observed that the resistance of the anatasephase of titania varies in the presence of carbon monoxideand hydrogen at temperatures above 500 C but on dop-ing with 10 alumina it becomes selective for hydrogen

287

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Barrier layer

Titanium Nano tube array

Platinum electrodes

Fig 19 Schematic representation of the hydrogen sensor geometry

Munuera et al134 used rhenium doping in the anatasephase to perform room-temperature hydrogen monitor-ing Recently Shimizu et al122 reported that anodizednanoporous titania lms with a Pd Schottky barrier aresensitive to hydrogen at 250 C

Titania nanotubes were grown from titanium foil(995 pure from Alfa Aesar Ward Hill MA) with athickness of 025 mm The anodization was performed inan electrolyte medium of 05 hydro uoric acid in waterwith the use of a platinum foil cathode Well-de nednanotube arrays were grown with anodizing potentialsranging from 12 V to 20 V Nanotube length increaseswith anodization time reaching 400 nm in approximately20 min and then remains constant For the present studythe samples were anodized for 25 min The samples werethen annealed at 500 C in a pure oxygen ambient for 6 hwith a heating and cooling rate of 1 Cmin

The electrode geometry of the titania nanotube sen-sors is shown in Figure 19 The sensor consists of a basetitanium metal foil with a nanotube array grown on topAn insulating barrier layer separates the nanotubes fromthe conducting titanium foil Electrical connections weremade by two parallel 10 mm pound 2 mm platinum pads of100-Œm thickness The test chamber consists of a 13-literquartz tube with stainless steel end caps placed insidea furnace (Thermolyne USA model 21100 tubular fur-nace) Gas ow through the test chamber was controlledvia a computer-controlled mass ow controller The elec-trical resistance of the titania sensors was measured witha computer-controlled Keithly digital multimeter Prior todata collection the test chamber was initially evacuatedwith a mechanical pump whereupon nitrogen (99999pure) was passed while the sensor under test was heatedto the desired temperature The test gases were mixed inappropriate ratios with nitrogen to create the necessarytest gas ambient

The surface morphology of nanotube arrays preparedwith an anodization potential of 20 V and annealed at500 C for 6 h in a pure oxygen ambient is shown inFigure 20a and b It can be seen from these images thatthe nanotubes are uniform over the surface The nano-tubes are approximately 400 nm in length and have abarrier layer62 thickness of ordm50 nm For the nanotubes

100 nm

(a)

(b)

1 mm (c)

100 nm Fig 20 The surface morphology of the titania nanotubes after anneal-ing at 500 C (a) High- and (b) low-magni cation images of a 20-Vsample (c) A high-magni cation image of a 12-V sample

fabricated with 20-V anodization the average pore diam-eter as determined from FESEM images is 76 nm (thestandard deviation 15 nm) with a wall thickness of 27 nm(standard deviation 6 nm) The sample anodized at 12 Vshown in Figure 2c was found to have an average porediameter of 46 nm (standard deviation 8 nm) with a wallthickness of 17 nm (standard deviation 2 nm) The porosi-ties of the 20-V and 12-V samples were calculated as 45and 61 respectively

Glancing-angle X-ray diffraction patterns of a 20-Vsample annealed at 500 C for 6 h in oxygen ambientis shown in Figure 21 It can be seen that both anataseand rutile phases of titania are present in the sampleA detailed study135 of these structures by high-resolutiontransmission electron microscope (HRTEM) showed that

288

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

20 30 40 50 60 70 80

2 q (deg)

(101

)

(110

)

A

R

TR

T

T

R R A TRR T R R T TT

Fig 21 Glancing-angle X-ray diffraction pattern of a 20-V sample(glancing angle D 2) annealed at 500 C in oxygen ambient A R andT represent respectively the re ections from anatase crystallites rutilecrystallites and the titanium substrate

the anatase crystallites were concentrated on the walls ofthe nanotubes and rutile on the barrier layer

Nanotubes annealed in a pure oxygen ambient werefound to be stable (intact) to temperatures of approxi-mately 580 C Above this temperature protrusions wereseen to be coming out through the nanotubes an effectthat spread with increasing temperature These protru-sions which are due to oxidation of the titanium substratecollapse the nanotubes

Figure 22 shows the response of the 20-V (76 nm) sam-ple as a function of ambient temperature as it is switchedfrom a nitrogen environment to one containing 1000 ppmhydrogen and then back to nitrogen The plot was madeusing 4Rg=R05

ƒ1 versus time where R0 is the base resis-tance of the sensor that is the sensor resistance before thetest gas was introduced and Rg is the measured resistancein the presence of test gas The sensor shows increasinghydrogen sensitivity with temperature with a change by3 orders of magnitude in resistance at temperatures aboveordm300 C At all of the temperatures the original resistancerecovered without hysteresis

The sensitivity S is de ned by the formula

S DR0

ƒ Rgs

Rgs

10 0

10 1

10 2

10 3

0 10 0 5 10 2 1 10 3 15 103 2 10 3 25 103 3 10 3 35 103 4 10 3

Time (Sec)

(Rg

R0)

-1

180degC

240degC

265degC

290degC

345degC

375degC

H2+N

2

N2

Fig 22 Variation of resistance Rg normalized with respect to baselineresistance R0 of a 20-V sample with time on exposure to 1000 ppmhydrogen at different temperatures It may be noted that the inverse ofRg=R0 was used in the plot to represent data in the positive y direction

10-1

100

101

102

103

150 200 250 300 350 400 450

Temperature (degC)

Sens

itivi

ty (

R0-

Rgs

)R

gs

Fig 23 The sensitivity temperature dependence of a 20-V sample to1000 ppm hydrogen

where R0 is the resistance of the sensor before the gas ispassed and Rgs is the resistance after gas is passed andreaches the saturation value The temperature-dependentsensitivity of a 20-V sample to 1000 ppm hydrogen isshown in Figure 23 Sensitivity is seen to increase withtemperature to approximately 380 C where the increasein sensitivity with temperature begins to saturate

The response time de ned as the time needed for thesensor to reach 90 of the nal signal for a given concen-tration of gas is plotted against temperature in Figure 24(the time includes that required for the gas to equili-brate inside the measurement chamber estimated to beordm30 s) The response time decreases exponentially withtemperature

To check the behavior of the sensor on repeated hydro-gen exposure the hydrogen concentration was varied indiscrete steps of 100 ppm from 0 to 500 ppm while thetemperature was kept constant at 290 C the chamber was ushed with nitrogen after each exposure to hydrogenThe response of the 20-V prepared nanotube sensorskept at 290 C is shown in Figure 25 The behaviorof the sensor is consistent recovering its original resis-tance after repeated exposure to varying hydrogen gasconcentrations The sensitivities of the 76-nm- 53-nm-and 22-nm-diameter nanotube sensors to hydrogen con-centrations ranging from 100 ppm to 1 are shown inFigure 26 The sensitivity of the 76-nm (20-V) sensor tolow concentrations is shown in Figure 27 there is a linearincrease in sensitivity at low concentrations The lower

10 1

10 2

10 3

10 4

150 200 250 300 350 400 450

Temperature (degC)

Res

pons

e T

ime

(Sec

)

Fig 24 Variation in response time for a 20-V sensor as a function oftemperature The dots represent measured data

289

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

10 2

0 4 4 4 4 4

10 3

0 10 1 10 2 10 3 10 4 10 5 10

Time (Sec)

Res

ista

nce

(W)

H2 (ppm)

200

300

400

500

400

300

200

100

100

Fig 25 Resistance of a 20-V sample when exposed to different con-centrations of hydrogen at 290 C The nitrogen-hydrogen mixture waspassed for 1500 s the chamber was then ushed with nitrogen for 3000 sbefore the nitrogen-hydrogen mixture was passed again

test limit of 20 ppm is determined by the experimentalapparatus and not the sensor

Figure 28 shows the change in resistance of the 76-nm-46-nm- and 22-nm-diameter nanotube sensors with expo-sure to 1000 ppm hydrogen at 290 C Whereas smallerdiameter nanotubes had greater sensitivity to hydrogenthe samples made at lower anodizing voltages tended tobe more brittle and harder to handle mechanically withoutbreaking

The 20-V 76-nm-diameter sample was exposed to oxy-gen carbon monoxide ammonia and carbon dioxide at290 C to study the cross-sensitivity The sensor wasfound to have no detectable variation in resistance onexposure to carbon dioxide The sensitivities of the tita-nia nanotubes to the other gases are shown in Figure 29The sensitivities of the nanotubes to carbon monoxide andammonia are negligible compared with that of hydrogenThe resistance of the nanotubes increased in the presenceof oxygen and did not regain their original electrical con-ductivity even after several hours in a 290 C nitrogenenvironment

1

10

100

1000

104

105

0 02 04 06 08 1

Hydrogen Concentration ()

Pore diameter = 76 nm

Pore diameter = 53 nm

Pore diameter = 22nm

Sen

sitiv

ity (

R0-

Rg)

Rg

Fig 26 The sensitivity variation of a samples with pore diameters of22 nm 53 nm and 76 nm versus time and hydrogen concentrations

0 100 200 300 400 5001

10

100

Hydrogen concentration (ppm)

Sen

sitiv

ity

(R0-

Rgs

)R

gs

Fig 27 Sensitivity of a 20-V (76 nm) sample at 290 C for low hydro-gen concentrations

Since the sensor measurements were made in atmo-spheres without oxygen the increase in conductivitycannot be due to hydrogen removing oxygen from thelattice132ndash134 or the removal of chemisorbed oxygen135ndash137

Hydrogen molecules can dissociate at defects on thetitania surface which can subsequently diffuse into thetitania lattice acting as electron donors1281 1381 139 How-ever a dissociation-driven process would result in aslow response and recovery times with complete recov-ery unlikely since the sensor completely regains its orig-inal resistance with hydrogen cycling it appears that thisis not the dominant mechanism behind high hydrogensensitivity We believe chemisorption of the dissociatedhydrogen on the titania surface is the underlying sensingmechanism137 During chemisorption hydrogen acts as asurface state and a partial charge transfer from hydrogento the titania conduction band takes place This createsan electron accumulation layer on the nanotube surfaceenhancing its electrical conductance Upon removal of thehydrogen ambient electron transfer back to the hydro-gen molecule takes place which subsequently desorbsrestoring the original electrical resistance of the mate-rial Another factor that may play a role in the hydrogensensitivity (and selectivity) is the platinum electrodes Atelevated temperatures hydrogen dissociation can occur onplatinum surfaces These dissociated hydrogen atoms mayspill1361 140 onto the nanotube surface where they diffuse

10-1

101

103

105

0 05 1 15

Time (Hour)

Pore size - 76 nm

Pore size - 53 nm

Pore size - 22 nm

(Rg

R0)

-1

Fig 28 A comparison of the variation in resistance of samples havingpore diameters of 22 nm 53 nm and 76 nm versus time upon exposureto 1000 ppm of hydrogen at 290 C

290

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

10-2

10-1

100

0 100 2 102 4 102 6 102 8 102 1 103

OxygenAmmoniaCarbon monoxide

Time (Sec)

(Rg

R0)

-1

Fig 29 Variation of resistance of a 20-V sample when exposed to1000 ppm carbon monoxide 5 ammonia and 1 oxygen at 290 C

into the material affecting its electrical properties Fromthe present study it was not clear how signi cant a rolethe platinum electrodes play

Anatase the polymorph of titania has been reported tohave high sensitivity for reducing gases like hydrogen andcarbon monoxide1211 1241 133 Our nanotube samples containanatase phase mainly on the walls and rutile in the barrierlayer As the diffusing hydrogen atoms go to the intersti-tial sites1331 142 and as the ca ratio of anatase is almostfour times that of rutile it appears that the anatase latticeaccommodates hydrogen easily and hence makes a highercontribution to hydrogen sensitivity

The effect of chemisorption can be neglected in theoxygen-sensing experiments As the recovery requiresseveral hours it appears that the nanotubes contain oxy-gen vacancies or titanium interstitial defects in the pres-ence of nitrogen On exposure of the sensor to oxygenambient the lattice reoxidizes and hence the conductiv-ity of the sensor decreases On the removal of oxygenthe reduction of the lattice will not immediately occurhence the sensor requires several hours to regain its orig-inal conductivity

4 CONCLUSIONS

This paper summarizes different routes of fabrication ofmetal oxide nanoarchitectures useful for sensing applica-tions Structures made via anodization of a starting metaland breath gure formation from a sol-gel are discussedand presented as are TiO2 nanowires made by lling ananoporous alumina membrane with subsequent removalof the supporting template Two test cases illustratingthe utility of these nanoporous metal oxide structures forsensing are presented highly ordered nanoporous aluminais used for humidity sensing and TiO2 nanotubes are usedfor hydrogen sensing These study test cases reveal sev-eral points worth brie y noting

In the rst sensing test case uniform nanoporous alu-mina sensors exhibit pore size and frequency-dependent

humidity-sensing characteristics The mechanism respon-sible for the sensing action of the nanoporous alumina sen-sors was studied by representing the impedance spectra atdifferent humidity levels in Cole-Cole plots This makes itpossible to identify the competing impedance-determiningprocesses occurring in the sensors namely charge transferand mass transport through the physisorbed water layeras dependent upon the humidity level and input signalfrequency Charge transfer was found to enhance the sen-sitivity by causing a large variation in impedance withrespect to humidity Mass transport leads to accumulationof ions at the electrodes which tends to decrease the mea-surement sensitivity by increasing the impedance Hencea proper choice of operating frequency high enough toavoid ion accumulation is essential for optimum sensorperformance

It has been inferred from our results that anions becomeincorporated into the alumina lm during the anodiza-tion process supporting the hypothesis of Khanna andNahar104ndash106 where they act as proton donors and facili-tate the adsorption of water molecules Hence liquid-likecharge transfer networks are formed at relatively lowerhumidity levels for the smaller pore-size sensors Thisprocess rather than capillary condensation is primarilyresponsible for the observed humidity-sensing behavior ofthe sensors Our studies show that a distribution of poresizes as suggested by Shimizu et al94 is not necessary toobtain sensitive performance over a wide humidity rangerather such optimal performance can be obtained with uni-form nanodimensional pores containing adsorbed anionicimpurities

In the second sensing test case titania nanotubes pre-pared by anodization and annealed in an oxygen atmo-sphere at a temperature of 500 C were found to behighly sensitive to hydrogen The nanotube sensors con-tained both anatase and rutile phases of titania and showedappreciable sensitivity toward hydrogen at temperaturesas low as 180 C The sensitivity increased drasticallywith temperature showing a variation of 3 orders in mag-nitude of resistance to 1000 ppm of hydrogen at 400 CThe response time decreased with increasing temperatureat 290 C full switching of the sensor took approximately3 min Results were highly reproducible with no indica-tion of hysteresis Our results showed that these sensorsare at least capable of monitoring hydrogen levels from20 ppm to 4 At 290 C nanotubes with smaller porediameters showed higher sensitivity to hydrogen com-pared with their larger pore counterparts The sensorsshowed high selectivity to hydrogen compared with car-bon monoxide ammonia and carbon dioxide Althoughthe sensor was sensitive to high concentrations of oxy-gen the response time was high and the sensor did notcompletely regain the original condition We believe thehydrogen sensitivity of the nanotubes is due to hydrogendiffusion into the titania lattice where they act as electrondonors

291

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Acknowledgments Support of this work by the NationalScience Foundation through grant ECS-0210033 is grate-fully acknowledged

References and Notes

1 K Kajihara K Tanaka K Horao and N Soga Jpn J ApplPhys 35 6110 (1996)

2 B O Regan and M Gratzel A low cost Nature 353 737 (1991)3 A Fujishima and K Honda Nature 238 37 (1972)4 K Sato A Tsuzuki H Taoda Y Torii T Kato and Y Butsugan

J Mater Sci 29 5911 (1994)5 K Sato A Tsuzuki Y Torii H Taoda T Kato and Y Butsugan

J Mater Sci 30 837 (1995)6 N Ozer and C M Lampert Sol Energy Mater Sol Cells 54 147

(1998)7 A Manz and H Becker editors Microsystem Technology in

Chemistry and Life Sciences Springer-Verlag Berlin 19988 E Kim Y Xia and G M Whitesides Nature 376 581 (1995)9 M Templin A Franck A Du Chesne A Leist A Zhang

R Ulrich V Schadler and U Weisner Science 278 1795 (1997)10 D Zhao J Feng Q Huo N Melosh G H Frederickson

B Chmelka and G D Stucky Science 279 548 (1998)11 O D Velev T A Jede R F Lobo and A M Lenhoff Nature

389 447 (1997)12 B T Holland C F Blanford and A Stein Science 281 538

(1998)13 S Matsushita T Miwa and A Fujishima Chem Lett 9 925

(1997)14 A Imhof and D J Pine Adv Mater 10 697 (1998)15 T Tatsuma A Ikezawa Y Ohko T Miwa T Matsue and

A Fujishima Adv Mater 12 643 (2000)16 Z Zhong Y Yin B Gates and Y Xia Adv Mater 12 206 (2000)17 O D Velev and A M Lenhoff Curr Opin Colloids Interface

Sci 5 56 (2000)18 O D Velev and E W Kaler Adv Mater 12 531 (2000)19 P Jiang J Cizeron J F Bertone and V L Colvin J Am Chem

Soc 121 7957 (1999)20 M E Turner T J Trentler and V L Colvin Adv Mater 13 180

(2001)21 K M Kulinowski P Jiang H Vaswani and V L Colvin Adv

Mater 12 833 (2000)22 P Jiang J F Bertone and V L Colvin Science 291 453 (2001)23 R Seshadri and F C Meldrum Adv Mater 12 1149 (2000)24 F C Meldrum and R Seshadri Chem Commun 29 (2000)25 Q B Meng Z Z Gu O Sato and A Fujishima Appl Phys

Lett 77 4313 (2000)26 G Widawski B Rawiso and B Francois Nature 369 3897

(1994)27 B Francois O Pitois and J Francois Adv Mater 7 1041 (1995)28 O Pitois and B Francois Eur Phys J B 8 225 (1999)29 S A Jenekhe and X L Chen Science 283 372 (1999)30 M Shimomuara and T Sawadaishi Curr Opin Colloid Interface

Sci 6 11 (2001)31 O Karthaus N Maruyama X Cieren M Shimomura

H Hasegawa and T Hashimoto Langmuir 16 6071 (2000)32 T Nishikawa J Nishida R Ookura S Nishimura S Wada

T Karino and M Shimomura Mater Sci Eng C 10 141 (1999)33 K Kajihara K Nakanishi K Tanaka K Hirao and N Soga

J Am Ceram Soc 81 2670 (1998)34 C A Grimes R S Singh E C Dickey and O K Varghese

J Mater Res 16 1686 (2001)35 M Srinivasrao D Collings A Phillips and S Patel Science 292

79 (2001)36 P G de Gennes Rev Mod Phys 57 827 (1985)

37 D Beysens A Steyer P Guenoun D Fritter and C M KnoblerPhase Transitions 31 219 (1991)

38 B J Briscoe and K P Galvin J Phys D 23 422 (1990)39 B J Briscoe and K P Galvin J Phys D 23 1265 (1990)40 D Beysens and C M Knobler Phys Rev Lett 57 1433 (1986)41 F Family and P Meakin Phys Rev Lett 61 428 (1988)42 A V Limaye R D Narhe A M Dhote and S B Ogale Phys

Rev Lett 76 3762 (1996)43 P Dellrsquo Aversana and G P Neitzel Phys Today 51 38 (1998)44 P Dellrsquo Aversana J R Banavar and J Koplik Phys Fluids 8 15

(1986)45 P Dellrsquo Aversana V Tontodonato and L Carotenuto Phys Fluids

9 2475 (1997)46 M Q Li S Q Xu and E Kumacheva Langmuir 16 7275 (2000)47 P Freymuth Rev Sci Instrum 64 1 (1993)48 P Ball The Self-Made Tapestry Pattern Formation in Nature

Oxford University Press New York 199949 M F Schatz S J VanHook W D McCormick J B Swift and

H L Swinney Phys Fluids 11 2577 (1999)50 S R Coriell M R Cordes and W J Boettinger J Cryst Growth

49 13 (1980)51 T S Sullivan Y M Liu and R E Ecke Phys Rev E 54 486

(1996)52 J L Rogers M F Schatz J L Bougie and J B Swift 84 87

(2000)53 K Nitschke 52 R5772 (1995)54 E Bodenschatz J R De Bruyn G Ahlers and D S Cannell

Phys Rev Lett 67 3078 (1991)55 C W Meyer D S Cannell G Ahlers J B Swift and P C

Hohenberg Phys Rev Lett 61 947 (1988)56 R F Ismagilov D Rosmarin D H Gracias A D Stroock and

G M Whitesides Appl Phys Lett 79 439 (2001)57 E A Barringer and H K Bowen Langmuir 1 420 (1985)58 M Tschapek C Wasowski and R M T Sanchez J Electroanal

Chem 74 167 (1976)59 H Masuda and K Fukuda Science 268 1466 (1995)60 H Masuda F Hasegwa and S Ono J Electrochem Soc 144

L 127 (1997)61 O Jessensky F Muller and U Gosele Appl Phys Lett 72 1173

(1998)62 D Gong C A Grimes O K Varghese W Hu R S Singh

Z Chen and E C Dickey J Mater Res 16 3331 (2001)63 B M Kulwicki J Am Ceram Soc 74 697 (1991)64 H Arai and T Seiyama in Sensors A Comprehensive Survey

edited by W Gopel J Hesse and J N Zemel VCH Weinheim(1992) Vol 3 pp 981ndash1012

65 E Traversa Sens Actuators B 23 135 (1995)66 N Yamazoe and Y Shimizu Sens Actuators 10 379 (1986)67 A Bearzotti I Fratoddi L Palummo S Petrocco A Furlani

C Lo Sterzo and M V Russo Sens Actuators B 76 316 (2001)68 Y Sakai Y Sadaoka and M Matsuguchi Sens Actuators B 35

85 (1996)69 C A Grimes and D Kouzoudis Sens Actuators A 84 205

(2000)70 K G Ong C A Grimes C L Robbins and R S Singh Sens

Actuators A 93 33 (2001)71 T Seiyama N Yamazoe and H Arai Sens Actuators 4 85

(1983)72 L Ketron Ceram Bull 68 860 (1989)73 E Traversa G Gusmano and A Montenero Eur J Solid State

Inorg Chem 32 719 (1995)74 M K Jain M C Bhatnagar and G L Sharma Sens Actuators

B 55 180 (1999)75 O K Varghese and L K Malhotra J Appl Phys 87 7457 (2000)76 G Sberveglieri R Murri and N Pinto Sens Actuators B 23

177 (1995)

292

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Stimulus (s)

Electrical Signal S(s)

Circuit Design Signal Processing

Materials Design -useful transduction mechanism

-chemistry -nano to microstructure

Device Design

Fig 1 Schematic drawing of design variables inherent in the fabrica-tion of a useful sensor

2 FABRICATION OF METAL OXIDENANOARCHITECTURES

21 Mesoporous Thin Films

Nanoarchitectured thin lms are of considerable interestfor applications including photocatalysis and sensing andas templates for cell growth1ndash6 Apart from the conven-tional top-down engineering approach of translating a con-structed pattern onto a substrate by lithography7 or softlithography8 self-assembly and self-organization of mate-rials offer a rapid fabrication route at low cost Consider-able effort has focused on the use of a template aroundwhich the material of interest is assembled Depend-ing on the required pore size block copolymers91 10

latex spheres11ndash13 water-in-oil emulsions14 polystyreneparticles151 16 colloidal crystals17ndash22 and bioskeletons23ndash25

have been used as templates Problems associated withtemplate-assisted fabrication of porous structures includepreparation of a high-quality template complete lling ofthe voids in the template and minimization of shrinkageupon template removal Since any of these factors can

Oomman K Varghese received his PhD degree in physics from the Indian Institute of Technology Delhi India in2001 Since graduation he has worked as a postdoctoral fellow in the research group of Professor Grimes Departmentof Electrical Engineering and the Materials Science and Engineering Department the Pennsylvania State University His elds of interest include gas sensors based on metal oxide nanodimensional structures and carbon nanotubes impedancespectroscopy and transmission electron microscopy for materials characterization grain boundary space charge segregationeffects in binary oxides and sol-gel synthesis of metal oxide thin lms

Craig A Grimes received BS degrees in electrical engineering and physics from the Pennsylvania State University in1984 and the MS and PhD degrees in electrical engineering from the University of Texas at Austin in 1986 and 1990respectively He was employed by Lockheed Research Laboratories from 1990 to 1994 From 1994 to 2001 Dr Grimeswas a member of the Electrical and Computer Engineering Department at the University of Kentucky where he wasthe Frank J Derbyshire Professor He is currently an associate professor at the Pennsylvania State University in theDepartment of Electrical Engineering and Materials Science and Engineering His research interests include remote querychemical and environmental sensors nanodimensional metal-oxide thin- lm architectures for sensing and biotemplatingpropagation and control of electromagnetic energy and carbon nanotube-based electronic devices He has contributed over120 archival journal publications and seven book chapters and is co-author of the book The Electromagnetic Origin ofQuantum Theory and Light published by the World Scienti c Publishing Company (2002) He is North American editorof Sensors and an editorial board member of IEEE Transactions on Magnetics

in uence the nal quality of the porous structure all threerequirements must be ful lled at the same time

A nontemplate method for the synthesis of orderedmicrometer-sized honeycomb structures by self-assemblyof block copolymers was reported by Francois et al26ndash28

and Jenekhe and co-workers29 Shimomura et al301 31

helped pioneer efforts to fabricate patterned thin lmsby self-assembly An organic chloroform solution withamphiphiles containing metal acetylacetonates or alkox-ides cast at high atmospheric humidity was found toform a closely packed layer of water droplets on topof the organic solvent with the water droplets acting asa template301 31 after evaporation of the chloroform andwater a honeycomb structure remains Finally the pyrol-ysis of the metal alkoxide lm leads to the formationof microporous metal oxide such as anatase with poresize and wall thickness of the resulting lm controlledby solution concentration and ambient humidity level Ina similar vein Nishikawa and co-workers32 reported thefabrication of porous lms by casting of a dilute solu-tion of amphiphillic polymers onto solid substrates at highhumidity levels

Several sol-gel methods have been employed to fab-ricate patterned TiO2 lms typically by the incorpora-tion of organic polymers with the precursor solution toobtain precisely controlled macroscopic structures Tat-suma et al15 for example reported the fabrication ofmicroporous TiO2 lms prepared with the use of a two-dimensional array of polystyrene microspheres as a tem-plate A TiO2 aqueous sol was introduced into the gapbetween the polystyrene spheres upon the evaporationof water the remaining lms were calcined to incineratethe polystyrene particles leaving a porous TiO2 lmIn an analogous approach Kajihara et al33 reported the

278

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

fabrication of porous TiO2 lms with the use of analkoxide-based sol-gel containing polyethylene glycolwhich served as the template into which the substrateswere dip coated

A disadvantage of using a template is that the dom-inant length scale of the resulting porous structure is xed by the template size therefore dynamic control ofthe length scale becomes almost impossible Our inter-est lies in the fabrication of nanoporous lms with-out the use of a template for example we recentlyreported the fabrication of metal oxide lms with magnet-ically modulated nanodimensionalporosity34 The absenceof a template makes the ambient humidity level andthe sol pH key process variables for the fabrication ofcontrolled structures35 We describe here the fabricationof TiO2 mesoporous structures via sol-gel breath gureformation36ndash48 as well as thermocapillary and surfacetension-driven Benard-Marangoni convection49ndash56

The overall hydrolysis and condensation reaction fortitanium isopropoxide the sol-gel precursor for TiO2 canbe represented as Ti(OC3H7)4 C 2H2O TiO2 C4HOC3H7 The condensation reaction leads to the for-mation of colloidal particles which can be polymeric orparticulate depending on the type of precursors and pHof the sol Colloidal particle aggregation may be inhib-ited by the formation of surface charge developed byeither preferential dissociation of one of the lattice ionsof the sol particle or preferential adsorption of chargedspecies from solution The surface charge is formed byeither protonation (Ti-OH C HC Ti-OHC

2 5 or depro-tonation (Ti-OH C OHƒ Ti-Oƒ C H2O) of the Ti-OHbonds The pH at which the surface is electrically neutralis called the point of zero charge (PZC) The surface isnegatively charged at a pH gt PZC and positively chargedat pH lt PZC For TiO2 the PZC varies at values at leastbetween 52 (Ref 57) and 55 (Ref 58)

In a nitrogen environment the precursor titaniumtetraisopropoxide (TTIP) was dissolved in isopropanolto which deionized water and then nitric acid wereadded The reagents used in the experiment namely TTIP(99999) isopropanol (995) and nitric acid (70redistilled) were procured from Aldrich The solution wascontinuously stirred for 2 h and then stored in a nitrogenenvironment In a typical preparation of 01 M TiO2 sol1 ml of TTIP 005 ml of HNO3 (70 distilled) 01 ml ofdeionized water and 327 ml of isopropanol were usedThe lms were deposited by either dip coating of theglass or silicon substrate or simply by placing a dropletof solution on a clean substrate The ambient humidityin which the TiO2 lms dried was controlled by passingnitrogen through a room-temperature bubbler The humid-ity and temperature of the chamber were monitored with adigital hydrometer Upon drying all lms were annealedat 100 C for 1 h in a nitrogen environment

A 001 M sol was modi ed by a 12 nitric aciddeionizedwater solution with 15 Œl of acidwater solution added to

the original 3385 ml sol the pH of the resulting solutionwas approximately 05 Topology formation is found to bea function of (1) ambient humidity (2) atmosphere owvelocity during drying (3) sol concentration and (4) solpH Figure 2andashc illustrates the effect of drying rate on thestructure Figure 2a shows the fastest drying rate and Fig-ure 2c the slowest (the dark areas are pores ie absenceof lm) In the initial stages of the lm growth process thecondensed droplets grow as isolated objects without inter-action between neighbors as seen in Figure 2a Longerdrying times permit the drops to begin to coalesce as seenin Figure 2b with further coalescence seen in Figure 2cBreath gures36ndash48 are formed when a liquid surface is

Fig 2 Drying-time-dependent FE-SEM images of a thin lmdeposited from a 01 M TiO2 sol 1 ml of TTIP 005 ml of HNO3 (70distilled) 01 ml of deionized water and 327 ml of isopropanol towhich 15 Œl of a 12 nitric aciddeionized water solution was added(a) the fastest drying region and (c) the slowest drying region wherethe liquid has coalesced into drops

279

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

brought into contact with moist air the solvent vapor pres-sure and air velocity across the surface drive solvent evap-oration rapidly cooling the surface This in turn facili-tates the nucleation and growth of water droplets from theatmosphere36ndash41 The temperature difference between theliquid surface and air results in thermocapillary convec-tion within the liquid that acts to stabilize the condensingwater droplets on or at the solution surface43ndash48 Air owacross the surface coupled with surface convection cur-rents drives the ordering of water droplets into hexagonalarrays44ndash46 Once the surface is completely covered withwater droplets the temperature difference between the sur-face and the droplets diminishes and the droplets beingdenser than the solvent sink into the solution de ning theresidual structure Figure 3 shows the effect of sol pHshowing the pH-dependent variation in the resulting struc-ture achieved

Benard-Marangoni convection typically results from avertical temperature gradient due to solvent evaporationfrom a thin liquid lm46ndash54 Preferential solvent evapo-ration removes heat from the lm surface resulting inBenard-Marangoni convection50ndash56 across the uid layerwith the suspension welling up in the center of a convec-tion cell and then owing back down the cell boundary Asevaporation proceeds the colloidal suspension becomesmore concentrated changing the convection character-istics and in turn de ning the ultimate lm structureWell-de ned hexagonal pentagonal or square patternsare expected from an ideal time-independent Benard-Marangoni convection50 in a homogeneous liquid whereasthe cell patterns of our lms seen in Figure 2 are typicallyirregular This deviation of structure from the theoreticalideal may be the result of time-dependent convection owthrough changes in sol viscosity with preferential evapora-tion of the propanol solvent leading to coupled thermoso-lutal Benard-Marangoni convection51 in the nonhomoge-neous uid

22 Nanotube Nanopore and NanowireFabrication via Anodization of Al and Ti

Highly ordered nanoporous alumina lms (see Fig 4aandb)aremadethrougha two-stepanodizationprocess59ndash61

The aluminum substrate is rst anodized in an oxalic orsulfuric acid solution The anodization is stopped after afew microns of aluminum are consumed and the porousalumina lm is removed through etching The etchant amixture of chromic acid and phosphoric acid is highlyselective attacking alumina much faster than aluminumThe remaining aluminum is dimpled with the dimplesserving as a uniform seed layer upon which a highly uni-form porous layer can then be achieved through a sec-ond anodization step at the same voltage Alumina tem-plates are widely used as templates for the fabrication ofnanowires Figure 4c shows a TiO2 nanowire mat fab-ricated by lling pores of an alumina membrane via sol

Fig 3 Images taken of a dried 01 M TiO2 sol 1 ml of TTIP 005 mlof HNO3 (70 distilled) 01 ml of deionized water and 327 ml ofisopropanol to which varying amounts of a 12 nitric aciddeionizedwater solution were added to vary pH values (a) pH 017 (b) pH 029(c) pH 037

gel letting the lm dry and then subsequently removingthe alumina template by a sodium hydroxide etch

Anodization has also been used to fabricate TiO2 nano-tube arrays62 with pore size linearly proportional toanodization voltage (see Fig 5) The nanotube array inFigure 5 was fabricated by anodization of titanium at 20 Vin 05 wt HF solution for 20 min resulting in a well-aligned titanium oxide nanotube array with an approxi-mate average tube diameter of 60 nm and a tube lengthof 400 nm Diameters of fabricated tubes have ranged insize from 25 nm to 65 nm

280

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

Fig 4 (a) The surface of a two-step anodized alumina membrane (the akes seen in the image are the ldquodustrdquo associated with tissue paper)(b) An image of a cleaved sample showing the pore channels (c) TiO2

nanowires made by lling an alumina template via sol-gel then subse-quently removing the template by a sodium hydroxide etch resulting ina TiO2 nanowire mat

3 EXPERIMENTAL RESULTSAPPLICATION TO SENSING

31 The Water Vapor Sensing Performanceof Highly Ordered NanoporousAlumina Films

Humidity sensors have attracted considerable attentionover many years because of their great importancein applications ranging from monitoring food qual-ity to meteorological studies63ndash65 Ceramic humiditysensors651 71ndash82 are commercially available and offer majoradvantages with high resistance to chemical attack

Fig 5 TiO2 nanotubes made by anodization of titanium

thermal stability mechanical strength and quick responseHowever ceramic humidity sensors still suffer from insuf- cient sensitivity over wide humidity ranges as well aslack of reversibility and drift in base resistance with timebecause of water molecule chemisorption

The humidity-sensing properties of alumina discoveredalmost 50 years ago83ndash94 are based upon ionic conduc-tion the presence of an adsorbed layer of water at thesurface reduces the total sensor impedance because of theincrease in the ionic conductivity as well as capacitancedue to the high dielectric constant of water The humidity-sensing characteristics of a given sensing element dependupon the material used the method of preparation andthe resulting surface topology Porous alumina is usuallypreferred for sensing applications because of the largesurface area available for water adsorption

An illustration of water adsorption by a surfaceredrawn from Ref 95 is shown in Figure 6 Figure6a shows the adsorption of a water molecule on aclean alumina surface Physisorption of water initiallyoccurs on an active surface site and forms an adsorption

Fig 6 Schematic illustration of interaction of a water molecule withan alumina surface redrawn from Ref 95

281

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Fig 7 Schematic illustration of buildup of adsorbed water layersredrawn from Ref 96 building of a water layer upon an alumina surface

complex (Fig 6a and b) The negatively charged oxygenof the water molecule becomes attached to the positivelycharged cation site and one of the positively chargedhydrogen atoms becomes attached to the anionic site byelectrostatic attraction Eventually they are converted totwo chemisorbed hydroxyl (OH) groups (Fig 6c) Anincrease in humidity makes the water molecules physisorbto this chemisorbed hydroxyl layer The effectiveness ofphysisorption depends upon the cation charge complexes(from alumina or the surface impurities) and the hydroxylions on the surface of the alumina The physisorptionprocess is facilitated by higher surface charge densitiesAfter hydroxyl formation the next water molecule willbe physisorbed through hydrogen double bonds on thetwo neighbouring hydroxyl groups (Fig 6d) and a protonmay be transferred from a hydroxyl group to the watermolecule to form a H3OC ion At higher humidity levelseach water molecule will be singly bonded to a hydroxylgroup as illustrated in Figure 7 (redrawn from [96]) andproton hopping between adjacent water molecules in thecontinuous water layer takes place The conduction pro-cess in this case occurs by attachment of a proton to awater molecule thereby forming a hydronium ion thehydronium ion releases another proton to a second watermolecule that accepts this proton while releasing a thirdproton and so on through the liquid Hence the dominantcharge carrier in a high-moisture atmosphere is the HC

(proton) The concentration of HC increases with increas-ing humidity and it moves freely through the water-likelayer

In the case of porous alumina capillary condensationcan take place in pores with a radius up to rK at a partic-ular relative humidity and temperature which is given byKelvinrsquos relation

rKD 2ƒM

RT ln4Ps=P5

Here P is the water vapor pressure Ps is the water vaporpressure at saturation ƒ is the surface tension R is theuniversal gas constant T is the temperature in Kelvin

H O

H O

O

H

H +

H

H

O

H +H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H O

H O

O

H

H +

H

H

O

H +H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H O

H O

O

H

H +

H

H

O

H+H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H

OH

O

O

HH

O

H+H

H+

H

OH

OH

O

O

HH

O

H+

HH

+H

O

H

O

H

O

O

HH

O

H+ HH+ H

O

H

O

H

O

O

HH

O

H+ HH+ H

O

Aluminium

AluminaCapillary Condensation

Capillary Condensation Adsorption

Adsorbed Water Layer

Fig 8 Schematic representation of water vapor interaction withnanoporous alumina illustrating some but not all97 of the multiple fac-tors inherent in determining the response

and and M are respectively the density and molecularweight of water

According this relation the condensation occurs insidesmall pores at lower humidity levels as illustrated inFigure 8 which is an illustrative drawing presentingthe different water-surface interaction mechanisms (NoteFigure 8 is not meant to be an exhaustive considera-tion of all possible effects for example the drawingdoes not address effects associated with capillary surfacecurvature97)

This condensed water reduces the impedance of alu-mina considerably No condensation occurs inside poresof radius greater than rK at a particular humidity level (seethe largest pore shown in Fig 8) In such pores as in thecase of a plane surface it is the number of physisorbedlayers that decides the impedance

Since capillary condensation enhances the sensingcapabilities of a material pore size distribution has beenwidely considered to be an important parameter in deter-mining the sensitivity in a particular humidity range941 98

However the results of our study on uniform nanoporousalumina lms presented below show that an easily builtand highly reproducible wide-range humidity sensor canbe achieved with nanodimensional pores of a narrow sizedistribution made by anodization of aluminum59ndash611 99

Starting from adhesive-backed aluminum tape (99pure) purchased from Tesa Tape Inc100 and subsequentlyanodized sensors SO50 SO30 and DO15 were preparedunder the conditions given in Table I Per sample nota-tion the rst letter denotes a single S or double D stepanodization process the second letter denotes the use of

Table I Anodization conditions used for making nanoporous alumina lms

Anodizing Duration of No ofvoltage anodization anodization

Sample (v) (min) stages

SO50 50 120 1SO30 30 120 1DO15 15 120 2

The electrolyte used in all cases was 2 wt oxalic acid

282

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

oxalic acid (O) in the anodization bath the two-digit num-ber denotes the anodization voltage It was observed thatsingle-step anodization results in disordered pore struc-tures at low voltages Hence a double-step process wasused to make sample DO15 These samples were initiallyanodized under the conditions given in Table I The sam-ples were then immersed in an aqueous solution contain-ing 18 wt chromic acid and 4 wt phosphoric acid for12 h to remove the initial nanoporous layer while keep-ing the nonporous bottom layer of alumina intact Thesamples were rinsed in distilled water and again anodizedunder the same conditions used for the rst anodizationstep Estimates of pore size distribution were determinedfrom FE-SEM images and are shown in Figure 9andashcas a size distribution histograph Samples SO50 SO30and DO15 have respectively average pore diameters of

0

10

20

30

375 40 425 45 475 50

o

f po

res

Pore diameter (nm)

0

10

20

30

25 325 375 425 475 525

o

f po

res

0

5

10

15

20

25

175 20 225 25 275 30 325 375

DO15

o

f po

res

(a)

Pore diameter (nm)

(c)

Pore diameter (nm)

(b)

Fig 9 Pore size distribution of the tested sensors (a) SO50 (b) SO30(c) DO15

Fig 10 Digital image of an impedance-based interdigital nanoporousalumina humidity sensor

452 nm 384 nm and 136 nm with a correspondingstandard deviation of 34 nm 78 nm and 49 nm Allsamples have a predominantly ordered pore structure

Immediately after anodization all samples were thor-oughly washed in distilled water and then dried in a nitro-gen atmosphere at 100 C An interdigital capacitor pat-tern with dimensions of 12 cm pound 12 cm was formed onthe surface of the alumina lms by the evaporation of gold(thickness ordm 250 nm) through a mask A digital cameraimage of a typical sensor is shown in Figure 10 To createthe necessary ambient humidity argon passed through amass ow controller (MFC) was bubbled through a bot-tle containing deionized water and then mixed with dryargon coming from another MFC in appropriate ratiosbefore being passed to the test chamber A constant total ow was maintained throughout the experiment Cham-ber humidity was monitored with the humidity probeof a Keithley 6517A electrometer and all measurementswere carried out at room temperature (23 C) Sensorimpedance was measured over the frequency range 5 Hzto 13 MHz with a computer-controlled Hewlett Packardimpedance analyzer (4192A) tted with an impedance test xture (Agilent 16034E) A signal amplitude of 90 mVwas used for all measurements Electrical contact wasmade between the sensor under test and the impedanceanalyzer test xture with 42-gauge 5-cm-long jumpersattached to the interdigital electrodes with silver pasteThe contacts were subsequently annealed at 100 C for2 h to cure the silver paste

Figure 11 shows the measured 5-kHz sensor impedanceof the different sensors as a function of humidity Thehumidity-sensitive region of DO15 is sup145 to 95 RHSensors SO30 and SO50 become sensitive to humidity atapproximately 65 and 75 RH respectively A smallerpore size increases the range of humidity values overwhich the sensor is responsive and increases the sensitiv-ity at lower humidity levels Therefore an optimal poresize can be selected for the operating region of interest

283

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

103

104

105

106

0 20 40 60 80 100

Z (SO30)

Z(SO50)

Z(DO15)

| Z| (

W)

RH ()

Fig 11 Measured 5-kHz impedance of sensors over different humiditylevels

Figure 12 shows the response of the different sensorsto humidity at different measurement frequencies It canbe seen from the gure that as the frequency decreasesthe width of the sensitive region increases and the point atwhich the sensor becomes responsive shifts to lower rela-tive humidity values For a given pore size the humidity-sensitive region of operation can be selected by frequencytuning

The response time de ned as the time needed to reach90 of the nal signal for a given relative humidityand the recovery time de ned as the time taken forthe signal to come to within 10 of the initial valuewere determined by alternately exposing the sensors toa 2 and a 45 RH ambient with impedance measuredat 5 kHz A typical responserecovery graph of sensorSO50 is shown in Figure 13 All sensors were com-pletely reversible regaining their original impedance val-ues even after repeated exposure to high humidity levelsThe response time and recovery times corresponding todifferent pore diameters are shown in Figure 14 The min-imum responserecovery times were observed for SO50the sample with the largest pore size

The impedance spectra of the sensors were taken overa range of humidity levels The total impedance wasresolved into real (Z0) and imaginary (Z00) parts and Cole-Cole impedance plots were constructed Figure 15andashcshows respectively the Cole-Cole plot of sensors SO50SO30 and DO15 The equivalent circuit models shown inFigure 16 were used to t the experimental data The t-ting was accomplished with the complex nonlinear least-square- tting program101 The dots in the plots repre-sent experimental data and lines represent the equiva-lent circuit model t The equivalent circuit denoted (a)or (b) with reference to Figure 16 used for tting isdenoted near the corresponding curves in Figure 15 Theequivalent circuit models are different from the empiri-cal model suggested by Falk et al102 for porous aluminasensors fabricated by anodization of aluminum thin lmsIn the equivalent circuit models R1 and R2 represent twofrequency-independent resistors in parallel with disper-sive frequency-dependent capacitors Cn14mdash5 and Cn24mdash5respectively103 These non-Debye capacitances can be

103

104

105

106

107

SO50

5kHz10kHz100kHz1MHz

0 20 40 60 80 100

0 20 40 60 80 100

0 20 40 60 80 100

| Z| (

W)

| Z| (

W)

| Z| (

W)

RH ()(a)

(c)

(b)

103

104

105

106

107

SO30

5kHz10kHz100kHz1MHz

RH ()

102

103

104

105

106

DO15

5kHz10kHz100kHz1MHz

RH ()

Fig 12 Variation of sensor impedance with humidity and measure-ment frequency (a) SO50 (b) SO30 (c) DO15

68

7

72

74

76

78

0 200 400 600 800

Time (Sec)

45 RH

2 RH

45 RH 45 RH

2 RH

| Z| (

x105

W)

Fig 13 Response of sensor SO50 upon repeated cycling between rel-ative humidity environments of 2 and 45

284

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

10

15

20

25

30

35

40

45

50

20 25 30 35 40 45 50

Res

pons

eR

ecov

ery

Tim

e (S

ec)

Pore diameter (nm)

DO15

SO30

SO50

Fig 14 Dependence of response time and recovery time on pore size

expressed in a general form as Cn4mdash5 D Bn4imdash5nƒ1 whereBn is a constant for a given set of experimental conditionsand n takes a value between 0 and 1 according to the localmicroscopic environment through which charge transporttakes place751 1021 103 Bn behaves as an ideal frequency-independent capacitor as n approaches 1 and as an idealconductor when the value of n approaches 0101

By examination of Figure 15a SO50 it can be seenthat at low humidity levels the Cole-Cole plot describesan arc with a very large radius of curvature As the humid-ity increases the radius decreases and at an RH of about80 the complete semicircle comes inside the measur-able impedance range However at the same humiditylevel a spur appears in the low-frequency region that dis-torts the semicircle The spur which can be viewed aspart of a second semicircle dominates at higher humid-ity levels while the semicircle described by the high-frequency impedance values diminishes in size Thisbehavior is similar for all sensors For sensors SO50SO30 and DO15 respectively the RH at which thehigh-frequency semicircle comes completely within themeasurable impedance range is ordm75 65 and 35these values roughly match the in ection points seen inFigure 11

The fundamental mechanism that enables ceramic sen-sors to sense humidity is the physisorption of watermolecules on an initially chemisorbed layer of hydroxylions651 96 The chemisorbed hydroxyl ions enhance theelectrical conductivity of the sensor either by donatingelectrons to the conduction band of the base materialor through proton hopping between adjacent hydroxylgroups upon the application of an electric eld Theprocess of chemisorption occurs at very low humiditylevels and is unaffected by further changes in humid-ity However an increase in humidity makes the watermolecules physisorb to this hydroxyl layer The effec-tiveness of physisorption depends upon the cation chargecomplexes (from the material or the impurities) and thehydroxyl ions (water molecules present at the surface ofthe base material)96 The physisorption process is facili-tated by higher surface charge densities During formationof the rst physisorbed layer a water molecule becomesattached to two neighboring hydroxyl groups through

0

02

04

06

08

1

12

0 02 04 06 08 1 12

Zrsquo (MW)

35 (a) 55 (a)

65 (a)

75 (b)

81 (b)

93 (b) SO50

0

02

04

06

08

1

12

14

0 02 04 06 08 1 12 14

Zrsquo (W)

Z

(W)

Z

(MW

)Z

(M

W)

35 (a)

55 (a)

65 (b)

75 (b)

81 (b)

93 (b)

SO30

0

2 105

4 105

6 105

8 105

1 106

12 106

0 2 105 4 105 6 105 8 105 1 106 12 106

17 (a)27 (a)

35 (b)45 (b)55 (b)

93 (b)

DO15

(a)

Zrsquo (MW)

(b)

(c)

Fig 15 Cole-Cole plot of the sensors at different humidity levels(a) SO50 (b) SO30 (c) DO15 The points represent the experimentaldata and lines represent the t with the equivalent circuit model givenin Figure 9 The RH values at which the data were collected and theequivalent circuit used for tting each curve (given as (a) or (b) denot-ing the circuit of Figure 17a and b respectively) are shown adjacent toeach curve

hydrogen double bonds and a proton may be transferredfrom a hydroxyl group to the water molecule to form aH3OC ion

When the physisorption occurs within less than a monolayer that is when clusters of physisorbed molecules

285

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

(b)

(a)

Cn1 (w)

R1

Cn2 (w) Cn1 (w)

R1 R2

Fig 16 Equivalent circuit used for tting the experimental data

are present at the surface it is thought that H3OC diffu-sion within hydroxyl groups and proton transfer betweenadjacent water molecules clusters take place96 HoweverKhanna and Nahar104ndash106 refute this theory They arguethat the proton diffusion is less probable in a surfacethat contains hydroxyl ions and clusters of physisorbedmolecules and hence cannot account for the conductiv-ity of porous alumina at an RH less than 40 that isbefore the rst monolayer is formed Instead they sug-gested a phonon-assisted electron tunneling mechanism atlow humidity levels in which electron tunneling occursfrom one water molecule to another as enhanced by thesurface anions of the alumina sensors

At higher humidity levels the number of physisorbedlayers increases allowing each water molecule to besingly bonded to a hydroxyl group and proton hop-ping between adjacent water molecules in the continu-ous water layer takes place The conduction process isthe same as that of pure water and is called the Grot-thuss chain reaction96 The dominant charge carrier ina high-humidity environment is therefore HC ions (pro-tons) The concentration of HC increases with increas-ing humidity and HC move freely through the water-likelayer According to the theory of Khanna and Nahar104

in porous alumina the protons for conduction are donatednot only by H2O but also by the Al(OH)3 formed asa result of water adsorption and the impurity anionsincorporated into the lm from the electrolyte duringanodization107ndash109 The water-like network consists ofsingly bonded water molecules which possess a highdielectric constant as they form dipoles and reorient freelyunder an externally applied electric eld The impedanceof the nanoporous alumina sensors includes this capaci-tive contribution as well

From the analysis of Figure 15 the high-frequencysemicircle that diminishes in size with increasing RHarises because of the effect of a physisorbed water layerat the surface Therefore R1 in the equivalent circuitof Figure 16 represents the charge transfer resistancethrough the adsorbed water layer The effect of electronic

102

104

106

108

SO50SO30

DO15

R1

(W)

0 20 40 60 80 100

RH ()

Fig 17 Humidity dependence of resistance R1 of different sensorsobtained by tting the equivalent circuit model with experimental data

conduction through the alumina bulk is parallel to thatof the surface conduction affected by the adsorbed waterlayer This bulk conduction process is dominant only atextremely low humidity values when there are only afew adsorbed molecules Hence the incomplete semicir-cle or arc in the low-humidity region is partially deter-mined from the base alumina Bn1 represents the capac-itance contribution from the adsorbed water layer Thelarge spurs seen in the low-frequency region are a con-sequence of ion migration in the adsorbed layer towardelectrodes1011 103 This migration leads to accumulation ofions at the electrodes Since the relaxation time of ionmigration is larger than the charge transfer process itis seen only at low frequencies The accumulated ionsin the adsorbed water layer give rise to R2 and Bn2From the equivalent circuits given in Figure 16 it canbe seen that the presence of ion accumulation repre-sented by the combination of R2 and Cn24mdash5 increasesthe overall impedance of the sensor Hence the reductionin impedance caused at high humidity due to the chargetransfer process is countered by ion accumulation

To better understand the fundamental sensing mecha-nisms R1 and Bn1 obtained through the numerical ttingprogram for the different sensors are plotted in Figures 17and 18 respectively It can be seen from Figure 17 thatR1 which is the resistance to charge transport in a staticelectric eld by the material containing an adsorbed waterlayer in the absence of ion migration is linear with RH

-11

10-10

10-9

10-8

10-7

50V30VDO15

Bn1

0 20 40 60 80 100

RH ()

Fig 18 Humidity dependence of Bn1 of different sensors obtained by tting the equivalent circuit model (Fig 17) with experimental data

286

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

in a semilog plot For sensor DO15 R1 decreases to afew hundred ohms at high RH levels indicating that theobserved conductivity is not due merely to the protonsprovided by the adsorbed water layer Since deionizedwater and argon gas are used to create the required humid-ity environment it is reasonable to assume that the sur-faces of the alumina lms contain anions presumablyfrom the electrolyte used for anodization which controlsconduction through the adsorbed layer108

With reference to Figure 18 Bn1 of the larger pore sam-ples SO50 and SO30 increases signi cantly only at thehigher humidity levels This result is in contrast to thatof the smaller pore size sensors which shows that Bn1

increases at lower humidity levels The relative humid-ity values where a signi cant increase in Bn1 takes placegenerally match the point seen in Figure 11 where theimpedance drops signi cantly which in turn is a func-tion of pore size These same relative humidity valuescorrelate with the point at which the spur appears inthe impedance diagrams given in Figure 15 The resultslead to the following conclusions Since hydrogen-bondedwater molecules possess a high dielectric constant theincrease in Bn1 indicates the presence of these bondson the surface which means that a liquid-like networkmore than one monolayer of physisorbed molecules hasformed at that humidity level This liquid-like network isresponsible for the sharp reduction in impedance seen inFigure 11 and the ion migration that creates the spur inthe low-frequency portion of the Cole-Cole plots as seenin Figure 15 According to Eq (1) capillary condensationdoes not take place in the alumina test sensors

In light of the above observations we propose thefollowing model to explain the behavior of nanoporousalumina in response to humidity The anodization pro-cess results in a certain amount of electrolyte anionstrapped within the pores1041 1071 108 Even in a low-humidityregime the presence of these anions on the sensor sur-face provides a high charge density for easy physisorp-tion of water molecules The result is the formationof a liquid-like network within the pores The impu-rity anions also act as proton donors by a dissociativemechanism104 joining the protons donated by the watermolecules The smaller the pore size the lower the RHlevel at which the liquid-like network is formed Hencefor the small pore sensors there is variation in Bn1 aswell as impedance magnitude in low-humidity regimesthe exponential dependence of R1 on RH also suggeststhe same phenomenon

In the case of DO15 the rst semicircle is highly dis-torted by the low-frequency spur originating with ionmigration A relatively large number of mobile ions mightbe present in these sensors leading to an enhancedaccumulation of protons at the electrode-sample contactAs mentioned earlier the dipoles in the singly bondedadsorbed water layer are free to rotate in an electric eld

Hence in the region close to the electrode-alumina inter-face the adsorbed water layer exhibits a net dipolar ori-entation and therefore it does not display its normaldielectric constant1011 103 Depending on ion size up toa monolayer of charge can exist at the electrode101 thediffuse (Gouy) layer of counter-ions spreads away fromthis inner (Stern) layer The number of ions available toform the Stern and Gouy layers at a particular frequencyincreases with increasing number of physisorbed layersHence as the number of physisorbed layers increases theStern and Gouy layers start forming at higher frequencies

With reference to Figure 12 the relative humiditylevel at which the sharp decrease in impedance occursshifts higher with increasing frequency This result can beexplained with the equivalent circuit shown in Figure 16athat is the circuit in the absence of ion accumulation Athigh frequencies the reactive impedance Xc of Cn14mdash5equal to Xc

D 1=jmdashCn14mdash5 is low and hence impedanceis largely determined by R1 As the humidity increasesR1 decreases however this translates into a reduction inimpedance only at relatively low frequencies where thevalue of R1 is comparable to or less than the capacitivereactance of Cn1(mdash) As the measurement frequency isreduced the reactance of Cn1(mdash) decreases hence the effectof R1 becomes more signi cant at lower humidity levels

32 Hydrogen Sensing with Titania Nanotubes

We recently reported62 the fabrication of self-organizedtitania nanotube arrays with an anodization techniqueAlthough the as-prepared nanotubes are amorphous theycrystallize on annealing at elevated temperatures and arestructurally stable to at least 600 C This stability ofstructure which is an essential criterion of a gas-sensingmaterial prompted us to study the gas-sensing behavior ofthese nanotubes for technologically important gases suchas oxygen carbon monoxide ammonia carbon dioxideand hydrogen

Titania has earned much attention for its oxygen-sensing capabilities110ndash115 Furthermore with propermanipulation of the microstructure or crystalline phaseandor the addition of the proper impurities or surfacefunctionalization titania can also be used as a reducinggas sensor116ndash127 The interaction of a gas with a metaloxide semiconductor is primarily a surface phenomenonand nanoporous metal oxides1221 1231 1281 129 offer the advan-tage of providing large sensing surface areas

Room-temperature metal oxide hydrogen sensors aregenerally based on Schottky barrier modulation of deviceslike PdTiO2 or PtTiO2 by hydrogen130ndash132 Elevated-temperature hydrogen sensors examine electrical resis-tance with hydrogen concentration for example Birke-feld et al133 observed that the resistance of the anatasephase of titania varies in the presence of carbon monoxideand hydrogen at temperatures above 500 C but on dop-ing with 10 alumina it becomes selective for hydrogen

287

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Barrier layer

Titanium Nano tube array

Platinum electrodes

Fig 19 Schematic representation of the hydrogen sensor geometry

Munuera et al134 used rhenium doping in the anatasephase to perform room-temperature hydrogen monitor-ing Recently Shimizu et al122 reported that anodizednanoporous titania lms with a Pd Schottky barrier aresensitive to hydrogen at 250 C

Titania nanotubes were grown from titanium foil(995 pure from Alfa Aesar Ward Hill MA) with athickness of 025 mm The anodization was performed inan electrolyte medium of 05 hydro uoric acid in waterwith the use of a platinum foil cathode Well-de nednanotube arrays were grown with anodizing potentialsranging from 12 V to 20 V Nanotube length increaseswith anodization time reaching 400 nm in approximately20 min and then remains constant For the present studythe samples were anodized for 25 min The samples werethen annealed at 500 C in a pure oxygen ambient for 6 hwith a heating and cooling rate of 1 Cmin

The electrode geometry of the titania nanotube sen-sors is shown in Figure 19 The sensor consists of a basetitanium metal foil with a nanotube array grown on topAn insulating barrier layer separates the nanotubes fromthe conducting titanium foil Electrical connections weremade by two parallel 10 mm pound 2 mm platinum pads of100-Œm thickness The test chamber consists of a 13-literquartz tube with stainless steel end caps placed insidea furnace (Thermolyne USA model 21100 tubular fur-nace) Gas ow through the test chamber was controlledvia a computer-controlled mass ow controller The elec-trical resistance of the titania sensors was measured witha computer-controlled Keithly digital multimeter Prior todata collection the test chamber was initially evacuatedwith a mechanical pump whereupon nitrogen (99999pure) was passed while the sensor under test was heatedto the desired temperature The test gases were mixed inappropriate ratios with nitrogen to create the necessarytest gas ambient

The surface morphology of nanotube arrays preparedwith an anodization potential of 20 V and annealed at500 C for 6 h in a pure oxygen ambient is shown inFigure 20a and b It can be seen from these images thatthe nanotubes are uniform over the surface The nano-tubes are approximately 400 nm in length and have abarrier layer62 thickness of ordm50 nm For the nanotubes

100 nm

(a)

(b)

1 mm (c)

100 nm Fig 20 The surface morphology of the titania nanotubes after anneal-ing at 500 C (a) High- and (b) low-magni cation images of a 20-Vsample (c) A high-magni cation image of a 12-V sample

fabricated with 20-V anodization the average pore diam-eter as determined from FESEM images is 76 nm (thestandard deviation 15 nm) with a wall thickness of 27 nm(standard deviation 6 nm) The sample anodized at 12 Vshown in Figure 2c was found to have an average porediameter of 46 nm (standard deviation 8 nm) with a wallthickness of 17 nm (standard deviation 2 nm) The porosi-ties of the 20-V and 12-V samples were calculated as 45and 61 respectively

Glancing-angle X-ray diffraction patterns of a 20-Vsample annealed at 500 C for 6 h in oxygen ambientis shown in Figure 21 It can be seen that both anataseand rutile phases of titania are present in the sampleA detailed study135 of these structures by high-resolutiontransmission electron microscope (HRTEM) showed that

288

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

20 30 40 50 60 70 80

2 q (deg)

(101

)

(110

)

A

R

TR

T

T

R R A TRR T R R T TT

Fig 21 Glancing-angle X-ray diffraction pattern of a 20-V sample(glancing angle D 2) annealed at 500 C in oxygen ambient A R andT represent respectively the re ections from anatase crystallites rutilecrystallites and the titanium substrate

the anatase crystallites were concentrated on the walls ofthe nanotubes and rutile on the barrier layer

Nanotubes annealed in a pure oxygen ambient werefound to be stable (intact) to temperatures of approxi-mately 580 C Above this temperature protrusions wereseen to be coming out through the nanotubes an effectthat spread with increasing temperature These protru-sions which are due to oxidation of the titanium substratecollapse the nanotubes

Figure 22 shows the response of the 20-V (76 nm) sam-ple as a function of ambient temperature as it is switchedfrom a nitrogen environment to one containing 1000 ppmhydrogen and then back to nitrogen The plot was madeusing 4Rg=R05

ƒ1 versus time where R0 is the base resis-tance of the sensor that is the sensor resistance before thetest gas was introduced and Rg is the measured resistancein the presence of test gas The sensor shows increasinghydrogen sensitivity with temperature with a change by3 orders of magnitude in resistance at temperatures aboveordm300 C At all of the temperatures the original resistancerecovered without hysteresis

The sensitivity S is de ned by the formula

S DR0

ƒ Rgs

Rgs

10 0

10 1

10 2

10 3

0 10 0 5 10 2 1 10 3 15 103 2 10 3 25 103 3 10 3 35 103 4 10 3

Time (Sec)

(Rg

R0)

-1

180degC

240degC

265degC

290degC

345degC

375degC

H2+N

2

N2

Fig 22 Variation of resistance Rg normalized with respect to baselineresistance R0 of a 20-V sample with time on exposure to 1000 ppmhydrogen at different temperatures It may be noted that the inverse ofRg=R0 was used in the plot to represent data in the positive y direction

10-1

100

101

102

103

150 200 250 300 350 400 450

Temperature (degC)

Sens

itivi

ty (

R0-

Rgs

)R

gs

Fig 23 The sensitivity temperature dependence of a 20-V sample to1000 ppm hydrogen

where R0 is the resistance of the sensor before the gas ispassed and Rgs is the resistance after gas is passed andreaches the saturation value The temperature-dependentsensitivity of a 20-V sample to 1000 ppm hydrogen isshown in Figure 23 Sensitivity is seen to increase withtemperature to approximately 380 C where the increasein sensitivity with temperature begins to saturate

The response time de ned as the time needed for thesensor to reach 90 of the nal signal for a given concen-tration of gas is plotted against temperature in Figure 24(the time includes that required for the gas to equili-brate inside the measurement chamber estimated to beordm30 s) The response time decreases exponentially withtemperature

To check the behavior of the sensor on repeated hydro-gen exposure the hydrogen concentration was varied indiscrete steps of 100 ppm from 0 to 500 ppm while thetemperature was kept constant at 290 C the chamber was ushed with nitrogen after each exposure to hydrogenThe response of the 20-V prepared nanotube sensorskept at 290 C is shown in Figure 25 The behaviorof the sensor is consistent recovering its original resis-tance after repeated exposure to varying hydrogen gasconcentrations The sensitivities of the 76-nm- 53-nm-and 22-nm-diameter nanotube sensors to hydrogen con-centrations ranging from 100 ppm to 1 are shown inFigure 26 The sensitivity of the 76-nm (20-V) sensor tolow concentrations is shown in Figure 27 there is a linearincrease in sensitivity at low concentrations The lower

10 1

10 2

10 3

10 4

150 200 250 300 350 400 450

Temperature (degC)

Res

pons

e T

ime

(Sec

)

Fig 24 Variation in response time for a 20-V sensor as a function oftemperature The dots represent measured data

289

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

10 2

0 4 4 4 4 4

10 3

0 10 1 10 2 10 3 10 4 10 5 10

Time (Sec)

Res

ista

nce

(W)

H2 (ppm)

200

300

400

500

400

300

200

100

100

Fig 25 Resistance of a 20-V sample when exposed to different con-centrations of hydrogen at 290 C The nitrogen-hydrogen mixture waspassed for 1500 s the chamber was then ushed with nitrogen for 3000 sbefore the nitrogen-hydrogen mixture was passed again

test limit of 20 ppm is determined by the experimentalapparatus and not the sensor

Figure 28 shows the change in resistance of the 76-nm-46-nm- and 22-nm-diameter nanotube sensors with expo-sure to 1000 ppm hydrogen at 290 C Whereas smallerdiameter nanotubes had greater sensitivity to hydrogenthe samples made at lower anodizing voltages tended tobe more brittle and harder to handle mechanically withoutbreaking

The 20-V 76-nm-diameter sample was exposed to oxy-gen carbon monoxide ammonia and carbon dioxide at290 C to study the cross-sensitivity The sensor wasfound to have no detectable variation in resistance onexposure to carbon dioxide The sensitivities of the tita-nia nanotubes to the other gases are shown in Figure 29The sensitivities of the nanotubes to carbon monoxide andammonia are negligible compared with that of hydrogenThe resistance of the nanotubes increased in the presenceof oxygen and did not regain their original electrical con-ductivity even after several hours in a 290 C nitrogenenvironment

1

10

100

1000

104

105

0 02 04 06 08 1

Hydrogen Concentration ()

Pore diameter = 76 nm

Pore diameter = 53 nm

Pore diameter = 22nm

Sen

sitiv

ity (

R0-

Rg)

Rg

Fig 26 The sensitivity variation of a samples with pore diameters of22 nm 53 nm and 76 nm versus time and hydrogen concentrations

0 100 200 300 400 5001

10

100

Hydrogen concentration (ppm)

Sen

sitiv

ity

(R0-

Rgs

)R

gs

Fig 27 Sensitivity of a 20-V (76 nm) sample at 290 C for low hydro-gen concentrations

Since the sensor measurements were made in atmo-spheres without oxygen the increase in conductivitycannot be due to hydrogen removing oxygen from thelattice132ndash134 or the removal of chemisorbed oxygen135ndash137

Hydrogen molecules can dissociate at defects on thetitania surface which can subsequently diffuse into thetitania lattice acting as electron donors1281 1381 139 How-ever a dissociation-driven process would result in aslow response and recovery times with complete recov-ery unlikely since the sensor completely regains its orig-inal resistance with hydrogen cycling it appears that thisis not the dominant mechanism behind high hydrogensensitivity We believe chemisorption of the dissociatedhydrogen on the titania surface is the underlying sensingmechanism137 During chemisorption hydrogen acts as asurface state and a partial charge transfer from hydrogento the titania conduction band takes place This createsan electron accumulation layer on the nanotube surfaceenhancing its electrical conductance Upon removal of thehydrogen ambient electron transfer back to the hydro-gen molecule takes place which subsequently desorbsrestoring the original electrical resistance of the mate-rial Another factor that may play a role in the hydrogensensitivity (and selectivity) is the platinum electrodes Atelevated temperatures hydrogen dissociation can occur onplatinum surfaces These dissociated hydrogen atoms mayspill1361 140 onto the nanotube surface where they diffuse

10-1

101

103

105

0 05 1 15

Time (Hour)

Pore size - 76 nm

Pore size - 53 nm

Pore size - 22 nm

(Rg

R0)

-1

Fig 28 A comparison of the variation in resistance of samples havingpore diameters of 22 nm 53 nm and 76 nm versus time upon exposureto 1000 ppm of hydrogen at 290 C

290

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

10-2

10-1

100

0 100 2 102 4 102 6 102 8 102 1 103

OxygenAmmoniaCarbon monoxide

Time (Sec)

(Rg

R0)

-1

Fig 29 Variation of resistance of a 20-V sample when exposed to1000 ppm carbon monoxide 5 ammonia and 1 oxygen at 290 C

into the material affecting its electrical properties Fromthe present study it was not clear how signi cant a rolethe platinum electrodes play

Anatase the polymorph of titania has been reported tohave high sensitivity for reducing gases like hydrogen andcarbon monoxide1211 1241 133 Our nanotube samples containanatase phase mainly on the walls and rutile in the barrierlayer As the diffusing hydrogen atoms go to the intersti-tial sites1331 142 and as the ca ratio of anatase is almostfour times that of rutile it appears that the anatase latticeaccommodates hydrogen easily and hence makes a highercontribution to hydrogen sensitivity

The effect of chemisorption can be neglected in theoxygen-sensing experiments As the recovery requiresseveral hours it appears that the nanotubes contain oxy-gen vacancies or titanium interstitial defects in the pres-ence of nitrogen On exposure of the sensor to oxygenambient the lattice reoxidizes and hence the conductiv-ity of the sensor decreases On the removal of oxygenthe reduction of the lattice will not immediately occurhence the sensor requires several hours to regain its orig-inal conductivity

4 CONCLUSIONS

This paper summarizes different routes of fabrication ofmetal oxide nanoarchitectures useful for sensing applica-tions Structures made via anodization of a starting metaland breath gure formation from a sol-gel are discussedand presented as are TiO2 nanowires made by lling ananoporous alumina membrane with subsequent removalof the supporting template Two test cases illustratingthe utility of these nanoporous metal oxide structures forsensing are presented highly ordered nanoporous aluminais used for humidity sensing and TiO2 nanotubes are usedfor hydrogen sensing These study test cases reveal sev-eral points worth brie y noting

In the rst sensing test case uniform nanoporous alu-mina sensors exhibit pore size and frequency-dependent

humidity-sensing characteristics The mechanism respon-sible for the sensing action of the nanoporous alumina sen-sors was studied by representing the impedance spectra atdifferent humidity levels in Cole-Cole plots This makes itpossible to identify the competing impedance-determiningprocesses occurring in the sensors namely charge transferand mass transport through the physisorbed water layeras dependent upon the humidity level and input signalfrequency Charge transfer was found to enhance the sen-sitivity by causing a large variation in impedance withrespect to humidity Mass transport leads to accumulationof ions at the electrodes which tends to decrease the mea-surement sensitivity by increasing the impedance Hencea proper choice of operating frequency high enough toavoid ion accumulation is essential for optimum sensorperformance

It has been inferred from our results that anions becomeincorporated into the alumina lm during the anodiza-tion process supporting the hypothesis of Khanna andNahar104ndash106 where they act as proton donors and facili-tate the adsorption of water molecules Hence liquid-likecharge transfer networks are formed at relatively lowerhumidity levels for the smaller pore-size sensors Thisprocess rather than capillary condensation is primarilyresponsible for the observed humidity-sensing behavior ofthe sensors Our studies show that a distribution of poresizes as suggested by Shimizu et al94 is not necessary toobtain sensitive performance over a wide humidity rangerather such optimal performance can be obtained with uni-form nanodimensional pores containing adsorbed anionicimpurities

In the second sensing test case titania nanotubes pre-pared by anodization and annealed in an oxygen atmo-sphere at a temperature of 500 C were found to behighly sensitive to hydrogen The nanotube sensors con-tained both anatase and rutile phases of titania and showedappreciable sensitivity toward hydrogen at temperaturesas low as 180 C The sensitivity increased drasticallywith temperature showing a variation of 3 orders in mag-nitude of resistance to 1000 ppm of hydrogen at 400 CThe response time decreased with increasing temperatureat 290 C full switching of the sensor took approximately3 min Results were highly reproducible with no indica-tion of hysteresis Our results showed that these sensorsare at least capable of monitoring hydrogen levels from20 ppm to 4 At 290 C nanotubes with smaller porediameters showed higher sensitivity to hydrogen com-pared with their larger pore counterparts The sensorsshowed high selectivity to hydrogen compared with car-bon monoxide ammonia and carbon dioxide Althoughthe sensor was sensitive to high concentrations of oxy-gen the response time was high and the sensor did notcompletely regain the original condition We believe thehydrogen sensitivity of the nanotubes is due to hydrogendiffusion into the titania lattice where they act as electrondonors

291

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Acknowledgments Support of this work by the NationalScience Foundation through grant ECS-0210033 is grate-fully acknowledged

References and Notes

1 K Kajihara K Tanaka K Horao and N Soga Jpn J ApplPhys 35 6110 (1996)

2 B O Regan and M Gratzel A low cost Nature 353 737 (1991)3 A Fujishima and K Honda Nature 238 37 (1972)4 K Sato A Tsuzuki H Taoda Y Torii T Kato and Y Butsugan

J Mater Sci 29 5911 (1994)5 K Sato A Tsuzuki Y Torii H Taoda T Kato and Y Butsugan

J Mater Sci 30 837 (1995)6 N Ozer and C M Lampert Sol Energy Mater Sol Cells 54 147

(1998)7 A Manz and H Becker editors Microsystem Technology in

Chemistry and Life Sciences Springer-Verlag Berlin 19988 E Kim Y Xia and G M Whitesides Nature 376 581 (1995)9 M Templin A Franck A Du Chesne A Leist A Zhang

R Ulrich V Schadler and U Weisner Science 278 1795 (1997)10 D Zhao J Feng Q Huo N Melosh G H Frederickson

B Chmelka and G D Stucky Science 279 548 (1998)11 O D Velev T A Jede R F Lobo and A M Lenhoff Nature

389 447 (1997)12 B T Holland C F Blanford and A Stein Science 281 538

(1998)13 S Matsushita T Miwa and A Fujishima Chem Lett 9 925

(1997)14 A Imhof and D J Pine Adv Mater 10 697 (1998)15 T Tatsuma A Ikezawa Y Ohko T Miwa T Matsue and

A Fujishima Adv Mater 12 643 (2000)16 Z Zhong Y Yin B Gates and Y Xia Adv Mater 12 206 (2000)17 O D Velev and A M Lenhoff Curr Opin Colloids Interface

Sci 5 56 (2000)18 O D Velev and E W Kaler Adv Mater 12 531 (2000)19 P Jiang J Cizeron J F Bertone and V L Colvin J Am Chem

Soc 121 7957 (1999)20 M E Turner T J Trentler and V L Colvin Adv Mater 13 180

(2001)21 K M Kulinowski P Jiang H Vaswani and V L Colvin Adv

Mater 12 833 (2000)22 P Jiang J F Bertone and V L Colvin Science 291 453 (2001)23 R Seshadri and F C Meldrum Adv Mater 12 1149 (2000)24 F C Meldrum and R Seshadri Chem Commun 29 (2000)25 Q B Meng Z Z Gu O Sato and A Fujishima Appl Phys

Lett 77 4313 (2000)26 G Widawski B Rawiso and B Francois Nature 369 3897

(1994)27 B Francois O Pitois and J Francois Adv Mater 7 1041 (1995)28 O Pitois and B Francois Eur Phys J B 8 225 (1999)29 S A Jenekhe and X L Chen Science 283 372 (1999)30 M Shimomuara and T Sawadaishi Curr Opin Colloid Interface

Sci 6 11 (2001)31 O Karthaus N Maruyama X Cieren M Shimomura

H Hasegawa and T Hashimoto Langmuir 16 6071 (2000)32 T Nishikawa J Nishida R Ookura S Nishimura S Wada

T Karino and M Shimomura Mater Sci Eng C 10 141 (1999)33 K Kajihara K Nakanishi K Tanaka K Hirao and N Soga

J Am Ceram Soc 81 2670 (1998)34 C A Grimes R S Singh E C Dickey and O K Varghese

J Mater Res 16 1686 (2001)35 M Srinivasrao D Collings A Phillips and S Patel Science 292

79 (2001)36 P G de Gennes Rev Mod Phys 57 827 (1985)

37 D Beysens A Steyer P Guenoun D Fritter and C M KnoblerPhase Transitions 31 219 (1991)

38 B J Briscoe and K P Galvin J Phys D 23 422 (1990)39 B J Briscoe and K P Galvin J Phys D 23 1265 (1990)40 D Beysens and C M Knobler Phys Rev Lett 57 1433 (1986)41 F Family and P Meakin Phys Rev Lett 61 428 (1988)42 A V Limaye R D Narhe A M Dhote and S B Ogale Phys

Rev Lett 76 3762 (1996)43 P Dellrsquo Aversana and G P Neitzel Phys Today 51 38 (1998)44 P Dellrsquo Aversana J R Banavar and J Koplik Phys Fluids 8 15

(1986)45 P Dellrsquo Aversana V Tontodonato and L Carotenuto Phys Fluids

9 2475 (1997)46 M Q Li S Q Xu and E Kumacheva Langmuir 16 7275 (2000)47 P Freymuth Rev Sci Instrum 64 1 (1993)48 P Ball The Self-Made Tapestry Pattern Formation in Nature

Oxford University Press New York 199949 M F Schatz S J VanHook W D McCormick J B Swift and

H L Swinney Phys Fluids 11 2577 (1999)50 S R Coriell M R Cordes and W J Boettinger J Cryst Growth

49 13 (1980)51 T S Sullivan Y M Liu and R E Ecke Phys Rev E 54 486

(1996)52 J L Rogers M F Schatz J L Bougie and J B Swift 84 87

(2000)53 K Nitschke 52 R5772 (1995)54 E Bodenschatz J R De Bruyn G Ahlers and D S Cannell

Phys Rev Lett 67 3078 (1991)55 C W Meyer D S Cannell G Ahlers J B Swift and P C

Hohenberg Phys Rev Lett 61 947 (1988)56 R F Ismagilov D Rosmarin D H Gracias A D Stroock and

G M Whitesides Appl Phys Lett 79 439 (2001)57 E A Barringer and H K Bowen Langmuir 1 420 (1985)58 M Tschapek C Wasowski and R M T Sanchez J Electroanal

Chem 74 167 (1976)59 H Masuda and K Fukuda Science 268 1466 (1995)60 H Masuda F Hasegwa and S Ono J Electrochem Soc 144

L 127 (1997)61 O Jessensky F Muller and U Gosele Appl Phys Lett 72 1173

(1998)62 D Gong C A Grimes O K Varghese W Hu R S Singh

Z Chen and E C Dickey J Mater Res 16 3331 (2001)63 B M Kulwicki J Am Ceram Soc 74 697 (1991)64 H Arai and T Seiyama in Sensors A Comprehensive Survey

edited by W Gopel J Hesse and J N Zemel VCH Weinheim(1992) Vol 3 pp 981ndash1012

65 E Traversa Sens Actuators B 23 135 (1995)66 N Yamazoe and Y Shimizu Sens Actuators 10 379 (1986)67 A Bearzotti I Fratoddi L Palummo S Petrocco A Furlani

C Lo Sterzo and M V Russo Sens Actuators B 76 316 (2001)68 Y Sakai Y Sadaoka and M Matsuguchi Sens Actuators B 35

85 (1996)69 C A Grimes and D Kouzoudis Sens Actuators A 84 205

(2000)70 K G Ong C A Grimes C L Robbins and R S Singh Sens

Actuators A 93 33 (2001)71 T Seiyama N Yamazoe and H Arai Sens Actuators 4 85

(1983)72 L Ketron Ceram Bull 68 860 (1989)73 E Traversa G Gusmano and A Montenero Eur J Solid State

Inorg Chem 32 719 (1995)74 M K Jain M C Bhatnagar and G L Sharma Sens Actuators

B 55 180 (1999)75 O K Varghese and L K Malhotra J Appl Phys 87 7457 (2000)76 G Sberveglieri R Murri and N Pinto Sens Actuators B 23

177 (1995)

292

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

fabrication of porous TiO2 lms with the use of analkoxide-based sol-gel containing polyethylene glycolwhich served as the template into which the substrateswere dip coated

A disadvantage of using a template is that the dom-inant length scale of the resulting porous structure is xed by the template size therefore dynamic control ofthe length scale becomes almost impossible Our inter-est lies in the fabrication of nanoporous lms with-out the use of a template for example we recentlyreported the fabrication of metal oxide lms with magnet-ically modulated nanodimensionalporosity34 The absenceof a template makes the ambient humidity level andthe sol pH key process variables for the fabrication ofcontrolled structures35 We describe here the fabricationof TiO2 mesoporous structures via sol-gel breath gureformation36ndash48 as well as thermocapillary and surfacetension-driven Benard-Marangoni convection49ndash56

The overall hydrolysis and condensation reaction fortitanium isopropoxide the sol-gel precursor for TiO2 canbe represented as Ti(OC3H7)4 C 2H2O TiO2 C4HOC3H7 The condensation reaction leads to the for-mation of colloidal particles which can be polymeric orparticulate depending on the type of precursors and pHof the sol Colloidal particle aggregation may be inhib-ited by the formation of surface charge developed byeither preferential dissociation of one of the lattice ionsof the sol particle or preferential adsorption of chargedspecies from solution The surface charge is formed byeither protonation (Ti-OH C HC Ti-OHC

2 5 or depro-tonation (Ti-OH C OHƒ Ti-Oƒ C H2O) of the Ti-OHbonds The pH at which the surface is electrically neutralis called the point of zero charge (PZC) The surface isnegatively charged at a pH gt PZC and positively chargedat pH lt PZC For TiO2 the PZC varies at values at leastbetween 52 (Ref 57) and 55 (Ref 58)

In a nitrogen environment the precursor titaniumtetraisopropoxide (TTIP) was dissolved in isopropanolto which deionized water and then nitric acid wereadded The reagents used in the experiment namely TTIP(99999) isopropanol (995) and nitric acid (70redistilled) were procured from Aldrich The solution wascontinuously stirred for 2 h and then stored in a nitrogenenvironment In a typical preparation of 01 M TiO2 sol1 ml of TTIP 005 ml of HNO3 (70 distilled) 01 ml ofdeionized water and 327 ml of isopropanol were usedThe lms were deposited by either dip coating of theglass or silicon substrate or simply by placing a dropletof solution on a clean substrate The ambient humidityin which the TiO2 lms dried was controlled by passingnitrogen through a room-temperature bubbler The humid-ity and temperature of the chamber were monitored with adigital hydrometer Upon drying all lms were annealedat 100 C for 1 h in a nitrogen environment

A 001 M sol was modi ed by a 12 nitric aciddeionizedwater solution with 15 Œl of acidwater solution added to

the original 3385 ml sol the pH of the resulting solutionwas approximately 05 Topology formation is found to bea function of (1) ambient humidity (2) atmosphere owvelocity during drying (3) sol concentration and (4) solpH Figure 2andashc illustrates the effect of drying rate on thestructure Figure 2a shows the fastest drying rate and Fig-ure 2c the slowest (the dark areas are pores ie absenceof lm) In the initial stages of the lm growth process thecondensed droplets grow as isolated objects without inter-action between neighbors as seen in Figure 2a Longerdrying times permit the drops to begin to coalesce as seenin Figure 2b with further coalescence seen in Figure 2cBreath gures36ndash48 are formed when a liquid surface is

Fig 2 Drying-time-dependent FE-SEM images of a thin lmdeposited from a 01 M TiO2 sol 1 ml of TTIP 005 ml of HNO3 (70distilled) 01 ml of deionized water and 327 ml of isopropanol towhich 15 Œl of a 12 nitric aciddeionized water solution was added(a) the fastest drying region and (c) the slowest drying region wherethe liquid has coalesced into drops

279

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

brought into contact with moist air the solvent vapor pres-sure and air velocity across the surface drive solvent evap-oration rapidly cooling the surface This in turn facili-tates the nucleation and growth of water droplets from theatmosphere36ndash41 The temperature difference between theliquid surface and air results in thermocapillary convec-tion within the liquid that acts to stabilize the condensingwater droplets on or at the solution surface43ndash48 Air owacross the surface coupled with surface convection cur-rents drives the ordering of water droplets into hexagonalarrays44ndash46 Once the surface is completely covered withwater droplets the temperature difference between the sur-face and the droplets diminishes and the droplets beingdenser than the solvent sink into the solution de ning theresidual structure Figure 3 shows the effect of sol pHshowing the pH-dependent variation in the resulting struc-ture achieved

Benard-Marangoni convection typically results from avertical temperature gradient due to solvent evaporationfrom a thin liquid lm46ndash54 Preferential solvent evapo-ration removes heat from the lm surface resulting inBenard-Marangoni convection50ndash56 across the uid layerwith the suspension welling up in the center of a convec-tion cell and then owing back down the cell boundary Asevaporation proceeds the colloidal suspension becomesmore concentrated changing the convection character-istics and in turn de ning the ultimate lm structureWell-de ned hexagonal pentagonal or square patternsare expected from an ideal time-independent Benard-Marangoni convection50 in a homogeneous liquid whereasthe cell patterns of our lms seen in Figure 2 are typicallyirregular This deviation of structure from the theoreticalideal may be the result of time-dependent convection owthrough changes in sol viscosity with preferential evapora-tion of the propanol solvent leading to coupled thermoso-lutal Benard-Marangoni convection51 in the nonhomoge-neous uid

22 Nanotube Nanopore and NanowireFabrication via Anodization of Al and Ti

Highly ordered nanoporous alumina lms (see Fig 4aandb)aremadethrougha two-stepanodizationprocess59ndash61

The aluminum substrate is rst anodized in an oxalic orsulfuric acid solution The anodization is stopped after afew microns of aluminum are consumed and the porousalumina lm is removed through etching The etchant amixture of chromic acid and phosphoric acid is highlyselective attacking alumina much faster than aluminumThe remaining aluminum is dimpled with the dimplesserving as a uniform seed layer upon which a highly uni-form porous layer can then be achieved through a sec-ond anodization step at the same voltage Alumina tem-plates are widely used as templates for the fabrication ofnanowires Figure 4c shows a TiO2 nanowire mat fab-ricated by lling pores of an alumina membrane via sol

Fig 3 Images taken of a dried 01 M TiO2 sol 1 ml of TTIP 005 mlof HNO3 (70 distilled) 01 ml of deionized water and 327 ml ofisopropanol to which varying amounts of a 12 nitric aciddeionizedwater solution were added to vary pH values (a) pH 017 (b) pH 029(c) pH 037

gel letting the lm dry and then subsequently removingthe alumina template by a sodium hydroxide etch

Anodization has also been used to fabricate TiO2 nano-tube arrays62 with pore size linearly proportional toanodization voltage (see Fig 5) The nanotube array inFigure 5 was fabricated by anodization of titanium at 20 Vin 05 wt HF solution for 20 min resulting in a well-aligned titanium oxide nanotube array with an approxi-mate average tube diameter of 60 nm and a tube lengthof 400 nm Diameters of fabricated tubes have ranged insize from 25 nm to 65 nm

280

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

Fig 4 (a) The surface of a two-step anodized alumina membrane (the akes seen in the image are the ldquodustrdquo associated with tissue paper)(b) An image of a cleaved sample showing the pore channels (c) TiO2

nanowires made by lling an alumina template via sol-gel then subse-quently removing the template by a sodium hydroxide etch resulting ina TiO2 nanowire mat

3 EXPERIMENTAL RESULTSAPPLICATION TO SENSING

31 The Water Vapor Sensing Performanceof Highly Ordered NanoporousAlumina Films

Humidity sensors have attracted considerable attentionover many years because of their great importancein applications ranging from monitoring food qual-ity to meteorological studies63ndash65 Ceramic humiditysensors651 71ndash82 are commercially available and offer majoradvantages with high resistance to chemical attack

Fig 5 TiO2 nanotubes made by anodization of titanium

thermal stability mechanical strength and quick responseHowever ceramic humidity sensors still suffer from insuf- cient sensitivity over wide humidity ranges as well aslack of reversibility and drift in base resistance with timebecause of water molecule chemisorption

The humidity-sensing properties of alumina discoveredalmost 50 years ago83ndash94 are based upon ionic conduc-tion the presence of an adsorbed layer of water at thesurface reduces the total sensor impedance because of theincrease in the ionic conductivity as well as capacitancedue to the high dielectric constant of water The humidity-sensing characteristics of a given sensing element dependupon the material used the method of preparation andthe resulting surface topology Porous alumina is usuallypreferred for sensing applications because of the largesurface area available for water adsorption

An illustration of water adsorption by a surfaceredrawn from Ref 95 is shown in Figure 6 Figure6a shows the adsorption of a water molecule on aclean alumina surface Physisorption of water initiallyoccurs on an active surface site and forms an adsorption

Fig 6 Schematic illustration of interaction of a water molecule withan alumina surface redrawn from Ref 95

281

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Fig 7 Schematic illustration of buildup of adsorbed water layersredrawn from Ref 96 building of a water layer upon an alumina surface

complex (Fig 6a and b) The negatively charged oxygenof the water molecule becomes attached to the positivelycharged cation site and one of the positively chargedhydrogen atoms becomes attached to the anionic site byelectrostatic attraction Eventually they are converted totwo chemisorbed hydroxyl (OH) groups (Fig 6c) Anincrease in humidity makes the water molecules physisorbto this chemisorbed hydroxyl layer The effectiveness ofphysisorption depends upon the cation charge complexes(from alumina or the surface impurities) and the hydroxylions on the surface of the alumina The physisorptionprocess is facilitated by higher surface charge densitiesAfter hydroxyl formation the next water molecule willbe physisorbed through hydrogen double bonds on thetwo neighbouring hydroxyl groups (Fig 6d) and a protonmay be transferred from a hydroxyl group to the watermolecule to form a H3OC ion At higher humidity levelseach water molecule will be singly bonded to a hydroxylgroup as illustrated in Figure 7 (redrawn from [96]) andproton hopping between adjacent water molecules in thecontinuous water layer takes place The conduction pro-cess in this case occurs by attachment of a proton to awater molecule thereby forming a hydronium ion thehydronium ion releases another proton to a second watermolecule that accepts this proton while releasing a thirdproton and so on through the liquid Hence the dominantcharge carrier in a high-moisture atmosphere is the HC

(proton) The concentration of HC increases with increas-ing humidity and it moves freely through the water-likelayer

In the case of porous alumina capillary condensationcan take place in pores with a radius up to rK at a partic-ular relative humidity and temperature which is given byKelvinrsquos relation

rKD 2ƒM

RT ln4Ps=P5

Here P is the water vapor pressure Ps is the water vaporpressure at saturation ƒ is the surface tension R is theuniversal gas constant T is the temperature in Kelvin

H O

H O

O

H

H +

H

H

O

H +H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H O

H O

O

H

H +

H

H

O

H +H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H O

H O

O

H

H +

H

H

O

H+H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H

OH

O

O

HH

O

H+H

H+

H

OH

OH

O

O

HH

O

H+

HH

+H

O

H

O

H

O

O

HH

O

H+ HH+ H

O

H

O

H

O

O

HH

O

H+ HH+ H

O

Aluminium

AluminaCapillary Condensation

Capillary Condensation Adsorption

Adsorbed Water Layer

Fig 8 Schematic representation of water vapor interaction withnanoporous alumina illustrating some but not all97 of the multiple fac-tors inherent in determining the response

and and M are respectively the density and molecularweight of water

According this relation the condensation occurs insidesmall pores at lower humidity levels as illustrated inFigure 8 which is an illustrative drawing presentingthe different water-surface interaction mechanisms (NoteFigure 8 is not meant to be an exhaustive considera-tion of all possible effects for example the drawingdoes not address effects associated with capillary surfacecurvature97)

This condensed water reduces the impedance of alu-mina considerably No condensation occurs inside poresof radius greater than rK at a particular humidity level (seethe largest pore shown in Fig 8) In such pores as in thecase of a plane surface it is the number of physisorbedlayers that decides the impedance

Since capillary condensation enhances the sensingcapabilities of a material pore size distribution has beenwidely considered to be an important parameter in deter-mining the sensitivity in a particular humidity range941 98

However the results of our study on uniform nanoporousalumina lms presented below show that an easily builtand highly reproducible wide-range humidity sensor canbe achieved with nanodimensional pores of a narrow sizedistribution made by anodization of aluminum59ndash611 99

Starting from adhesive-backed aluminum tape (99pure) purchased from Tesa Tape Inc100 and subsequentlyanodized sensors SO50 SO30 and DO15 were preparedunder the conditions given in Table I Per sample nota-tion the rst letter denotes a single S or double D stepanodization process the second letter denotes the use of

Table I Anodization conditions used for making nanoporous alumina lms

Anodizing Duration of No ofvoltage anodization anodization

Sample (v) (min) stages

SO50 50 120 1SO30 30 120 1DO15 15 120 2

The electrolyte used in all cases was 2 wt oxalic acid

282

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

oxalic acid (O) in the anodization bath the two-digit num-ber denotes the anodization voltage It was observed thatsingle-step anodization results in disordered pore struc-tures at low voltages Hence a double-step process wasused to make sample DO15 These samples were initiallyanodized under the conditions given in Table I The sam-ples were then immersed in an aqueous solution contain-ing 18 wt chromic acid and 4 wt phosphoric acid for12 h to remove the initial nanoporous layer while keep-ing the nonporous bottom layer of alumina intact Thesamples were rinsed in distilled water and again anodizedunder the same conditions used for the rst anodizationstep Estimates of pore size distribution were determinedfrom FE-SEM images and are shown in Figure 9andashcas a size distribution histograph Samples SO50 SO30and DO15 have respectively average pore diameters of

0

10

20

30

375 40 425 45 475 50

o

f po

res

Pore diameter (nm)

0

10

20

30

25 325 375 425 475 525

o

f po

res

0

5

10

15

20

25

175 20 225 25 275 30 325 375

DO15

o

f po

res

(a)

Pore diameter (nm)

(c)

Pore diameter (nm)

(b)

Fig 9 Pore size distribution of the tested sensors (a) SO50 (b) SO30(c) DO15

Fig 10 Digital image of an impedance-based interdigital nanoporousalumina humidity sensor

452 nm 384 nm and 136 nm with a correspondingstandard deviation of 34 nm 78 nm and 49 nm Allsamples have a predominantly ordered pore structure

Immediately after anodization all samples were thor-oughly washed in distilled water and then dried in a nitro-gen atmosphere at 100 C An interdigital capacitor pat-tern with dimensions of 12 cm pound 12 cm was formed onthe surface of the alumina lms by the evaporation of gold(thickness ordm 250 nm) through a mask A digital cameraimage of a typical sensor is shown in Figure 10 To createthe necessary ambient humidity argon passed through amass ow controller (MFC) was bubbled through a bot-tle containing deionized water and then mixed with dryargon coming from another MFC in appropriate ratiosbefore being passed to the test chamber A constant total ow was maintained throughout the experiment Cham-ber humidity was monitored with the humidity probeof a Keithley 6517A electrometer and all measurementswere carried out at room temperature (23 C) Sensorimpedance was measured over the frequency range 5 Hzto 13 MHz with a computer-controlled Hewlett Packardimpedance analyzer (4192A) tted with an impedance test xture (Agilent 16034E) A signal amplitude of 90 mVwas used for all measurements Electrical contact wasmade between the sensor under test and the impedanceanalyzer test xture with 42-gauge 5-cm-long jumpersattached to the interdigital electrodes with silver pasteThe contacts were subsequently annealed at 100 C for2 h to cure the silver paste

Figure 11 shows the measured 5-kHz sensor impedanceof the different sensors as a function of humidity Thehumidity-sensitive region of DO15 is sup145 to 95 RHSensors SO30 and SO50 become sensitive to humidity atapproximately 65 and 75 RH respectively A smallerpore size increases the range of humidity values overwhich the sensor is responsive and increases the sensitiv-ity at lower humidity levels Therefore an optimal poresize can be selected for the operating region of interest

283

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

103

104

105

106

0 20 40 60 80 100

Z (SO30)

Z(SO50)

Z(DO15)

| Z| (

W)

RH ()

Fig 11 Measured 5-kHz impedance of sensors over different humiditylevels

Figure 12 shows the response of the different sensorsto humidity at different measurement frequencies It canbe seen from the gure that as the frequency decreasesthe width of the sensitive region increases and the point atwhich the sensor becomes responsive shifts to lower rela-tive humidity values For a given pore size the humidity-sensitive region of operation can be selected by frequencytuning

The response time de ned as the time needed to reach90 of the nal signal for a given relative humidityand the recovery time de ned as the time taken forthe signal to come to within 10 of the initial valuewere determined by alternately exposing the sensors toa 2 and a 45 RH ambient with impedance measuredat 5 kHz A typical responserecovery graph of sensorSO50 is shown in Figure 13 All sensors were com-pletely reversible regaining their original impedance val-ues even after repeated exposure to high humidity levelsThe response time and recovery times corresponding todifferent pore diameters are shown in Figure 14 The min-imum responserecovery times were observed for SO50the sample with the largest pore size

The impedance spectra of the sensors were taken overa range of humidity levels The total impedance wasresolved into real (Z0) and imaginary (Z00) parts and Cole-Cole impedance plots were constructed Figure 15andashcshows respectively the Cole-Cole plot of sensors SO50SO30 and DO15 The equivalent circuit models shown inFigure 16 were used to t the experimental data The t-ting was accomplished with the complex nonlinear least-square- tting program101 The dots in the plots repre-sent experimental data and lines represent the equiva-lent circuit model t The equivalent circuit denoted (a)or (b) with reference to Figure 16 used for tting isdenoted near the corresponding curves in Figure 15 Theequivalent circuit models are different from the empiri-cal model suggested by Falk et al102 for porous aluminasensors fabricated by anodization of aluminum thin lmsIn the equivalent circuit models R1 and R2 represent twofrequency-independent resistors in parallel with disper-sive frequency-dependent capacitors Cn14mdash5 and Cn24mdash5respectively103 These non-Debye capacitances can be

103

104

105

106

107

SO50

5kHz10kHz100kHz1MHz

0 20 40 60 80 100

0 20 40 60 80 100

0 20 40 60 80 100

| Z| (

W)

| Z| (

W)

| Z| (

W)

RH ()(a)

(c)

(b)

103

104

105

106

107

SO30

5kHz10kHz100kHz1MHz

RH ()

102

103

104

105

106

DO15

5kHz10kHz100kHz1MHz

RH ()

Fig 12 Variation of sensor impedance with humidity and measure-ment frequency (a) SO50 (b) SO30 (c) DO15

68

7

72

74

76

78

0 200 400 600 800

Time (Sec)

45 RH

2 RH

45 RH 45 RH

2 RH

| Z| (

x105

W)

Fig 13 Response of sensor SO50 upon repeated cycling between rel-ative humidity environments of 2 and 45

284

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

10

15

20

25

30

35

40

45

50

20 25 30 35 40 45 50

Res

pons

eR

ecov

ery

Tim

e (S

ec)

Pore diameter (nm)

DO15

SO30

SO50

Fig 14 Dependence of response time and recovery time on pore size

expressed in a general form as Cn4mdash5 D Bn4imdash5nƒ1 whereBn is a constant for a given set of experimental conditionsand n takes a value between 0 and 1 according to the localmicroscopic environment through which charge transporttakes place751 1021 103 Bn behaves as an ideal frequency-independent capacitor as n approaches 1 and as an idealconductor when the value of n approaches 0101

By examination of Figure 15a SO50 it can be seenthat at low humidity levels the Cole-Cole plot describesan arc with a very large radius of curvature As the humid-ity increases the radius decreases and at an RH of about80 the complete semicircle comes inside the measur-able impedance range However at the same humiditylevel a spur appears in the low-frequency region that dis-torts the semicircle The spur which can be viewed aspart of a second semicircle dominates at higher humid-ity levels while the semicircle described by the high-frequency impedance values diminishes in size Thisbehavior is similar for all sensors For sensors SO50SO30 and DO15 respectively the RH at which thehigh-frequency semicircle comes completely within themeasurable impedance range is ordm75 65 and 35these values roughly match the in ection points seen inFigure 11

The fundamental mechanism that enables ceramic sen-sors to sense humidity is the physisorption of watermolecules on an initially chemisorbed layer of hydroxylions651 96 The chemisorbed hydroxyl ions enhance theelectrical conductivity of the sensor either by donatingelectrons to the conduction band of the base materialor through proton hopping between adjacent hydroxylgroups upon the application of an electric eld Theprocess of chemisorption occurs at very low humiditylevels and is unaffected by further changes in humid-ity However an increase in humidity makes the watermolecules physisorb to this hydroxyl layer The effec-tiveness of physisorption depends upon the cation chargecomplexes (from the material or the impurities) and thehydroxyl ions (water molecules present at the surface ofthe base material)96 The physisorption process is facili-tated by higher surface charge densities During formationof the rst physisorbed layer a water molecule becomesattached to two neighboring hydroxyl groups through

0

02

04

06

08

1

12

0 02 04 06 08 1 12

Zrsquo (MW)

35 (a) 55 (a)

65 (a)

75 (b)

81 (b)

93 (b) SO50

0

02

04

06

08

1

12

14

0 02 04 06 08 1 12 14

Zrsquo (W)

Z

(W)

Z

(MW

)Z

(M

W)

35 (a)

55 (a)

65 (b)

75 (b)

81 (b)

93 (b)

SO30

0

2 105

4 105

6 105

8 105

1 106

12 106

0 2 105 4 105 6 105 8 105 1 106 12 106

17 (a)27 (a)

35 (b)45 (b)55 (b)

93 (b)

DO15

(a)

Zrsquo (MW)

(b)

(c)

Fig 15 Cole-Cole plot of the sensors at different humidity levels(a) SO50 (b) SO30 (c) DO15 The points represent the experimentaldata and lines represent the t with the equivalent circuit model givenin Figure 9 The RH values at which the data were collected and theequivalent circuit used for tting each curve (given as (a) or (b) denot-ing the circuit of Figure 17a and b respectively) are shown adjacent toeach curve

hydrogen double bonds and a proton may be transferredfrom a hydroxyl group to the water molecule to form aH3OC ion

When the physisorption occurs within less than a monolayer that is when clusters of physisorbed molecules

285

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

(b)

(a)

Cn1 (w)

R1

Cn2 (w) Cn1 (w)

R1 R2

Fig 16 Equivalent circuit used for tting the experimental data

are present at the surface it is thought that H3OC diffu-sion within hydroxyl groups and proton transfer betweenadjacent water molecules clusters take place96 HoweverKhanna and Nahar104ndash106 refute this theory They arguethat the proton diffusion is less probable in a surfacethat contains hydroxyl ions and clusters of physisorbedmolecules and hence cannot account for the conductiv-ity of porous alumina at an RH less than 40 that isbefore the rst monolayer is formed Instead they sug-gested a phonon-assisted electron tunneling mechanism atlow humidity levels in which electron tunneling occursfrom one water molecule to another as enhanced by thesurface anions of the alumina sensors

At higher humidity levels the number of physisorbedlayers increases allowing each water molecule to besingly bonded to a hydroxyl group and proton hop-ping between adjacent water molecules in the continu-ous water layer takes place The conduction process isthe same as that of pure water and is called the Grot-thuss chain reaction96 The dominant charge carrier ina high-humidity environment is therefore HC ions (pro-tons) The concentration of HC increases with increas-ing humidity and HC move freely through the water-likelayer According to the theory of Khanna and Nahar104

in porous alumina the protons for conduction are donatednot only by H2O but also by the Al(OH)3 formed asa result of water adsorption and the impurity anionsincorporated into the lm from the electrolyte duringanodization107ndash109 The water-like network consists ofsingly bonded water molecules which possess a highdielectric constant as they form dipoles and reorient freelyunder an externally applied electric eld The impedanceof the nanoporous alumina sensors includes this capaci-tive contribution as well

From the analysis of Figure 15 the high-frequencysemicircle that diminishes in size with increasing RHarises because of the effect of a physisorbed water layerat the surface Therefore R1 in the equivalent circuitof Figure 16 represents the charge transfer resistancethrough the adsorbed water layer The effect of electronic

102

104

106

108

SO50SO30

DO15

R1

(W)

0 20 40 60 80 100

RH ()

Fig 17 Humidity dependence of resistance R1 of different sensorsobtained by tting the equivalent circuit model with experimental data

conduction through the alumina bulk is parallel to thatof the surface conduction affected by the adsorbed waterlayer This bulk conduction process is dominant only atextremely low humidity values when there are only afew adsorbed molecules Hence the incomplete semicir-cle or arc in the low-humidity region is partially deter-mined from the base alumina Bn1 represents the capac-itance contribution from the adsorbed water layer Thelarge spurs seen in the low-frequency region are a con-sequence of ion migration in the adsorbed layer towardelectrodes1011 103 This migration leads to accumulation ofions at the electrodes Since the relaxation time of ionmigration is larger than the charge transfer process itis seen only at low frequencies The accumulated ionsin the adsorbed water layer give rise to R2 and Bn2From the equivalent circuits given in Figure 16 it canbe seen that the presence of ion accumulation repre-sented by the combination of R2 and Cn24mdash5 increasesthe overall impedance of the sensor Hence the reductionin impedance caused at high humidity due to the chargetransfer process is countered by ion accumulation

To better understand the fundamental sensing mecha-nisms R1 and Bn1 obtained through the numerical ttingprogram for the different sensors are plotted in Figures 17and 18 respectively It can be seen from Figure 17 thatR1 which is the resistance to charge transport in a staticelectric eld by the material containing an adsorbed waterlayer in the absence of ion migration is linear with RH

-11

10-10

10-9

10-8

10-7

50V30VDO15

Bn1

0 20 40 60 80 100

RH ()

Fig 18 Humidity dependence of Bn1 of different sensors obtained by tting the equivalent circuit model (Fig 17) with experimental data

286

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

in a semilog plot For sensor DO15 R1 decreases to afew hundred ohms at high RH levels indicating that theobserved conductivity is not due merely to the protonsprovided by the adsorbed water layer Since deionizedwater and argon gas are used to create the required humid-ity environment it is reasonable to assume that the sur-faces of the alumina lms contain anions presumablyfrom the electrolyte used for anodization which controlsconduction through the adsorbed layer108

With reference to Figure 18 Bn1 of the larger pore sam-ples SO50 and SO30 increases signi cantly only at thehigher humidity levels This result is in contrast to thatof the smaller pore size sensors which shows that Bn1

increases at lower humidity levels The relative humid-ity values where a signi cant increase in Bn1 takes placegenerally match the point seen in Figure 11 where theimpedance drops signi cantly which in turn is a func-tion of pore size These same relative humidity valuescorrelate with the point at which the spur appears inthe impedance diagrams given in Figure 15 The resultslead to the following conclusions Since hydrogen-bondedwater molecules possess a high dielectric constant theincrease in Bn1 indicates the presence of these bondson the surface which means that a liquid-like networkmore than one monolayer of physisorbed molecules hasformed at that humidity level This liquid-like network isresponsible for the sharp reduction in impedance seen inFigure 11 and the ion migration that creates the spur inthe low-frequency portion of the Cole-Cole plots as seenin Figure 15 According to Eq (1) capillary condensationdoes not take place in the alumina test sensors

In light of the above observations we propose thefollowing model to explain the behavior of nanoporousalumina in response to humidity The anodization pro-cess results in a certain amount of electrolyte anionstrapped within the pores1041 1071 108 Even in a low-humidityregime the presence of these anions on the sensor sur-face provides a high charge density for easy physisorp-tion of water molecules The result is the formationof a liquid-like network within the pores The impu-rity anions also act as proton donors by a dissociativemechanism104 joining the protons donated by the watermolecules The smaller the pore size the lower the RHlevel at which the liquid-like network is formed Hencefor the small pore sensors there is variation in Bn1 aswell as impedance magnitude in low-humidity regimesthe exponential dependence of R1 on RH also suggeststhe same phenomenon

In the case of DO15 the rst semicircle is highly dis-torted by the low-frequency spur originating with ionmigration A relatively large number of mobile ions mightbe present in these sensors leading to an enhancedaccumulation of protons at the electrode-sample contactAs mentioned earlier the dipoles in the singly bondedadsorbed water layer are free to rotate in an electric eld

Hence in the region close to the electrode-alumina inter-face the adsorbed water layer exhibits a net dipolar ori-entation and therefore it does not display its normaldielectric constant1011 103 Depending on ion size up toa monolayer of charge can exist at the electrode101 thediffuse (Gouy) layer of counter-ions spreads away fromthis inner (Stern) layer The number of ions available toform the Stern and Gouy layers at a particular frequencyincreases with increasing number of physisorbed layersHence as the number of physisorbed layers increases theStern and Gouy layers start forming at higher frequencies

With reference to Figure 12 the relative humiditylevel at which the sharp decrease in impedance occursshifts higher with increasing frequency This result can beexplained with the equivalent circuit shown in Figure 16athat is the circuit in the absence of ion accumulation Athigh frequencies the reactive impedance Xc of Cn14mdash5equal to Xc

D 1=jmdashCn14mdash5 is low and hence impedanceis largely determined by R1 As the humidity increasesR1 decreases however this translates into a reduction inimpedance only at relatively low frequencies where thevalue of R1 is comparable to or less than the capacitivereactance of Cn1(mdash) As the measurement frequency isreduced the reactance of Cn1(mdash) decreases hence the effectof R1 becomes more signi cant at lower humidity levels

32 Hydrogen Sensing with Titania Nanotubes

We recently reported62 the fabrication of self-organizedtitania nanotube arrays with an anodization techniqueAlthough the as-prepared nanotubes are amorphous theycrystallize on annealing at elevated temperatures and arestructurally stable to at least 600 C This stability ofstructure which is an essential criterion of a gas-sensingmaterial prompted us to study the gas-sensing behavior ofthese nanotubes for technologically important gases suchas oxygen carbon monoxide ammonia carbon dioxideand hydrogen

Titania has earned much attention for its oxygen-sensing capabilities110ndash115 Furthermore with propermanipulation of the microstructure or crystalline phaseandor the addition of the proper impurities or surfacefunctionalization titania can also be used as a reducinggas sensor116ndash127 The interaction of a gas with a metaloxide semiconductor is primarily a surface phenomenonand nanoporous metal oxides1221 1231 1281 129 offer the advan-tage of providing large sensing surface areas

Room-temperature metal oxide hydrogen sensors aregenerally based on Schottky barrier modulation of deviceslike PdTiO2 or PtTiO2 by hydrogen130ndash132 Elevated-temperature hydrogen sensors examine electrical resis-tance with hydrogen concentration for example Birke-feld et al133 observed that the resistance of the anatasephase of titania varies in the presence of carbon monoxideand hydrogen at temperatures above 500 C but on dop-ing with 10 alumina it becomes selective for hydrogen

287

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Barrier layer

Titanium Nano tube array

Platinum electrodes

Fig 19 Schematic representation of the hydrogen sensor geometry

Munuera et al134 used rhenium doping in the anatasephase to perform room-temperature hydrogen monitor-ing Recently Shimizu et al122 reported that anodizednanoporous titania lms with a Pd Schottky barrier aresensitive to hydrogen at 250 C

Titania nanotubes were grown from titanium foil(995 pure from Alfa Aesar Ward Hill MA) with athickness of 025 mm The anodization was performed inan electrolyte medium of 05 hydro uoric acid in waterwith the use of a platinum foil cathode Well-de nednanotube arrays were grown with anodizing potentialsranging from 12 V to 20 V Nanotube length increaseswith anodization time reaching 400 nm in approximately20 min and then remains constant For the present studythe samples were anodized for 25 min The samples werethen annealed at 500 C in a pure oxygen ambient for 6 hwith a heating and cooling rate of 1 Cmin

The electrode geometry of the titania nanotube sen-sors is shown in Figure 19 The sensor consists of a basetitanium metal foil with a nanotube array grown on topAn insulating barrier layer separates the nanotubes fromthe conducting titanium foil Electrical connections weremade by two parallel 10 mm pound 2 mm platinum pads of100-Œm thickness The test chamber consists of a 13-literquartz tube with stainless steel end caps placed insidea furnace (Thermolyne USA model 21100 tubular fur-nace) Gas ow through the test chamber was controlledvia a computer-controlled mass ow controller The elec-trical resistance of the titania sensors was measured witha computer-controlled Keithly digital multimeter Prior todata collection the test chamber was initially evacuatedwith a mechanical pump whereupon nitrogen (99999pure) was passed while the sensor under test was heatedto the desired temperature The test gases were mixed inappropriate ratios with nitrogen to create the necessarytest gas ambient

The surface morphology of nanotube arrays preparedwith an anodization potential of 20 V and annealed at500 C for 6 h in a pure oxygen ambient is shown inFigure 20a and b It can be seen from these images thatthe nanotubes are uniform over the surface The nano-tubes are approximately 400 nm in length and have abarrier layer62 thickness of ordm50 nm For the nanotubes

100 nm

(a)

(b)

1 mm (c)

100 nm Fig 20 The surface morphology of the titania nanotubes after anneal-ing at 500 C (a) High- and (b) low-magni cation images of a 20-Vsample (c) A high-magni cation image of a 12-V sample

fabricated with 20-V anodization the average pore diam-eter as determined from FESEM images is 76 nm (thestandard deviation 15 nm) with a wall thickness of 27 nm(standard deviation 6 nm) The sample anodized at 12 Vshown in Figure 2c was found to have an average porediameter of 46 nm (standard deviation 8 nm) with a wallthickness of 17 nm (standard deviation 2 nm) The porosi-ties of the 20-V and 12-V samples were calculated as 45and 61 respectively

Glancing-angle X-ray diffraction patterns of a 20-Vsample annealed at 500 C for 6 h in oxygen ambientis shown in Figure 21 It can be seen that both anataseand rutile phases of titania are present in the sampleA detailed study135 of these structures by high-resolutiontransmission electron microscope (HRTEM) showed that

288

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

20 30 40 50 60 70 80

2 q (deg)

(101

)

(110

)

A

R

TR

T

T

R R A TRR T R R T TT

Fig 21 Glancing-angle X-ray diffraction pattern of a 20-V sample(glancing angle D 2) annealed at 500 C in oxygen ambient A R andT represent respectively the re ections from anatase crystallites rutilecrystallites and the titanium substrate

the anatase crystallites were concentrated on the walls ofthe nanotubes and rutile on the barrier layer

Nanotubes annealed in a pure oxygen ambient werefound to be stable (intact) to temperatures of approxi-mately 580 C Above this temperature protrusions wereseen to be coming out through the nanotubes an effectthat spread with increasing temperature These protru-sions which are due to oxidation of the titanium substratecollapse the nanotubes

Figure 22 shows the response of the 20-V (76 nm) sam-ple as a function of ambient temperature as it is switchedfrom a nitrogen environment to one containing 1000 ppmhydrogen and then back to nitrogen The plot was madeusing 4Rg=R05

ƒ1 versus time where R0 is the base resis-tance of the sensor that is the sensor resistance before thetest gas was introduced and Rg is the measured resistancein the presence of test gas The sensor shows increasinghydrogen sensitivity with temperature with a change by3 orders of magnitude in resistance at temperatures aboveordm300 C At all of the temperatures the original resistancerecovered without hysteresis

The sensitivity S is de ned by the formula

S DR0

ƒ Rgs

Rgs

10 0

10 1

10 2

10 3

0 10 0 5 10 2 1 10 3 15 103 2 10 3 25 103 3 10 3 35 103 4 10 3

Time (Sec)

(Rg

R0)

-1

180degC

240degC

265degC

290degC

345degC

375degC

H2+N

2

N2

Fig 22 Variation of resistance Rg normalized with respect to baselineresistance R0 of a 20-V sample with time on exposure to 1000 ppmhydrogen at different temperatures It may be noted that the inverse ofRg=R0 was used in the plot to represent data in the positive y direction

10-1

100

101

102

103

150 200 250 300 350 400 450

Temperature (degC)

Sens

itivi

ty (

R0-

Rgs

)R

gs

Fig 23 The sensitivity temperature dependence of a 20-V sample to1000 ppm hydrogen

where R0 is the resistance of the sensor before the gas ispassed and Rgs is the resistance after gas is passed andreaches the saturation value The temperature-dependentsensitivity of a 20-V sample to 1000 ppm hydrogen isshown in Figure 23 Sensitivity is seen to increase withtemperature to approximately 380 C where the increasein sensitivity with temperature begins to saturate

The response time de ned as the time needed for thesensor to reach 90 of the nal signal for a given concen-tration of gas is plotted against temperature in Figure 24(the time includes that required for the gas to equili-brate inside the measurement chamber estimated to beordm30 s) The response time decreases exponentially withtemperature

To check the behavior of the sensor on repeated hydro-gen exposure the hydrogen concentration was varied indiscrete steps of 100 ppm from 0 to 500 ppm while thetemperature was kept constant at 290 C the chamber was ushed with nitrogen after each exposure to hydrogenThe response of the 20-V prepared nanotube sensorskept at 290 C is shown in Figure 25 The behaviorof the sensor is consistent recovering its original resis-tance after repeated exposure to varying hydrogen gasconcentrations The sensitivities of the 76-nm- 53-nm-and 22-nm-diameter nanotube sensors to hydrogen con-centrations ranging from 100 ppm to 1 are shown inFigure 26 The sensitivity of the 76-nm (20-V) sensor tolow concentrations is shown in Figure 27 there is a linearincrease in sensitivity at low concentrations The lower

10 1

10 2

10 3

10 4

150 200 250 300 350 400 450

Temperature (degC)

Res

pons

e T

ime

(Sec

)

Fig 24 Variation in response time for a 20-V sensor as a function oftemperature The dots represent measured data

289

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

10 2

0 4 4 4 4 4

10 3

0 10 1 10 2 10 3 10 4 10 5 10

Time (Sec)

Res

ista

nce

(W)

H2 (ppm)

200

300

400

500

400

300

200

100

100

Fig 25 Resistance of a 20-V sample when exposed to different con-centrations of hydrogen at 290 C The nitrogen-hydrogen mixture waspassed for 1500 s the chamber was then ushed with nitrogen for 3000 sbefore the nitrogen-hydrogen mixture was passed again

test limit of 20 ppm is determined by the experimentalapparatus and not the sensor

Figure 28 shows the change in resistance of the 76-nm-46-nm- and 22-nm-diameter nanotube sensors with expo-sure to 1000 ppm hydrogen at 290 C Whereas smallerdiameter nanotubes had greater sensitivity to hydrogenthe samples made at lower anodizing voltages tended tobe more brittle and harder to handle mechanically withoutbreaking

The 20-V 76-nm-diameter sample was exposed to oxy-gen carbon monoxide ammonia and carbon dioxide at290 C to study the cross-sensitivity The sensor wasfound to have no detectable variation in resistance onexposure to carbon dioxide The sensitivities of the tita-nia nanotubes to the other gases are shown in Figure 29The sensitivities of the nanotubes to carbon monoxide andammonia are negligible compared with that of hydrogenThe resistance of the nanotubes increased in the presenceof oxygen and did not regain their original electrical con-ductivity even after several hours in a 290 C nitrogenenvironment

1

10

100

1000

104

105

0 02 04 06 08 1

Hydrogen Concentration ()

Pore diameter = 76 nm

Pore diameter = 53 nm

Pore diameter = 22nm

Sen

sitiv

ity (

R0-

Rg)

Rg

Fig 26 The sensitivity variation of a samples with pore diameters of22 nm 53 nm and 76 nm versus time and hydrogen concentrations

0 100 200 300 400 5001

10

100

Hydrogen concentration (ppm)

Sen

sitiv

ity

(R0-

Rgs

)R

gs

Fig 27 Sensitivity of a 20-V (76 nm) sample at 290 C for low hydro-gen concentrations

Since the sensor measurements were made in atmo-spheres without oxygen the increase in conductivitycannot be due to hydrogen removing oxygen from thelattice132ndash134 or the removal of chemisorbed oxygen135ndash137

Hydrogen molecules can dissociate at defects on thetitania surface which can subsequently diffuse into thetitania lattice acting as electron donors1281 1381 139 How-ever a dissociation-driven process would result in aslow response and recovery times with complete recov-ery unlikely since the sensor completely regains its orig-inal resistance with hydrogen cycling it appears that thisis not the dominant mechanism behind high hydrogensensitivity We believe chemisorption of the dissociatedhydrogen on the titania surface is the underlying sensingmechanism137 During chemisorption hydrogen acts as asurface state and a partial charge transfer from hydrogento the titania conduction band takes place This createsan electron accumulation layer on the nanotube surfaceenhancing its electrical conductance Upon removal of thehydrogen ambient electron transfer back to the hydro-gen molecule takes place which subsequently desorbsrestoring the original electrical resistance of the mate-rial Another factor that may play a role in the hydrogensensitivity (and selectivity) is the platinum electrodes Atelevated temperatures hydrogen dissociation can occur onplatinum surfaces These dissociated hydrogen atoms mayspill1361 140 onto the nanotube surface where they diffuse

10-1

101

103

105

0 05 1 15

Time (Hour)

Pore size - 76 nm

Pore size - 53 nm

Pore size - 22 nm

(Rg

R0)

-1

Fig 28 A comparison of the variation in resistance of samples havingpore diameters of 22 nm 53 nm and 76 nm versus time upon exposureto 1000 ppm of hydrogen at 290 C

290

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

10-2

10-1

100

0 100 2 102 4 102 6 102 8 102 1 103

OxygenAmmoniaCarbon monoxide

Time (Sec)

(Rg

R0)

-1

Fig 29 Variation of resistance of a 20-V sample when exposed to1000 ppm carbon monoxide 5 ammonia and 1 oxygen at 290 C

into the material affecting its electrical properties Fromthe present study it was not clear how signi cant a rolethe platinum electrodes play

Anatase the polymorph of titania has been reported tohave high sensitivity for reducing gases like hydrogen andcarbon monoxide1211 1241 133 Our nanotube samples containanatase phase mainly on the walls and rutile in the barrierlayer As the diffusing hydrogen atoms go to the intersti-tial sites1331 142 and as the ca ratio of anatase is almostfour times that of rutile it appears that the anatase latticeaccommodates hydrogen easily and hence makes a highercontribution to hydrogen sensitivity

The effect of chemisorption can be neglected in theoxygen-sensing experiments As the recovery requiresseveral hours it appears that the nanotubes contain oxy-gen vacancies or titanium interstitial defects in the pres-ence of nitrogen On exposure of the sensor to oxygenambient the lattice reoxidizes and hence the conductiv-ity of the sensor decreases On the removal of oxygenthe reduction of the lattice will not immediately occurhence the sensor requires several hours to regain its orig-inal conductivity

4 CONCLUSIONS

This paper summarizes different routes of fabrication ofmetal oxide nanoarchitectures useful for sensing applica-tions Structures made via anodization of a starting metaland breath gure formation from a sol-gel are discussedand presented as are TiO2 nanowires made by lling ananoporous alumina membrane with subsequent removalof the supporting template Two test cases illustratingthe utility of these nanoporous metal oxide structures forsensing are presented highly ordered nanoporous aluminais used for humidity sensing and TiO2 nanotubes are usedfor hydrogen sensing These study test cases reveal sev-eral points worth brie y noting

In the rst sensing test case uniform nanoporous alu-mina sensors exhibit pore size and frequency-dependent

humidity-sensing characteristics The mechanism respon-sible for the sensing action of the nanoporous alumina sen-sors was studied by representing the impedance spectra atdifferent humidity levels in Cole-Cole plots This makes itpossible to identify the competing impedance-determiningprocesses occurring in the sensors namely charge transferand mass transport through the physisorbed water layeras dependent upon the humidity level and input signalfrequency Charge transfer was found to enhance the sen-sitivity by causing a large variation in impedance withrespect to humidity Mass transport leads to accumulationof ions at the electrodes which tends to decrease the mea-surement sensitivity by increasing the impedance Hencea proper choice of operating frequency high enough toavoid ion accumulation is essential for optimum sensorperformance

It has been inferred from our results that anions becomeincorporated into the alumina lm during the anodiza-tion process supporting the hypothesis of Khanna andNahar104ndash106 where they act as proton donors and facili-tate the adsorption of water molecules Hence liquid-likecharge transfer networks are formed at relatively lowerhumidity levels for the smaller pore-size sensors Thisprocess rather than capillary condensation is primarilyresponsible for the observed humidity-sensing behavior ofthe sensors Our studies show that a distribution of poresizes as suggested by Shimizu et al94 is not necessary toobtain sensitive performance over a wide humidity rangerather such optimal performance can be obtained with uni-form nanodimensional pores containing adsorbed anionicimpurities

In the second sensing test case titania nanotubes pre-pared by anodization and annealed in an oxygen atmo-sphere at a temperature of 500 C were found to behighly sensitive to hydrogen The nanotube sensors con-tained both anatase and rutile phases of titania and showedappreciable sensitivity toward hydrogen at temperaturesas low as 180 C The sensitivity increased drasticallywith temperature showing a variation of 3 orders in mag-nitude of resistance to 1000 ppm of hydrogen at 400 CThe response time decreased with increasing temperatureat 290 C full switching of the sensor took approximately3 min Results were highly reproducible with no indica-tion of hysteresis Our results showed that these sensorsare at least capable of monitoring hydrogen levels from20 ppm to 4 At 290 C nanotubes with smaller porediameters showed higher sensitivity to hydrogen com-pared with their larger pore counterparts The sensorsshowed high selectivity to hydrogen compared with car-bon monoxide ammonia and carbon dioxide Althoughthe sensor was sensitive to high concentrations of oxy-gen the response time was high and the sensor did notcompletely regain the original condition We believe thehydrogen sensitivity of the nanotubes is due to hydrogendiffusion into the titania lattice where they act as electrondonors

291

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Acknowledgments Support of this work by the NationalScience Foundation through grant ECS-0210033 is grate-fully acknowledged

References and Notes

1 K Kajihara K Tanaka K Horao and N Soga Jpn J ApplPhys 35 6110 (1996)

2 B O Regan and M Gratzel A low cost Nature 353 737 (1991)3 A Fujishima and K Honda Nature 238 37 (1972)4 K Sato A Tsuzuki H Taoda Y Torii T Kato and Y Butsugan

J Mater Sci 29 5911 (1994)5 K Sato A Tsuzuki Y Torii H Taoda T Kato and Y Butsugan

J Mater Sci 30 837 (1995)6 N Ozer and C M Lampert Sol Energy Mater Sol Cells 54 147

(1998)7 A Manz and H Becker editors Microsystem Technology in

Chemistry and Life Sciences Springer-Verlag Berlin 19988 E Kim Y Xia and G M Whitesides Nature 376 581 (1995)9 M Templin A Franck A Du Chesne A Leist A Zhang

R Ulrich V Schadler and U Weisner Science 278 1795 (1997)10 D Zhao J Feng Q Huo N Melosh G H Frederickson

B Chmelka and G D Stucky Science 279 548 (1998)11 O D Velev T A Jede R F Lobo and A M Lenhoff Nature

389 447 (1997)12 B T Holland C F Blanford and A Stein Science 281 538

(1998)13 S Matsushita T Miwa and A Fujishima Chem Lett 9 925

(1997)14 A Imhof and D J Pine Adv Mater 10 697 (1998)15 T Tatsuma A Ikezawa Y Ohko T Miwa T Matsue and

A Fujishima Adv Mater 12 643 (2000)16 Z Zhong Y Yin B Gates and Y Xia Adv Mater 12 206 (2000)17 O D Velev and A M Lenhoff Curr Opin Colloids Interface

Sci 5 56 (2000)18 O D Velev and E W Kaler Adv Mater 12 531 (2000)19 P Jiang J Cizeron J F Bertone and V L Colvin J Am Chem

Soc 121 7957 (1999)20 M E Turner T J Trentler and V L Colvin Adv Mater 13 180

(2001)21 K M Kulinowski P Jiang H Vaswani and V L Colvin Adv

Mater 12 833 (2000)22 P Jiang J F Bertone and V L Colvin Science 291 453 (2001)23 R Seshadri and F C Meldrum Adv Mater 12 1149 (2000)24 F C Meldrum and R Seshadri Chem Commun 29 (2000)25 Q B Meng Z Z Gu O Sato and A Fujishima Appl Phys

Lett 77 4313 (2000)26 G Widawski B Rawiso and B Francois Nature 369 3897

(1994)27 B Francois O Pitois and J Francois Adv Mater 7 1041 (1995)28 O Pitois and B Francois Eur Phys J B 8 225 (1999)29 S A Jenekhe and X L Chen Science 283 372 (1999)30 M Shimomuara and T Sawadaishi Curr Opin Colloid Interface

Sci 6 11 (2001)31 O Karthaus N Maruyama X Cieren M Shimomura

H Hasegawa and T Hashimoto Langmuir 16 6071 (2000)32 T Nishikawa J Nishida R Ookura S Nishimura S Wada

T Karino and M Shimomura Mater Sci Eng C 10 141 (1999)33 K Kajihara K Nakanishi K Tanaka K Hirao and N Soga

J Am Ceram Soc 81 2670 (1998)34 C A Grimes R S Singh E C Dickey and O K Varghese

J Mater Res 16 1686 (2001)35 M Srinivasrao D Collings A Phillips and S Patel Science 292

79 (2001)36 P G de Gennes Rev Mod Phys 57 827 (1985)

37 D Beysens A Steyer P Guenoun D Fritter and C M KnoblerPhase Transitions 31 219 (1991)

38 B J Briscoe and K P Galvin J Phys D 23 422 (1990)39 B J Briscoe and K P Galvin J Phys D 23 1265 (1990)40 D Beysens and C M Knobler Phys Rev Lett 57 1433 (1986)41 F Family and P Meakin Phys Rev Lett 61 428 (1988)42 A V Limaye R D Narhe A M Dhote and S B Ogale Phys

Rev Lett 76 3762 (1996)43 P Dellrsquo Aversana and G P Neitzel Phys Today 51 38 (1998)44 P Dellrsquo Aversana J R Banavar and J Koplik Phys Fluids 8 15

(1986)45 P Dellrsquo Aversana V Tontodonato and L Carotenuto Phys Fluids

9 2475 (1997)46 M Q Li S Q Xu and E Kumacheva Langmuir 16 7275 (2000)47 P Freymuth Rev Sci Instrum 64 1 (1993)48 P Ball The Self-Made Tapestry Pattern Formation in Nature

Oxford University Press New York 199949 M F Schatz S J VanHook W D McCormick J B Swift and

H L Swinney Phys Fluids 11 2577 (1999)50 S R Coriell M R Cordes and W J Boettinger J Cryst Growth

49 13 (1980)51 T S Sullivan Y M Liu and R E Ecke Phys Rev E 54 486

(1996)52 J L Rogers M F Schatz J L Bougie and J B Swift 84 87

(2000)53 K Nitschke 52 R5772 (1995)54 E Bodenschatz J R De Bruyn G Ahlers and D S Cannell

Phys Rev Lett 67 3078 (1991)55 C W Meyer D S Cannell G Ahlers J B Swift and P C

Hohenberg Phys Rev Lett 61 947 (1988)56 R F Ismagilov D Rosmarin D H Gracias A D Stroock and

G M Whitesides Appl Phys Lett 79 439 (2001)57 E A Barringer and H K Bowen Langmuir 1 420 (1985)58 M Tschapek C Wasowski and R M T Sanchez J Electroanal

Chem 74 167 (1976)59 H Masuda and K Fukuda Science 268 1466 (1995)60 H Masuda F Hasegwa and S Ono J Electrochem Soc 144

L 127 (1997)61 O Jessensky F Muller and U Gosele Appl Phys Lett 72 1173

(1998)62 D Gong C A Grimes O K Varghese W Hu R S Singh

Z Chen and E C Dickey J Mater Res 16 3331 (2001)63 B M Kulwicki J Am Ceram Soc 74 697 (1991)64 H Arai and T Seiyama in Sensors A Comprehensive Survey

edited by W Gopel J Hesse and J N Zemel VCH Weinheim(1992) Vol 3 pp 981ndash1012

65 E Traversa Sens Actuators B 23 135 (1995)66 N Yamazoe and Y Shimizu Sens Actuators 10 379 (1986)67 A Bearzotti I Fratoddi L Palummo S Petrocco A Furlani

C Lo Sterzo and M V Russo Sens Actuators B 76 316 (2001)68 Y Sakai Y Sadaoka and M Matsuguchi Sens Actuators B 35

85 (1996)69 C A Grimes and D Kouzoudis Sens Actuators A 84 205

(2000)70 K G Ong C A Grimes C L Robbins and R S Singh Sens

Actuators A 93 33 (2001)71 T Seiyama N Yamazoe and H Arai Sens Actuators 4 85

(1983)72 L Ketron Ceram Bull 68 860 (1989)73 E Traversa G Gusmano and A Montenero Eur J Solid State

Inorg Chem 32 719 (1995)74 M K Jain M C Bhatnagar and G L Sharma Sens Actuators

B 55 180 (1999)75 O K Varghese and L K Malhotra J Appl Phys 87 7457 (2000)76 G Sberveglieri R Murri and N Pinto Sens Actuators B 23

177 (1995)

292

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

brought into contact with moist air the solvent vapor pres-sure and air velocity across the surface drive solvent evap-oration rapidly cooling the surface This in turn facili-tates the nucleation and growth of water droplets from theatmosphere36ndash41 The temperature difference between theliquid surface and air results in thermocapillary convec-tion within the liquid that acts to stabilize the condensingwater droplets on or at the solution surface43ndash48 Air owacross the surface coupled with surface convection cur-rents drives the ordering of water droplets into hexagonalarrays44ndash46 Once the surface is completely covered withwater droplets the temperature difference between the sur-face and the droplets diminishes and the droplets beingdenser than the solvent sink into the solution de ning theresidual structure Figure 3 shows the effect of sol pHshowing the pH-dependent variation in the resulting struc-ture achieved

Benard-Marangoni convection typically results from avertical temperature gradient due to solvent evaporationfrom a thin liquid lm46ndash54 Preferential solvent evapo-ration removes heat from the lm surface resulting inBenard-Marangoni convection50ndash56 across the uid layerwith the suspension welling up in the center of a convec-tion cell and then owing back down the cell boundary Asevaporation proceeds the colloidal suspension becomesmore concentrated changing the convection character-istics and in turn de ning the ultimate lm structureWell-de ned hexagonal pentagonal or square patternsare expected from an ideal time-independent Benard-Marangoni convection50 in a homogeneous liquid whereasthe cell patterns of our lms seen in Figure 2 are typicallyirregular This deviation of structure from the theoreticalideal may be the result of time-dependent convection owthrough changes in sol viscosity with preferential evapora-tion of the propanol solvent leading to coupled thermoso-lutal Benard-Marangoni convection51 in the nonhomoge-neous uid

22 Nanotube Nanopore and NanowireFabrication via Anodization of Al and Ti

Highly ordered nanoporous alumina lms (see Fig 4aandb)aremadethrougha two-stepanodizationprocess59ndash61

The aluminum substrate is rst anodized in an oxalic orsulfuric acid solution The anodization is stopped after afew microns of aluminum are consumed and the porousalumina lm is removed through etching The etchant amixture of chromic acid and phosphoric acid is highlyselective attacking alumina much faster than aluminumThe remaining aluminum is dimpled with the dimplesserving as a uniform seed layer upon which a highly uni-form porous layer can then be achieved through a sec-ond anodization step at the same voltage Alumina tem-plates are widely used as templates for the fabrication ofnanowires Figure 4c shows a TiO2 nanowire mat fab-ricated by lling pores of an alumina membrane via sol

Fig 3 Images taken of a dried 01 M TiO2 sol 1 ml of TTIP 005 mlof HNO3 (70 distilled) 01 ml of deionized water and 327 ml ofisopropanol to which varying amounts of a 12 nitric aciddeionizedwater solution were added to vary pH values (a) pH 017 (b) pH 029(c) pH 037

gel letting the lm dry and then subsequently removingthe alumina template by a sodium hydroxide etch

Anodization has also been used to fabricate TiO2 nano-tube arrays62 with pore size linearly proportional toanodization voltage (see Fig 5) The nanotube array inFigure 5 was fabricated by anodization of titanium at 20 Vin 05 wt HF solution for 20 min resulting in a well-aligned titanium oxide nanotube array with an approxi-mate average tube diameter of 60 nm and a tube lengthof 400 nm Diameters of fabricated tubes have ranged insize from 25 nm to 65 nm

280

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

Fig 4 (a) The surface of a two-step anodized alumina membrane (the akes seen in the image are the ldquodustrdquo associated with tissue paper)(b) An image of a cleaved sample showing the pore channels (c) TiO2

nanowires made by lling an alumina template via sol-gel then subse-quently removing the template by a sodium hydroxide etch resulting ina TiO2 nanowire mat

3 EXPERIMENTAL RESULTSAPPLICATION TO SENSING

31 The Water Vapor Sensing Performanceof Highly Ordered NanoporousAlumina Films

Humidity sensors have attracted considerable attentionover many years because of their great importancein applications ranging from monitoring food qual-ity to meteorological studies63ndash65 Ceramic humiditysensors651 71ndash82 are commercially available and offer majoradvantages with high resistance to chemical attack

Fig 5 TiO2 nanotubes made by anodization of titanium

thermal stability mechanical strength and quick responseHowever ceramic humidity sensors still suffer from insuf- cient sensitivity over wide humidity ranges as well aslack of reversibility and drift in base resistance with timebecause of water molecule chemisorption

The humidity-sensing properties of alumina discoveredalmost 50 years ago83ndash94 are based upon ionic conduc-tion the presence of an adsorbed layer of water at thesurface reduces the total sensor impedance because of theincrease in the ionic conductivity as well as capacitancedue to the high dielectric constant of water The humidity-sensing characteristics of a given sensing element dependupon the material used the method of preparation andthe resulting surface topology Porous alumina is usuallypreferred for sensing applications because of the largesurface area available for water adsorption

An illustration of water adsorption by a surfaceredrawn from Ref 95 is shown in Figure 6 Figure6a shows the adsorption of a water molecule on aclean alumina surface Physisorption of water initiallyoccurs on an active surface site and forms an adsorption

Fig 6 Schematic illustration of interaction of a water molecule withan alumina surface redrawn from Ref 95

281

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Fig 7 Schematic illustration of buildup of adsorbed water layersredrawn from Ref 96 building of a water layer upon an alumina surface

complex (Fig 6a and b) The negatively charged oxygenof the water molecule becomes attached to the positivelycharged cation site and one of the positively chargedhydrogen atoms becomes attached to the anionic site byelectrostatic attraction Eventually they are converted totwo chemisorbed hydroxyl (OH) groups (Fig 6c) Anincrease in humidity makes the water molecules physisorbto this chemisorbed hydroxyl layer The effectiveness ofphysisorption depends upon the cation charge complexes(from alumina or the surface impurities) and the hydroxylions on the surface of the alumina The physisorptionprocess is facilitated by higher surface charge densitiesAfter hydroxyl formation the next water molecule willbe physisorbed through hydrogen double bonds on thetwo neighbouring hydroxyl groups (Fig 6d) and a protonmay be transferred from a hydroxyl group to the watermolecule to form a H3OC ion At higher humidity levelseach water molecule will be singly bonded to a hydroxylgroup as illustrated in Figure 7 (redrawn from [96]) andproton hopping between adjacent water molecules in thecontinuous water layer takes place The conduction pro-cess in this case occurs by attachment of a proton to awater molecule thereby forming a hydronium ion thehydronium ion releases another proton to a second watermolecule that accepts this proton while releasing a thirdproton and so on through the liquid Hence the dominantcharge carrier in a high-moisture atmosphere is the HC

(proton) The concentration of HC increases with increas-ing humidity and it moves freely through the water-likelayer

In the case of porous alumina capillary condensationcan take place in pores with a radius up to rK at a partic-ular relative humidity and temperature which is given byKelvinrsquos relation

rKD 2ƒM

RT ln4Ps=P5

Here P is the water vapor pressure Ps is the water vaporpressure at saturation ƒ is the surface tension R is theuniversal gas constant T is the temperature in Kelvin

H O

H O

O

H

H +

H

H

O

H +H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H O

H O

O

H

H +

H

H

O

H +H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H O

H O

O

H

H +

H

H

O

H+H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H

OH

O

O

HH

O

H+H

H+

H

OH

OH

O

O

HH

O

H+

HH

+H

O

H

O

H

O

O

HH

O

H+ HH+ H

O

H

O

H

O

O

HH

O

H+ HH+ H

O

Aluminium

AluminaCapillary Condensation

Capillary Condensation Adsorption

Adsorbed Water Layer

Fig 8 Schematic representation of water vapor interaction withnanoporous alumina illustrating some but not all97 of the multiple fac-tors inherent in determining the response

and and M are respectively the density and molecularweight of water

According this relation the condensation occurs insidesmall pores at lower humidity levels as illustrated inFigure 8 which is an illustrative drawing presentingthe different water-surface interaction mechanisms (NoteFigure 8 is not meant to be an exhaustive considera-tion of all possible effects for example the drawingdoes not address effects associated with capillary surfacecurvature97)

This condensed water reduces the impedance of alu-mina considerably No condensation occurs inside poresof radius greater than rK at a particular humidity level (seethe largest pore shown in Fig 8) In such pores as in thecase of a plane surface it is the number of physisorbedlayers that decides the impedance

Since capillary condensation enhances the sensingcapabilities of a material pore size distribution has beenwidely considered to be an important parameter in deter-mining the sensitivity in a particular humidity range941 98

However the results of our study on uniform nanoporousalumina lms presented below show that an easily builtand highly reproducible wide-range humidity sensor canbe achieved with nanodimensional pores of a narrow sizedistribution made by anodization of aluminum59ndash611 99

Starting from adhesive-backed aluminum tape (99pure) purchased from Tesa Tape Inc100 and subsequentlyanodized sensors SO50 SO30 and DO15 were preparedunder the conditions given in Table I Per sample nota-tion the rst letter denotes a single S or double D stepanodization process the second letter denotes the use of

Table I Anodization conditions used for making nanoporous alumina lms

Anodizing Duration of No ofvoltage anodization anodization

Sample (v) (min) stages

SO50 50 120 1SO30 30 120 1DO15 15 120 2

The electrolyte used in all cases was 2 wt oxalic acid

282

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

oxalic acid (O) in the anodization bath the two-digit num-ber denotes the anodization voltage It was observed thatsingle-step anodization results in disordered pore struc-tures at low voltages Hence a double-step process wasused to make sample DO15 These samples were initiallyanodized under the conditions given in Table I The sam-ples were then immersed in an aqueous solution contain-ing 18 wt chromic acid and 4 wt phosphoric acid for12 h to remove the initial nanoporous layer while keep-ing the nonporous bottom layer of alumina intact Thesamples were rinsed in distilled water and again anodizedunder the same conditions used for the rst anodizationstep Estimates of pore size distribution were determinedfrom FE-SEM images and are shown in Figure 9andashcas a size distribution histograph Samples SO50 SO30and DO15 have respectively average pore diameters of

0

10

20

30

375 40 425 45 475 50

o

f po

res

Pore diameter (nm)

0

10

20

30

25 325 375 425 475 525

o

f po

res

0

5

10

15

20

25

175 20 225 25 275 30 325 375

DO15

o

f po

res

(a)

Pore diameter (nm)

(c)

Pore diameter (nm)

(b)

Fig 9 Pore size distribution of the tested sensors (a) SO50 (b) SO30(c) DO15

Fig 10 Digital image of an impedance-based interdigital nanoporousalumina humidity sensor

452 nm 384 nm and 136 nm with a correspondingstandard deviation of 34 nm 78 nm and 49 nm Allsamples have a predominantly ordered pore structure

Immediately after anodization all samples were thor-oughly washed in distilled water and then dried in a nitro-gen atmosphere at 100 C An interdigital capacitor pat-tern with dimensions of 12 cm pound 12 cm was formed onthe surface of the alumina lms by the evaporation of gold(thickness ordm 250 nm) through a mask A digital cameraimage of a typical sensor is shown in Figure 10 To createthe necessary ambient humidity argon passed through amass ow controller (MFC) was bubbled through a bot-tle containing deionized water and then mixed with dryargon coming from another MFC in appropriate ratiosbefore being passed to the test chamber A constant total ow was maintained throughout the experiment Cham-ber humidity was monitored with the humidity probeof a Keithley 6517A electrometer and all measurementswere carried out at room temperature (23 C) Sensorimpedance was measured over the frequency range 5 Hzto 13 MHz with a computer-controlled Hewlett Packardimpedance analyzer (4192A) tted with an impedance test xture (Agilent 16034E) A signal amplitude of 90 mVwas used for all measurements Electrical contact wasmade between the sensor under test and the impedanceanalyzer test xture with 42-gauge 5-cm-long jumpersattached to the interdigital electrodes with silver pasteThe contacts were subsequently annealed at 100 C for2 h to cure the silver paste

Figure 11 shows the measured 5-kHz sensor impedanceof the different sensors as a function of humidity Thehumidity-sensitive region of DO15 is sup145 to 95 RHSensors SO30 and SO50 become sensitive to humidity atapproximately 65 and 75 RH respectively A smallerpore size increases the range of humidity values overwhich the sensor is responsive and increases the sensitiv-ity at lower humidity levels Therefore an optimal poresize can be selected for the operating region of interest

283

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

103

104

105

106

0 20 40 60 80 100

Z (SO30)

Z(SO50)

Z(DO15)

| Z| (

W)

RH ()

Fig 11 Measured 5-kHz impedance of sensors over different humiditylevels

Figure 12 shows the response of the different sensorsto humidity at different measurement frequencies It canbe seen from the gure that as the frequency decreasesthe width of the sensitive region increases and the point atwhich the sensor becomes responsive shifts to lower rela-tive humidity values For a given pore size the humidity-sensitive region of operation can be selected by frequencytuning

The response time de ned as the time needed to reach90 of the nal signal for a given relative humidityand the recovery time de ned as the time taken forthe signal to come to within 10 of the initial valuewere determined by alternately exposing the sensors toa 2 and a 45 RH ambient with impedance measuredat 5 kHz A typical responserecovery graph of sensorSO50 is shown in Figure 13 All sensors were com-pletely reversible regaining their original impedance val-ues even after repeated exposure to high humidity levelsThe response time and recovery times corresponding todifferent pore diameters are shown in Figure 14 The min-imum responserecovery times were observed for SO50the sample with the largest pore size

The impedance spectra of the sensors were taken overa range of humidity levels The total impedance wasresolved into real (Z0) and imaginary (Z00) parts and Cole-Cole impedance plots were constructed Figure 15andashcshows respectively the Cole-Cole plot of sensors SO50SO30 and DO15 The equivalent circuit models shown inFigure 16 were used to t the experimental data The t-ting was accomplished with the complex nonlinear least-square- tting program101 The dots in the plots repre-sent experimental data and lines represent the equiva-lent circuit model t The equivalent circuit denoted (a)or (b) with reference to Figure 16 used for tting isdenoted near the corresponding curves in Figure 15 Theequivalent circuit models are different from the empiri-cal model suggested by Falk et al102 for porous aluminasensors fabricated by anodization of aluminum thin lmsIn the equivalent circuit models R1 and R2 represent twofrequency-independent resistors in parallel with disper-sive frequency-dependent capacitors Cn14mdash5 and Cn24mdash5respectively103 These non-Debye capacitances can be

103

104

105

106

107

SO50

5kHz10kHz100kHz1MHz

0 20 40 60 80 100

0 20 40 60 80 100

0 20 40 60 80 100

| Z| (

W)

| Z| (

W)

| Z| (

W)

RH ()(a)

(c)

(b)

103

104

105

106

107

SO30

5kHz10kHz100kHz1MHz

RH ()

102

103

104

105

106

DO15

5kHz10kHz100kHz1MHz

RH ()

Fig 12 Variation of sensor impedance with humidity and measure-ment frequency (a) SO50 (b) SO30 (c) DO15

68

7

72

74

76

78

0 200 400 600 800

Time (Sec)

45 RH

2 RH

45 RH 45 RH

2 RH

| Z| (

x105

W)

Fig 13 Response of sensor SO50 upon repeated cycling between rel-ative humidity environments of 2 and 45

284

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

10

15

20

25

30

35

40

45

50

20 25 30 35 40 45 50

Res

pons

eR

ecov

ery

Tim

e (S

ec)

Pore diameter (nm)

DO15

SO30

SO50

Fig 14 Dependence of response time and recovery time on pore size

expressed in a general form as Cn4mdash5 D Bn4imdash5nƒ1 whereBn is a constant for a given set of experimental conditionsand n takes a value between 0 and 1 according to the localmicroscopic environment through which charge transporttakes place751 1021 103 Bn behaves as an ideal frequency-independent capacitor as n approaches 1 and as an idealconductor when the value of n approaches 0101

By examination of Figure 15a SO50 it can be seenthat at low humidity levels the Cole-Cole plot describesan arc with a very large radius of curvature As the humid-ity increases the radius decreases and at an RH of about80 the complete semicircle comes inside the measur-able impedance range However at the same humiditylevel a spur appears in the low-frequency region that dis-torts the semicircle The spur which can be viewed aspart of a second semicircle dominates at higher humid-ity levels while the semicircle described by the high-frequency impedance values diminishes in size Thisbehavior is similar for all sensors For sensors SO50SO30 and DO15 respectively the RH at which thehigh-frequency semicircle comes completely within themeasurable impedance range is ordm75 65 and 35these values roughly match the in ection points seen inFigure 11

The fundamental mechanism that enables ceramic sen-sors to sense humidity is the physisorption of watermolecules on an initially chemisorbed layer of hydroxylions651 96 The chemisorbed hydroxyl ions enhance theelectrical conductivity of the sensor either by donatingelectrons to the conduction band of the base materialor through proton hopping between adjacent hydroxylgroups upon the application of an electric eld Theprocess of chemisorption occurs at very low humiditylevels and is unaffected by further changes in humid-ity However an increase in humidity makes the watermolecules physisorb to this hydroxyl layer The effec-tiveness of physisorption depends upon the cation chargecomplexes (from the material or the impurities) and thehydroxyl ions (water molecules present at the surface ofthe base material)96 The physisorption process is facili-tated by higher surface charge densities During formationof the rst physisorbed layer a water molecule becomesattached to two neighboring hydroxyl groups through

0

02

04

06

08

1

12

0 02 04 06 08 1 12

Zrsquo (MW)

35 (a) 55 (a)

65 (a)

75 (b)

81 (b)

93 (b) SO50

0

02

04

06

08

1

12

14

0 02 04 06 08 1 12 14

Zrsquo (W)

Z

(W)

Z

(MW

)Z

(M

W)

35 (a)

55 (a)

65 (b)

75 (b)

81 (b)

93 (b)

SO30

0

2 105

4 105

6 105

8 105

1 106

12 106

0 2 105 4 105 6 105 8 105 1 106 12 106

17 (a)27 (a)

35 (b)45 (b)55 (b)

93 (b)

DO15

(a)

Zrsquo (MW)

(b)

(c)

Fig 15 Cole-Cole plot of the sensors at different humidity levels(a) SO50 (b) SO30 (c) DO15 The points represent the experimentaldata and lines represent the t with the equivalent circuit model givenin Figure 9 The RH values at which the data were collected and theequivalent circuit used for tting each curve (given as (a) or (b) denot-ing the circuit of Figure 17a and b respectively) are shown adjacent toeach curve

hydrogen double bonds and a proton may be transferredfrom a hydroxyl group to the water molecule to form aH3OC ion

When the physisorption occurs within less than a monolayer that is when clusters of physisorbed molecules

285

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

(b)

(a)

Cn1 (w)

R1

Cn2 (w) Cn1 (w)

R1 R2

Fig 16 Equivalent circuit used for tting the experimental data

are present at the surface it is thought that H3OC diffu-sion within hydroxyl groups and proton transfer betweenadjacent water molecules clusters take place96 HoweverKhanna and Nahar104ndash106 refute this theory They arguethat the proton diffusion is less probable in a surfacethat contains hydroxyl ions and clusters of physisorbedmolecules and hence cannot account for the conductiv-ity of porous alumina at an RH less than 40 that isbefore the rst monolayer is formed Instead they sug-gested a phonon-assisted electron tunneling mechanism atlow humidity levels in which electron tunneling occursfrom one water molecule to another as enhanced by thesurface anions of the alumina sensors

At higher humidity levels the number of physisorbedlayers increases allowing each water molecule to besingly bonded to a hydroxyl group and proton hop-ping between adjacent water molecules in the continu-ous water layer takes place The conduction process isthe same as that of pure water and is called the Grot-thuss chain reaction96 The dominant charge carrier ina high-humidity environment is therefore HC ions (pro-tons) The concentration of HC increases with increas-ing humidity and HC move freely through the water-likelayer According to the theory of Khanna and Nahar104

in porous alumina the protons for conduction are donatednot only by H2O but also by the Al(OH)3 formed asa result of water adsorption and the impurity anionsincorporated into the lm from the electrolyte duringanodization107ndash109 The water-like network consists ofsingly bonded water molecules which possess a highdielectric constant as they form dipoles and reorient freelyunder an externally applied electric eld The impedanceof the nanoporous alumina sensors includes this capaci-tive contribution as well

From the analysis of Figure 15 the high-frequencysemicircle that diminishes in size with increasing RHarises because of the effect of a physisorbed water layerat the surface Therefore R1 in the equivalent circuitof Figure 16 represents the charge transfer resistancethrough the adsorbed water layer The effect of electronic

102

104

106

108

SO50SO30

DO15

R1

(W)

0 20 40 60 80 100

RH ()

Fig 17 Humidity dependence of resistance R1 of different sensorsobtained by tting the equivalent circuit model with experimental data

conduction through the alumina bulk is parallel to thatof the surface conduction affected by the adsorbed waterlayer This bulk conduction process is dominant only atextremely low humidity values when there are only afew adsorbed molecules Hence the incomplete semicir-cle or arc in the low-humidity region is partially deter-mined from the base alumina Bn1 represents the capac-itance contribution from the adsorbed water layer Thelarge spurs seen in the low-frequency region are a con-sequence of ion migration in the adsorbed layer towardelectrodes1011 103 This migration leads to accumulation ofions at the electrodes Since the relaxation time of ionmigration is larger than the charge transfer process itis seen only at low frequencies The accumulated ionsin the adsorbed water layer give rise to R2 and Bn2From the equivalent circuits given in Figure 16 it canbe seen that the presence of ion accumulation repre-sented by the combination of R2 and Cn24mdash5 increasesthe overall impedance of the sensor Hence the reductionin impedance caused at high humidity due to the chargetransfer process is countered by ion accumulation

To better understand the fundamental sensing mecha-nisms R1 and Bn1 obtained through the numerical ttingprogram for the different sensors are plotted in Figures 17and 18 respectively It can be seen from Figure 17 thatR1 which is the resistance to charge transport in a staticelectric eld by the material containing an adsorbed waterlayer in the absence of ion migration is linear with RH

-11

10-10

10-9

10-8

10-7

50V30VDO15

Bn1

0 20 40 60 80 100

RH ()

Fig 18 Humidity dependence of Bn1 of different sensors obtained by tting the equivalent circuit model (Fig 17) with experimental data

286

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

in a semilog plot For sensor DO15 R1 decreases to afew hundred ohms at high RH levels indicating that theobserved conductivity is not due merely to the protonsprovided by the adsorbed water layer Since deionizedwater and argon gas are used to create the required humid-ity environment it is reasonable to assume that the sur-faces of the alumina lms contain anions presumablyfrom the electrolyte used for anodization which controlsconduction through the adsorbed layer108

With reference to Figure 18 Bn1 of the larger pore sam-ples SO50 and SO30 increases signi cantly only at thehigher humidity levels This result is in contrast to thatof the smaller pore size sensors which shows that Bn1

increases at lower humidity levels The relative humid-ity values where a signi cant increase in Bn1 takes placegenerally match the point seen in Figure 11 where theimpedance drops signi cantly which in turn is a func-tion of pore size These same relative humidity valuescorrelate with the point at which the spur appears inthe impedance diagrams given in Figure 15 The resultslead to the following conclusions Since hydrogen-bondedwater molecules possess a high dielectric constant theincrease in Bn1 indicates the presence of these bondson the surface which means that a liquid-like networkmore than one monolayer of physisorbed molecules hasformed at that humidity level This liquid-like network isresponsible for the sharp reduction in impedance seen inFigure 11 and the ion migration that creates the spur inthe low-frequency portion of the Cole-Cole plots as seenin Figure 15 According to Eq (1) capillary condensationdoes not take place in the alumina test sensors

In light of the above observations we propose thefollowing model to explain the behavior of nanoporousalumina in response to humidity The anodization pro-cess results in a certain amount of electrolyte anionstrapped within the pores1041 1071 108 Even in a low-humidityregime the presence of these anions on the sensor sur-face provides a high charge density for easy physisorp-tion of water molecules The result is the formationof a liquid-like network within the pores The impu-rity anions also act as proton donors by a dissociativemechanism104 joining the protons donated by the watermolecules The smaller the pore size the lower the RHlevel at which the liquid-like network is formed Hencefor the small pore sensors there is variation in Bn1 aswell as impedance magnitude in low-humidity regimesthe exponential dependence of R1 on RH also suggeststhe same phenomenon

In the case of DO15 the rst semicircle is highly dis-torted by the low-frequency spur originating with ionmigration A relatively large number of mobile ions mightbe present in these sensors leading to an enhancedaccumulation of protons at the electrode-sample contactAs mentioned earlier the dipoles in the singly bondedadsorbed water layer are free to rotate in an electric eld

Hence in the region close to the electrode-alumina inter-face the adsorbed water layer exhibits a net dipolar ori-entation and therefore it does not display its normaldielectric constant1011 103 Depending on ion size up toa monolayer of charge can exist at the electrode101 thediffuse (Gouy) layer of counter-ions spreads away fromthis inner (Stern) layer The number of ions available toform the Stern and Gouy layers at a particular frequencyincreases with increasing number of physisorbed layersHence as the number of physisorbed layers increases theStern and Gouy layers start forming at higher frequencies

With reference to Figure 12 the relative humiditylevel at which the sharp decrease in impedance occursshifts higher with increasing frequency This result can beexplained with the equivalent circuit shown in Figure 16athat is the circuit in the absence of ion accumulation Athigh frequencies the reactive impedance Xc of Cn14mdash5equal to Xc

D 1=jmdashCn14mdash5 is low and hence impedanceis largely determined by R1 As the humidity increasesR1 decreases however this translates into a reduction inimpedance only at relatively low frequencies where thevalue of R1 is comparable to or less than the capacitivereactance of Cn1(mdash) As the measurement frequency isreduced the reactance of Cn1(mdash) decreases hence the effectof R1 becomes more signi cant at lower humidity levels

32 Hydrogen Sensing with Titania Nanotubes

We recently reported62 the fabrication of self-organizedtitania nanotube arrays with an anodization techniqueAlthough the as-prepared nanotubes are amorphous theycrystallize on annealing at elevated temperatures and arestructurally stable to at least 600 C This stability ofstructure which is an essential criterion of a gas-sensingmaterial prompted us to study the gas-sensing behavior ofthese nanotubes for technologically important gases suchas oxygen carbon monoxide ammonia carbon dioxideand hydrogen

Titania has earned much attention for its oxygen-sensing capabilities110ndash115 Furthermore with propermanipulation of the microstructure or crystalline phaseandor the addition of the proper impurities or surfacefunctionalization titania can also be used as a reducinggas sensor116ndash127 The interaction of a gas with a metaloxide semiconductor is primarily a surface phenomenonand nanoporous metal oxides1221 1231 1281 129 offer the advan-tage of providing large sensing surface areas

Room-temperature metal oxide hydrogen sensors aregenerally based on Schottky barrier modulation of deviceslike PdTiO2 or PtTiO2 by hydrogen130ndash132 Elevated-temperature hydrogen sensors examine electrical resis-tance with hydrogen concentration for example Birke-feld et al133 observed that the resistance of the anatasephase of titania varies in the presence of carbon monoxideand hydrogen at temperatures above 500 C but on dop-ing with 10 alumina it becomes selective for hydrogen

287

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Barrier layer

Titanium Nano tube array

Platinum electrodes

Fig 19 Schematic representation of the hydrogen sensor geometry

Munuera et al134 used rhenium doping in the anatasephase to perform room-temperature hydrogen monitor-ing Recently Shimizu et al122 reported that anodizednanoporous titania lms with a Pd Schottky barrier aresensitive to hydrogen at 250 C

Titania nanotubes were grown from titanium foil(995 pure from Alfa Aesar Ward Hill MA) with athickness of 025 mm The anodization was performed inan electrolyte medium of 05 hydro uoric acid in waterwith the use of a platinum foil cathode Well-de nednanotube arrays were grown with anodizing potentialsranging from 12 V to 20 V Nanotube length increaseswith anodization time reaching 400 nm in approximately20 min and then remains constant For the present studythe samples were anodized for 25 min The samples werethen annealed at 500 C in a pure oxygen ambient for 6 hwith a heating and cooling rate of 1 Cmin

The electrode geometry of the titania nanotube sen-sors is shown in Figure 19 The sensor consists of a basetitanium metal foil with a nanotube array grown on topAn insulating barrier layer separates the nanotubes fromthe conducting titanium foil Electrical connections weremade by two parallel 10 mm pound 2 mm platinum pads of100-Œm thickness The test chamber consists of a 13-literquartz tube with stainless steel end caps placed insidea furnace (Thermolyne USA model 21100 tubular fur-nace) Gas ow through the test chamber was controlledvia a computer-controlled mass ow controller The elec-trical resistance of the titania sensors was measured witha computer-controlled Keithly digital multimeter Prior todata collection the test chamber was initially evacuatedwith a mechanical pump whereupon nitrogen (99999pure) was passed while the sensor under test was heatedto the desired temperature The test gases were mixed inappropriate ratios with nitrogen to create the necessarytest gas ambient

The surface morphology of nanotube arrays preparedwith an anodization potential of 20 V and annealed at500 C for 6 h in a pure oxygen ambient is shown inFigure 20a and b It can be seen from these images thatthe nanotubes are uniform over the surface The nano-tubes are approximately 400 nm in length and have abarrier layer62 thickness of ordm50 nm For the nanotubes

100 nm

(a)

(b)

1 mm (c)

100 nm Fig 20 The surface morphology of the titania nanotubes after anneal-ing at 500 C (a) High- and (b) low-magni cation images of a 20-Vsample (c) A high-magni cation image of a 12-V sample

fabricated with 20-V anodization the average pore diam-eter as determined from FESEM images is 76 nm (thestandard deviation 15 nm) with a wall thickness of 27 nm(standard deviation 6 nm) The sample anodized at 12 Vshown in Figure 2c was found to have an average porediameter of 46 nm (standard deviation 8 nm) with a wallthickness of 17 nm (standard deviation 2 nm) The porosi-ties of the 20-V and 12-V samples were calculated as 45and 61 respectively

Glancing-angle X-ray diffraction patterns of a 20-Vsample annealed at 500 C for 6 h in oxygen ambientis shown in Figure 21 It can be seen that both anataseand rutile phases of titania are present in the sampleA detailed study135 of these structures by high-resolutiontransmission electron microscope (HRTEM) showed that

288

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

20 30 40 50 60 70 80

2 q (deg)

(101

)

(110

)

A

R

TR

T

T

R R A TRR T R R T TT

Fig 21 Glancing-angle X-ray diffraction pattern of a 20-V sample(glancing angle D 2) annealed at 500 C in oxygen ambient A R andT represent respectively the re ections from anatase crystallites rutilecrystallites and the titanium substrate

the anatase crystallites were concentrated on the walls ofthe nanotubes and rutile on the barrier layer

Nanotubes annealed in a pure oxygen ambient werefound to be stable (intact) to temperatures of approxi-mately 580 C Above this temperature protrusions wereseen to be coming out through the nanotubes an effectthat spread with increasing temperature These protru-sions which are due to oxidation of the titanium substratecollapse the nanotubes

Figure 22 shows the response of the 20-V (76 nm) sam-ple as a function of ambient temperature as it is switchedfrom a nitrogen environment to one containing 1000 ppmhydrogen and then back to nitrogen The plot was madeusing 4Rg=R05

ƒ1 versus time where R0 is the base resis-tance of the sensor that is the sensor resistance before thetest gas was introduced and Rg is the measured resistancein the presence of test gas The sensor shows increasinghydrogen sensitivity with temperature with a change by3 orders of magnitude in resistance at temperatures aboveordm300 C At all of the temperatures the original resistancerecovered without hysteresis

The sensitivity S is de ned by the formula

S DR0

ƒ Rgs

Rgs

10 0

10 1

10 2

10 3

0 10 0 5 10 2 1 10 3 15 103 2 10 3 25 103 3 10 3 35 103 4 10 3

Time (Sec)

(Rg

R0)

-1

180degC

240degC

265degC

290degC

345degC

375degC

H2+N

2

N2

Fig 22 Variation of resistance Rg normalized with respect to baselineresistance R0 of a 20-V sample with time on exposure to 1000 ppmhydrogen at different temperatures It may be noted that the inverse ofRg=R0 was used in the plot to represent data in the positive y direction

10-1

100

101

102

103

150 200 250 300 350 400 450

Temperature (degC)

Sens

itivi

ty (

R0-

Rgs

)R

gs

Fig 23 The sensitivity temperature dependence of a 20-V sample to1000 ppm hydrogen

where R0 is the resistance of the sensor before the gas ispassed and Rgs is the resistance after gas is passed andreaches the saturation value The temperature-dependentsensitivity of a 20-V sample to 1000 ppm hydrogen isshown in Figure 23 Sensitivity is seen to increase withtemperature to approximately 380 C where the increasein sensitivity with temperature begins to saturate

The response time de ned as the time needed for thesensor to reach 90 of the nal signal for a given concen-tration of gas is plotted against temperature in Figure 24(the time includes that required for the gas to equili-brate inside the measurement chamber estimated to beordm30 s) The response time decreases exponentially withtemperature

To check the behavior of the sensor on repeated hydro-gen exposure the hydrogen concentration was varied indiscrete steps of 100 ppm from 0 to 500 ppm while thetemperature was kept constant at 290 C the chamber was ushed with nitrogen after each exposure to hydrogenThe response of the 20-V prepared nanotube sensorskept at 290 C is shown in Figure 25 The behaviorof the sensor is consistent recovering its original resis-tance after repeated exposure to varying hydrogen gasconcentrations The sensitivities of the 76-nm- 53-nm-and 22-nm-diameter nanotube sensors to hydrogen con-centrations ranging from 100 ppm to 1 are shown inFigure 26 The sensitivity of the 76-nm (20-V) sensor tolow concentrations is shown in Figure 27 there is a linearincrease in sensitivity at low concentrations The lower

10 1

10 2

10 3

10 4

150 200 250 300 350 400 450

Temperature (degC)

Res

pons

e T

ime

(Sec

)

Fig 24 Variation in response time for a 20-V sensor as a function oftemperature The dots represent measured data

289

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

10 2

0 4 4 4 4 4

10 3

0 10 1 10 2 10 3 10 4 10 5 10

Time (Sec)

Res

ista

nce

(W)

H2 (ppm)

200

300

400

500

400

300

200

100

100

Fig 25 Resistance of a 20-V sample when exposed to different con-centrations of hydrogen at 290 C The nitrogen-hydrogen mixture waspassed for 1500 s the chamber was then ushed with nitrogen for 3000 sbefore the nitrogen-hydrogen mixture was passed again

test limit of 20 ppm is determined by the experimentalapparatus and not the sensor

Figure 28 shows the change in resistance of the 76-nm-46-nm- and 22-nm-diameter nanotube sensors with expo-sure to 1000 ppm hydrogen at 290 C Whereas smallerdiameter nanotubes had greater sensitivity to hydrogenthe samples made at lower anodizing voltages tended tobe more brittle and harder to handle mechanically withoutbreaking

The 20-V 76-nm-diameter sample was exposed to oxy-gen carbon monoxide ammonia and carbon dioxide at290 C to study the cross-sensitivity The sensor wasfound to have no detectable variation in resistance onexposure to carbon dioxide The sensitivities of the tita-nia nanotubes to the other gases are shown in Figure 29The sensitivities of the nanotubes to carbon monoxide andammonia are negligible compared with that of hydrogenThe resistance of the nanotubes increased in the presenceof oxygen and did not regain their original electrical con-ductivity even after several hours in a 290 C nitrogenenvironment

1

10

100

1000

104

105

0 02 04 06 08 1

Hydrogen Concentration ()

Pore diameter = 76 nm

Pore diameter = 53 nm

Pore diameter = 22nm

Sen

sitiv

ity (

R0-

Rg)

Rg

Fig 26 The sensitivity variation of a samples with pore diameters of22 nm 53 nm and 76 nm versus time and hydrogen concentrations

0 100 200 300 400 5001

10

100

Hydrogen concentration (ppm)

Sen

sitiv

ity

(R0-

Rgs

)R

gs

Fig 27 Sensitivity of a 20-V (76 nm) sample at 290 C for low hydro-gen concentrations

Since the sensor measurements were made in atmo-spheres without oxygen the increase in conductivitycannot be due to hydrogen removing oxygen from thelattice132ndash134 or the removal of chemisorbed oxygen135ndash137

Hydrogen molecules can dissociate at defects on thetitania surface which can subsequently diffuse into thetitania lattice acting as electron donors1281 1381 139 How-ever a dissociation-driven process would result in aslow response and recovery times with complete recov-ery unlikely since the sensor completely regains its orig-inal resistance with hydrogen cycling it appears that thisis not the dominant mechanism behind high hydrogensensitivity We believe chemisorption of the dissociatedhydrogen on the titania surface is the underlying sensingmechanism137 During chemisorption hydrogen acts as asurface state and a partial charge transfer from hydrogento the titania conduction band takes place This createsan electron accumulation layer on the nanotube surfaceenhancing its electrical conductance Upon removal of thehydrogen ambient electron transfer back to the hydro-gen molecule takes place which subsequently desorbsrestoring the original electrical resistance of the mate-rial Another factor that may play a role in the hydrogensensitivity (and selectivity) is the platinum electrodes Atelevated temperatures hydrogen dissociation can occur onplatinum surfaces These dissociated hydrogen atoms mayspill1361 140 onto the nanotube surface where they diffuse

10-1

101

103

105

0 05 1 15

Time (Hour)

Pore size - 76 nm

Pore size - 53 nm

Pore size - 22 nm

(Rg

R0)

-1

Fig 28 A comparison of the variation in resistance of samples havingpore diameters of 22 nm 53 nm and 76 nm versus time upon exposureto 1000 ppm of hydrogen at 290 C

290

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

10-2

10-1

100

0 100 2 102 4 102 6 102 8 102 1 103

OxygenAmmoniaCarbon monoxide

Time (Sec)

(Rg

R0)

-1

Fig 29 Variation of resistance of a 20-V sample when exposed to1000 ppm carbon monoxide 5 ammonia and 1 oxygen at 290 C

into the material affecting its electrical properties Fromthe present study it was not clear how signi cant a rolethe platinum electrodes play

Anatase the polymorph of titania has been reported tohave high sensitivity for reducing gases like hydrogen andcarbon monoxide1211 1241 133 Our nanotube samples containanatase phase mainly on the walls and rutile in the barrierlayer As the diffusing hydrogen atoms go to the intersti-tial sites1331 142 and as the ca ratio of anatase is almostfour times that of rutile it appears that the anatase latticeaccommodates hydrogen easily and hence makes a highercontribution to hydrogen sensitivity

The effect of chemisorption can be neglected in theoxygen-sensing experiments As the recovery requiresseveral hours it appears that the nanotubes contain oxy-gen vacancies or titanium interstitial defects in the pres-ence of nitrogen On exposure of the sensor to oxygenambient the lattice reoxidizes and hence the conductiv-ity of the sensor decreases On the removal of oxygenthe reduction of the lattice will not immediately occurhence the sensor requires several hours to regain its orig-inal conductivity

4 CONCLUSIONS

This paper summarizes different routes of fabrication ofmetal oxide nanoarchitectures useful for sensing applica-tions Structures made via anodization of a starting metaland breath gure formation from a sol-gel are discussedand presented as are TiO2 nanowires made by lling ananoporous alumina membrane with subsequent removalof the supporting template Two test cases illustratingthe utility of these nanoporous metal oxide structures forsensing are presented highly ordered nanoporous aluminais used for humidity sensing and TiO2 nanotubes are usedfor hydrogen sensing These study test cases reveal sev-eral points worth brie y noting

In the rst sensing test case uniform nanoporous alu-mina sensors exhibit pore size and frequency-dependent

humidity-sensing characteristics The mechanism respon-sible for the sensing action of the nanoporous alumina sen-sors was studied by representing the impedance spectra atdifferent humidity levels in Cole-Cole plots This makes itpossible to identify the competing impedance-determiningprocesses occurring in the sensors namely charge transferand mass transport through the physisorbed water layeras dependent upon the humidity level and input signalfrequency Charge transfer was found to enhance the sen-sitivity by causing a large variation in impedance withrespect to humidity Mass transport leads to accumulationof ions at the electrodes which tends to decrease the mea-surement sensitivity by increasing the impedance Hencea proper choice of operating frequency high enough toavoid ion accumulation is essential for optimum sensorperformance

It has been inferred from our results that anions becomeincorporated into the alumina lm during the anodiza-tion process supporting the hypothesis of Khanna andNahar104ndash106 where they act as proton donors and facili-tate the adsorption of water molecules Hence liquid-likecharge transfer networks are formed at relatively lowerhumidity levels for the smaller pore-size sensors Thisprocess rather than capillary condensation is primarilyresponsible for the observed humidity-sensing behavior ofthe sensors Our studies show that a distribution of poresizes as suggested by Shimizu et al94 is not necessary toobtain sensitive performance over a wide humidity rangerather such optimal performance can be obtained with uni-form nanodimensional pores containing adsorbed anionicimpurities

In the second sensing test case titania nanotubes pre-pared by anodization and annealed in an oxygen atmo-sphere at a temperature of 500 C were found to behighly sensitive to hydrogen The nanotube sensors con-tained both anatase and rutile phases of titania and showedappreciable sensitivity toward hydrogen at temperaturesas low as 180 C The sensitivity increased drasticallywith temperature showing a variation of 3 orders in mag-nitude of resistance to 1000 ppm of hydrogen at 400 CThe response time decreased with increasing temperatureat 290 C full switching of the sensor took approximately3 min Results were highly reproducible with no indica-tion of hysteresis Our results showed that these sensorsare at least capable of monitoring hydrogen levels from20 ppm to 4 At 290 C nanotubes with smaller porediameters showed higher sensitivity to hydrogen com-pared with their larger pore counterparts The sensorsshowed high selectivity to hydrogen compared with car-bon monoxide ammonia and carbon dioxide Althoughthe sensor was sensitive to high concentrations of oxy-gen the response time was high and the sensor did notcompletely regain the original condition We believe thehydrogen sensitivity of the nanotubes is due to hydrogendiffusion into the titania lattice where they act as electrondonors

291

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Acknowledgments Support of this work by the NationalScience Foundation through grant ECS-0210033 is grate-fully acknowledged

References and Notes

1 K Kajihara K Tanaka K Horao and N Soga Jpn J ApplPhys 35 6110 (1996)

2 B O Regan and M Gratzel A low cost Nature 353 737 (1991)3 A Fujishima and K Honda Nature 238 37 (1972)4 K Sato A Tsuzuki H Taoda Y Torii T Kato and Y Butsugan

J Mater Sci 29 5911 (1994)5 K Sato A Tsuzuki Y Torii H Taoda T Kato and Y Butsugan

J Mater Sci 30 837 (1995)6 N Ozer and C M Lampert Sol Energy Mater Sol Cells 54 147

(1998)7 A Manz and H Becker editors Microsystem Technology in

Chemistry and Life Sciences Springer-Verlag Berlin 19988 E Kim Y Xia and G M Whitesides Nature 376 581 (1995)9 M Templin A Franck A Du Chesne A Leist A Zhang

R Ulrich V Schadler and U Weisner Science 278 1795 (1997)10 D Zhao J Feng Q Huo N Melosh G H Frederickson

B Chmelka and G D Stucky Science 279 548 (1998)11 O D Velev T A Jede R F Lobo and A M Lenhoff Nature

389 447 (1997)12 B T Holland C F Blanford and A Stein Science 281 538

(1998)13 S Matsushita T Miwa and A Fujishima Chem Lett 9 925

(1997)14 A Imhof and D J Pine Adv Mater 10 697 (1998)15 T Tatsuma A Ikezawa Y Ohko T Miwa T Matsue and

A Fujishima Adv Mater 12 643 (2000)16 Z Zhong Y Yin B Gates and Y Xia Adv Mater 12 206 (2000)17 O D Velev and A M Lenhoff Curr Opin Colloids Interface

Sci 5 56 (2000)18 O D Velev and E W Kaler Adv Mater 12 531 (2000)19 P Jiang J Cizeron J F Bertone and V L Colvin J Am Chem

Soc 121 7957 (1999)20 M E Turner T J Trentler and V L Colvin Adv Mater 13 180

(2001)21 K M Kulinowski P Jiang H Vaswani and V L Colvin Adv

Mater 12 833 (2000)22 P Jiang J F Bertone and V L Colvin Science 291 453 (2001)23 R Seshadri and F C Meldrum Adv Mater 12 1149 (2000)24 F C Meldrum and R Seshadri Chem Commun 29 (2000)25 Q B Meng Z Z Gu O Sato and A Fujishima Appl Phys

Lett 77 4313 (2000)26 G Widawski B Rawiso and B Francois Nature 369 3897

(1994)27 B Francois O Pitois and J Francois Adv Mater 7 1041 (1995)28 O Pitois and B Francois Eur Phys J B 8 225 (1999)29 S A Jenekhe and X L Chen Science 283 372 (1999)30 M Shimomuara and T Sawadaishi Curr Opin Colloid Interface

Sci 6 11 (2001)31 O Karthaus N Maruyama X Cieren M Shimomura

H Hasegawa and T Hashimoto Langmuir 16 6071 (2000)32 T Nishikawa J Nishida R Ookura S Nishimura S Wada

T Karino and M Shimomura Mater Sci Eng C 10 141 (1999)33 K Kajihara K Nakanishi K Tanaka K Hirao and N Soga

J Am Ceram Soc 81 2670 (1998)34 C A Grimes R S Singh E C Dickey and O K Varghese

J Mater Res 16 1686 (2001)35 M Srinivasrao D Collings A Phillips and S Patel Science 292

79 (2001)36 P G de Gennes Rev Mod Phys 57 827 (1985)

37 D Beysens A Steyer P Guenoun D Fritter and C M KnoblerPhase Transitions 31 219 (1991)

38 B J Briscoe and K P Galvin J Phys D 23 422 (1990)39 B J Briscoe and K P Galvin J Phys D 23 1265 (1990)40 D Beysens and C M Knobler Phys Rev Lett 57 1433 (1986)41 F Family and P Meakin Phys Rev Lett 61 428 (1988)42 A V Limaye R D Narhe A M Dhote and S B Ogale Phys

Rev Lett 76 3762 (1996)43 P Dellrsquo Aversana and G P Neitzel Phys Today 51 38 (1998)44 P Dellrsquo Aversana J R Banavar and J Koplik Phys Fluids 8 15

(1986)45 P Dellrsquo Aversana V Tontodonato and L Carotenuto Phys Fluids

9 2475 (1997)46 M Q Li S Q Xu and E Kumacheva Langmuir 16 7275 (2000)47 P Freymuth Rev Sci Instrum 64 1 (1993)48 P Ball The Self-Made Tapestry Pattern Formation in Nature

Oxford University Press New York 199949 M F Schatz S J VanHook W D McCormick J B Swift and

H L Swinney Phys Fluids 11 2577 (1999)50 S R Coriell M R Cordes and W J Boettinger J Cryst Growth

49 13 (1980)51 T S Sullivan Y M Liu and R E Ecke Phys Rev E 54 486

(1996)52 J L Rogers M F Schatz J L Bougie and J B Swift 84 87

(2000)53 K Nitschke 52 R5772 (1995)54 E Bodenschatz J R De Bruyn G Ahlers and D S Cannell

Phys Rev Lett 67 3078 (1991)55 C W Meyer D S Cannell G Ahlers J B Swift and P C

Hohenberg Phys Rev Lett 61 947 (1988)56 R F Ismagilov D Rosmarin D H Gracias A D Stroock and

G M Whitesides Appl Phys Lett 79 439 (2001)57 E A Barringer and H K Bowen Langmuir 1 420 (1985)58 M Tschapek C Wasowski and R M T Sanchez J Electroanal

Chem 74 167 (1976)59 H Masuda and K Fukuda Science 268 1466 (1995)60 H Masuda F Hasegwa and S Ono J Electrochem Soc 144

L 127 (1997)61 O Jessensky F Muller and U Gosele Appl Phys Lett 72 1173

(1998)62 D Gong C A Grimes O K Varghese W Hu R S Singh

Z Chen and E C Dickey J Mater Res 16 3331 (2001)63 B M Kulwicki J Am Ceram Soc 74 697 (1991)64 H Arai and T Seiyama in Sensors A Comprehensive Survey

edited by W Gopel J Hesse and J N Zemel VCH Weinheim(1992) Vol 3 pp 981ndash1012

65 E Traversa Sens Actuators B 23 135 (1995)66 N Yamazoe and Y Shimizu Sens Actuators 10 379 (1986)67 A Bearzotti I Fratoddi L Palummo S Petrocco A Furlani

C Lo Sterzo and M V Russo Sens Actuators B 76 316 (2001)68 Y Sakai Y Sadaoka and M Matsuguchi Sens Actuators B 35

85 (1996)69 C A Grimes and D Kouzoudis Sens Actuators A 84 205

(2000)70 K G Ong C A Grimes C L Robbins and R S Singh Sens

Actuators A 93 33 (2001)71 T Seiyama N Yamazoe and H Arai Sens Actuators 4 85

(1983)72 L Ketron Ceram Bull 68 860 (1989)73 E Traversa G Gusmano and A Montenero Eur J Solid State

Inorg Chem 32 719 (1995)74 M K Jain M C Bhatnagar and G L Sharma Sens Actuators

B 55 180 (1999)75 O K Varghese and L K Malhotra J Appl Phys 87 7457 (2000)76 G Sberveglieri R Murri and N Pinto Sens Actuators B 23

177 (1995)

292

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

Fig 4 (a) The surface of a two-step anodized alumina membrane (the akes seen in the image are the ldquodustrdquo associated with tissue paper)(b) An image of a cleaved sample showing the pore channels (c) TiO2

nanowires made by lling an alumina template via sol-gel then subse-quently removing the template by a sodium hydroxide etch resulting ina TiO2 nanowire mat

3 EXPERIMENTAL RESULTSAPPLICATION TO SENSING

31 The Water Vapor Sensing Performanceof Highly Ordered NanoporousAlumina Films

Humidity sensors have attracted considerable attentionover many years because of their great importancein applications ranging from monitoring food qual-ity to meteorological studies63ndash65 Ceramic humiditysensors651 71ndash82 are commercially available and offer majoradvantages with high resistance to chemical attack

Fig 5 TiO2 nanotubes made by anodization of titanium

thermal stability mechanical strength and quick responseHowever ceramic humidity sensors still suffer from insuf- cient sensitivity over wide humidity ranges as well aslack of reversibility and drift in base resistance with timebecause of water molecule chemisorption

The humidity-sensing properties of alumina discoveredalmost 50 years ago83ndash94 are based upon ionic conduc-tion the presence of an adsorbed layer of water at thesurface reduces the total sensor impedance because of theincrease in the ionic conductivity as well as capacitancedue to the high dielectric constant of water The humidity-sensing characteristics of a given sensing element dependupon the material used the method of preparation andthe resulting surface topology Porous alumina is usuallypreferred for sensing applications because of the largesurface area available for water adsorption

An illustration of water adsorption by a surfaceredrawn from Ref 95 is shown in Figure 6 Figure6a shows the adsorption of a water molecule on aclean alumina surface Physisorption of water initiallyoccurs on an active surface site and forms an adsorption

Fig 6 Schematic illustration of interaction of a water molecule withan alumina surface redrawn from Ref 95

281

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Fig 7 Schematic illustration of buildup of adsorbed water layersredrawn from Ref 96 building of a water layer upon an alumina surface

complex (Fig 6a and b) The negatively charged oxygenof the water molecule becomes attached to the positivelycharged cation site and one of the positively chargedhydrogen atoms becomes attached to the anionic site byelectrostatic attraction Eventually they are converted totwo chemisorbed hydroxyl (OH) groups (Fig 6c) Anincrease in humidity makes the water molecules physisorbto this chemisorbed hydroxyl layer The effectiveness ofphysisorption depends upon the cation charge complexes(from alumina or the surface impurities) and the hydroxylions on the surface of the alumina The physisorptionprocess is facilitated by higher surface charge densitiesAfter hydroxyl formation the next water molecule willbe physisorbed through hydrogen double bonds on thetwo neighbouring hydroxyl groups (Fig 6d) and a protonmay be transferred from a hydroxyl group to the watermolecule to form a H3OC ion At higher humidity levelseach water molecule will be singly bonded to a hydroxylgroup as illustrated in Figure 7 (redrawn from [96]) andproton hopping between adjacent water molecules in thecontinuous water layer takes place The conduction pro-cess in this case occurs by attachment of a proton to awater molecule thereby forming a hydronium ion thehydronium ion releases another proton to a second watermolecule that accepts this proton while releasing a thirdproton and so on through the liquid Hence the dominantcharge carrier in a high-moisture atmosphere is the HC

(proton) The concentration of HC increases with increas-ing humidity and it moves freely through the water-likelayer

In the case of porous alumina capillary condensationcan take place in pores with a radius up to rK at a partic-ular relative humidity and temperature which is given byKelvinrsquos relation

rKD 2ƒM

RT ln4Ps=P5

Here P is the water vapor pressure Ps is the water vaporpressure at saturation ƒ is the surface tension R is theuniversal gas constant T is the temperature in Kelvin

H O

H O

O

H

H +

H

H

O

H +H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H O

H O

O

H

H +

H

H

O

H +H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H O

H O

O

H

H +

H

H

O

H+H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H

OH

O

O

HH

O

H+H

H+

H

OH

OH

O

O

HH

O

H+

HH

+H

O

H

O

H

O

O

HH

O

H+ HH+ H

O

H

O

H

O

O

HH

O

H+ HH+ H

O

Aluminium

AluminaCapillary Condensation

Capillary Condensation Adsorption

Adsorbed Water Layer

Fig 8 Schematic representation of water vapor interaction withnanoporous alumina illustrating some but not all97 of the multiple fac-tors inherent in determining the response

and and M are respectively the density and molecularweight of water

According this relation the condensation occurs insidesmall pores at lower humidity levels as illustrated inFigure 8 which is an illustrative drawing presentingthe different water-surface interaction mechanisms (NoteFigure 8 is not meant to be an exhaustive considera-tion of all possible effects for example the drawingdoes not address effects associated with capillary surfacecurvature97)

This condensed water reduces the impedance of alu-mina considerably No condensation occurs inside poresof radius greater than rK at a particular humidity level (seethe largest pore shown in Fig 8) In such pores as in thecase of a plane surface it is the number of physisorbedlayers that decides the impedance

Since capillary condensation enhances the sensingcapabilities of a material pore size distribution has beenwidely considered to be an important parameter in deter-mining the sensitivity in a particular humidity range941 98

However the results of our study on uniform nanoporousalumina lms presented below show that an easily builtand highly reproducible wide-range humidity sensor canbe achieved with nanodimensional pores of a narrow sizedistribution made by anodization of aluminum59ndash611 99

Starting from adhesive-backed aluminum tape (99pure) purchased from Tesa Tape Inc100 and subsequentlyanodized sensors SO50 SO30 and DO15 were preparedunder the conditions given in Table I Per sample nota-tion the rst letter denotes a single S or double D stepanodization process the second letter denotes the use of

Table I Anodization conditions used for making nanoporous alumina lms

Anodizing Duration of No ofvoltage anodization anodization

Sample (v) (min) stages

SO50 50 120 1SO30 30 120 1DO15 15 120 2

The electrolyte used in all cases was 2 wt oxalic acid

282

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

oxalic acid (O) in the anodization bath the two-digit num-ber denotes the anodization voltage It was observed thatsingle-step anodization results in disordered pore struc-tures at low voltages Hence a double-step process wasused to make sample DO15 These samples were initiallyanodized under the conditions given in Table I The sam-ples were then immersed in an aqueous solution contain-ing 18 wt chromic acid and 4 wt phosphoric acid for12 h to remove the initial nanoporous layer while keep-ing the nonporous bottom layer of alumina intact Thesamples were rinsed in distilled water and again anodizedunder the same conditions used for the rst anodizationstep Estimates of pore size distribution were determinedfrom FE-SEM images and are shown in Figure 9andashcas a size distribution histograph Samples SO50 SO30and DO15 have respectively average pore diameters of

0

10

20

30

375 40 425 45 475 50

o

f po

res

Pore diameter (nm)

0

10

20

30

25 325 375 425 475 525

o

f po

res

0

5

10

15

20

25

175 20 225 25 275 30 325 375

DO15

o

f po

res

(a)

Pore diameter (nm)

(c)

Pore diameter (nm)

(b)

Fig 9 Pore size distribution of the tested sensors (a) SO50 (b) SO30(c) DO15

Fig 10 Digital image of an impedance-based interdigital nanoporousalumina humidity sensor

452 nm 384 nm and 136 nm with a correspondingstandard deviation of 34 nm 78 nm and 49 nm Allsamples have a predominantly ordered pore structure

Immediately after anodization all samples were thor-oughly washed in distilled water and then dried in a nitro-gen atmosphere at 100 C An interdigital capacitor pat-tern with dimensions of 12 cm pound 12 cm was formed onthe surface of the alumina lms by the evaporation of gold(thickness ordm 250 nm) through a mask A digital cameraimage of a typical sensor is shown in Figure 10 To createthe necessary ambient humidity argon passed through amass ow controller (MFC) was bubbled through a bot-tle containing deionized water and then mixed with dryargon coming from another MFC in appropriate ratiosbefore being passed to the test chamber A constant total ow was maintained throughout the experiment Cham-ber humidity was monitored with the humidity probeof a Keithley 6517A electrometer and all measurementswere carried out at room temperature (23 C) Sensorimpedance was measured over the frequency range 5 Hzto 13 MHz with a computer-controlled Hewlett Packardimpedance analyzer (4192A) tted with an impedance test xture (Agilent 16034E) A signal amplitude of 90 mVwas used for all measurements Electrical contact wasmade between the sensor under test and the impedanceanalyzer test xture with 42-gauge 5-cm-long jumpersattached to the interdigital electrodes with silver pasteThe contacts were subsequently annealed at 100 C for2 h to cure the silver paste

Figure 11 shows the measured 5-kHz sensor impedanceof the different sensors as a function of humidity Thehumidity-sensitive region of DO15 is sup145 to 95 RHSensors SO30 and SO50 become sensitive to humidity atapproximately 65 and 75 RH respectively A smallerpore size increases the range of humidity values overwhich the sensor is responsive and increases the sensitiv-ity at lower humidity levels Therefore an optimal poresize can be selected for the operating region of interest

283

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

103

104

105

106

0 20 40 60 80 100

Z (SO30)

Z(SO50)

Z(DO15)

| Z| (

W)

RH ()

Fig 11 Measured 5-kHz impedance of sensors over different humiditylevels

Figure 12 shows the response of the different sensorsto humidity at different measurement frequencies It canbe seen from the gure that as the frequency decreasesthe width of the sensitive region increases and the point atwhich the sensor becomes responsive shifts to lower rela-tive humidity values For a given pore size the humidity-sensitive region of operation can be selected by frequencytuning

The response time de ned as the time needed to reach90 of the nal signal for a given relative humidityand the recovery time de ned as the time taken forthe signal to come to within 10 of the initial valuewere determined by alternately exposing the sensors toa 2 and a 45 RH ambient with impedance measuredat 5 kHz A typical responserecovery graph of sensorSO50 is shown in Figure 13 All sensors were com-pletely reversible regaining their original impedance val-ues even after repeated exposure to high humidity levelsThe response time and recovery times corresponding todifferent pore diameters are shown in Figure 14 The min-imum responserecovery times were observed for SO50the sample with the largest pore size

The impedance spectra of the sensors were taken overa range of humidity levels The total impedance wasresolved into real (Z0) and imaginary (Z00) parts and Cole-Cole impedance plots were constructed Figure 15andashcshows respectively the Cole-Cole plot of sensors SO50SO30 and DO15 The equivalent circuit models shown inFigure 16 were used to t the experimental data The t-ting was accomplished with the complex nonlinear least-square- tting program101 The dots in the plots repre-sent experimental data and lines represent the equiva-lent circuit model t The equivalent circuit denoted (a)or (b) with reference to Figure 16 used for tting isdenoted near the corresponding curves in Figure 15 Theequivalent circuit models are different from the empiri-cal model suggested by Falk et al102 for porous aluminasensors fabricated by anodization of aluminum thin lmsIn the equivalent circuit models R1 and R2 represent twofrequency-independent resistors in parallel with disper-sive frequency-dependent capacitors Cn14mdash5 and Cn24mdash5respectively103 These non-Debye capacitances can be

103

104

105

106

107

SO50

5kHz10kHz100kHz1MHz

0 20 40 60 80 100

0 20 40 60 80 100

0 20 40 60 80 100

| Z| (

W)

| Z| (

W)

| Z| (

W)

RH ()(a)

(c)

(b)

103

104

105

106

107

SO30

5kHz10kHz100kHz1MHz

RH ()

102

103

104

105

106

DO15

5kHz10kHz100kHz1MHz

RH ()

Fig 12 Variation of sensor impedance with humidity and measure-ment frequency (a) SO50 (b) SO30 (c) DO15

68

7

72

74

76

78

0 200 400 600 800

Time (Sec)

45 RH

2 RH

45 RH 45 RH

2 RH

| Z| (

x105

W)

Fig 13 Response of sensor SO50 upon repeated cycling between rel-ative humidity environments of 2 and 45

284

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

10

15

20

25

30

35

40

45

50

20 25 30 35 40 45 50

Res

pons

eR

ecov

ery

Tim

e (S

ec)

Pore diameter (nm)

DO15

SO30

SO50

Fig 14 Dependence of response time and recovery time on pore size

expressed in a general form as Cn4mdash5 D Bn4imdash5nƒ1 whereBn is a constant for a given set of experimental conditionsand n takes a value between 0 and 1 according to the localmicroscopic environment through which charge transporttakes place751 1021 103 Bn behaves as an ideal frequency-independent capacitor as n approaches 1 and as an idealconductor when the value of n approaches 0101

By examination of Figure 15a SO50 it can be seenthat at low humidity levels the Cole-Cole plot describesan arc with a very large radius of curvature As the humid-ity increases the radius decreases and at an RH of about80 the complete semicircle comes inside the measur-able impedance range However at the same humiditylevel a spur appears in the low-frequency region that dis-torts the semicircle The spur which can be viewed aspart of a second semicircle dominates at higher humid-ity levels while the semicircle described by the high-frequency impedance values diminishes in size Thisbehavior is similar for all sensors For sensors SO50SO30 and DO15 respectively the RH at which thehigh-frequency semicircle comes completely within themeasurable impedance range is ordm75 65 and 35these values roughly match the in ection points seen inFigure 11

The fundamental mechanism that enables ceramic sen-sors to sense humidity is the physisorption of watermolecules on an initially chemisorbed layer of hydroxylions651 96 The chemisorbed hydroxyl ions enhance theelectrical conductivity of the sensor either by donatingelectrons to the conduction band of the base materialor through proton hopping between adjacent hydroxylgroups upon the application of an electric eld Theprocess of chemisorption occurs at very low humiditylevels and is unaffected by further changes in humid-ity However an increase in humidity makes the watermolecules physisorb to this hydroxyl layer The effec-tiveness of physisorption depends upon the cation chargecomplexes (from the material or the impurities) and thehydroxyl ions (water molecules present at the surface ofthe base material)96 The physisorption process is facili-tated by higher surface charge densities During formationof the rst physisorbed layer a water molecule becomesattached to two neighboring hydroxyl groups through

0

02

04

06

08

1

12

0 02 04 06 08 1 12

Zrsquo (MW)

35 (a) 55 (a)

65 (a)

75 (b)

81 (b)

93 (b) SO50

0

02

04

06

08

1

12

14

0 02 04 06 08 1 12 14

Zrsquo (W)

Z

(W)

Z

(MW

)Z

(M

W)

35 (a)

55 (a)

65 (b)

75 (b)

81 (b)

93 (b)

SO30

0

2 105

4 105

6 105

8 105

1 106

12 106

0 2 105 4 105 6 105 8 105 1 106 12 106

17 (a)27 (a)

35 (b)45 (b)55 (b)

93 (b)

DO15

(a)

Zrsquo (MW)

(b)

(c)

Fig 15 Cole-Cole plot of the sensors at different humidity levels(a) SO50 (b) SO30 (c) DO15 The points represent the experimentaldata and lines represent the t with the equivalent circuit model givenin Figure 9 The RH values at which the data were collected and theequivalent circuit used for tting each curve (given as (a) or (b) denot-ing the circuit of Figure 17a and b respectively) are shown adjacent toeach curve

hydrogen double bonds and a proton may be transferredfrom a hydroxyl group to the water molecule to form aH3OC ion

When the physisorption occurs within less than a monolayer that is when clusters of physisorbed molecules

285

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

(b)

(a)

Cn1 (w)

R1

Cn2 (w) Cn1 (w)

R1 R2

Fig 16 Equivalent circuit used for tting the experimental data

are present at the surface it is thought that H3OC diffu-sion within hydroxyl groups and proton transfer betweenadjacent water molecules clusters take place96 HoweverKhanna and Nahar104ndash106 refute this theory They arguethat the proton diffusion is less probable in a surfacethat contains hydroxyl ions and clusters of physisorbedmolecules and hence cannot account for the conductiv-ity of porous alumina at an RH less than 40 that isbefore the rst monolayer is formed Instead they sug-gested a phonon-assisted electron tunneling mechanism atlow humidity levels in which electron tunneling occursfrom one water molecule to another as enhanced by thesurface anions of the alumina sensors

At higher humidity levels the number of physisorbedlayers increases allowing each water molecule to besingly bonded to a hydroxyl group and proton hop-ping between adjacent water molecules in the continu-ous water layer takes place The conduction process isthe same as that of pure water and is called the Grot-thuss chain reaction96 The dominant charge carrier ina high-humidity environment is therefore HC ions (pro-tons) The concentration of HC increases with increas-ing humidity and HC move freely through the water-likelayer According to the theory of Khanna and Nahar104

in porous alumina the protons for conduction are donatednot only by H2O but also by the Al(OH)3 formed asa result of water adsorption and the impurity anionsincorporated into the lm from the electrolyte duringanodization107ndash109 The water-like network consists ofsingly bonded water molecules which possess a highdielectric constant as they form dipoles and reorient freelyunder an externally applied electric eld The impedanceof the nanoporous alumina sensors includes this capaci-tive contribution as well

From the analysis of Figure 15 the high-frequencysemicircle that diminishes in size with increasing RHarises because of the effect of a physisorbed water layerat the surface Therefore R1 in the equivalent circuitof Figure 16 represents the charge transfer resistancethrough the adsorbed water layer The effect of electronic

102

104

106

108

SO50SO30

DO15

R1

(W)

0 20 40 60 80 100

RH ()

Fig 17 Humidity dependence of resistance R1 of different sensorsobtained by tting the equivalent circuit model with experimental data

conduction through the alumina bulk is parallel to thatof the surface conduction affected by the adsorbed waterlayer This bulk conduction process is dominant only atextremely low humidity values when there are only afew adsorbed molecules Hence the incomplete semicir-cle or arc in the low-humidity region is partially deter-mined from the base alumina Bn1 represents the capac-itance contribution from the adsorbed water layer Thelarge spurs seen in the low-frequency region are a con-sequence of ion migration in the adsorbed layer towardelectrodes1011 103 This migration leads to accumulation ofions at the electrodes Since the relaxation time of ionmigration is larger than the charge transfer process itis seen only at low frequencies The accumulated ionsin the adsorbed water layer give rise to R2 and Bn2From the equivalent circuits given in Figure 16 it canbe seen that the presence of ion accumulation repre-sented by the combination of R2 and Cn24mdash5 increasesthe overall impedance of the sensor Hence the reductionin impedance caused at high humidity due to the chargetransfer process is countered by ion accumulation

To better understand the fundamental sensing mecha-nisms R1 and Bn1 obtained through the numerical ttingprogram for the different sensors are plotted in Figures 17and 18 respectively It can be seen from Figure 17 thatR1 which is the resistance to charge transport in a staticelectric eld by the material containing an adsorbed waterlayer in the absence of ion migration is linear with RH

-11

10-10

10-9

10-8

10-7

50V30VDO15

Bn1

0 20 40 60 80 100

RH ()

Fig 18 Humidity dependence of Bn1 of different sensors obtained by tting the equivalent circuit model (Fig 17) with experimental data

286

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

in a semilog plot For sensor DO15 R1 decreases to afew hundred ohms at high RH levels indicating that theobserved conductivity is not due merely to the protonsprovided by the adsorbed water layer Since deionizedwater and argon gas are used to create the required humid-ity environment it is reasonable to assume that the sur-faces of the alumina lms contain anions presumablyfrom the electrolyte used for anodization which controlsconduction through the adsorbed layer108

With reference to Figure 18 Bn1 of the larger pore sam-ples SO50 and SO30 increases signi cantly only at thehigher humidity levels This result is in contrast to thatof the smaller pore size sensors which shows that Bn1

increases at lower humidity levels The relative humid-ity values where a signi cant increase in Bn1 takes placegenerally match the point seen in Figure 11 where theimpedance drops signi cantly which in turn is a func-tion of pore size These same relative humidity valuescorrelate with the point at which the spur appears inthe impedance diagrams given in Figure 15 The resultslead to the following conclusions Since hydrogen-bondedwater molecules possess a high dielectric constant theincrease in Bn1 indicates the presence of these bondson the surface which means that a liquid-like networkmore than one monolayer of physisorbed molecules hasformed at that humidity level This liquid-like network isresponsible for the sharp reduction in impedance seen inFigure 11 and the ion migration that creates the spur inthe low-frequency portion of the Cole-Cole plots as seenin Figure 15 According to Eq (1) capillary condensationdoes not take place in the alumina test sensors

In light of the above observations we propose thefollowing model to explain the behavior of nanoporousalumina in response to humidity The anodization pro-cess results in a certain amount of electrolyte anionstrapped within the pores1041 1071 108 Even in a low-humidityregime the presence of these anions on the sensor sur-face provides a high charge density for easy physisorp-tion of water molecules The result is the formationof a liquid-like network within the pores The impu-rity anions also act as proton donors by a dissociativemechanism104 joining the protons donated by the watermolecules The smaller the pore size the lower the RHlevel at which the liquid-like network is formed Hencefor the small pore sensors there is variation in Bn1 aswell as impedance magnitude in low-humidity regimesthe exponential dependence of R1 on RH also suggeststhe same phenomenon

In the case of DO15 the rst semicircle is highly dis-torted by the low-frequency spur originating with ionmigration A relatively large number of mobile ions mightbe present in these sensors leading to an enhancedaccumulation of protons at the electrode-sample contactAs mentioned earlier the dipoles in the singly bondedadsorbed water layer are free to rotate in an electric eld

Hence in the region close to the electrode-alumina inter-face the adsorbed water layer exhibits a net dipolar ori-entation and therefore it does not display its normaldielectric constant1011 103 Depending on ion size up toa monolayer of charge can exist at the electrode101 thediffuse (Gouy) layer of counter-ions spreads away fromthis inner (Stern) layer The number of ions available toform the Stern and Gouy layers at a particular frequencyincreases with increasing number of physisorbed layersHence as the number of physisorbed layers increases theStern and Gouy layers start forming at higher frequencies

With reference to Figure 12 the relative humiditylevel at which the sharp decrease in impedance occursshifts higher with increasing frequency This result can beexplained with the equivalent circuit shown in Figure 16athat is the circuit in the absence of ion accumulation Athigh frequencies the reactive impedance Xc of Cn14mdash5equal to Xc

D 1=jmdashCn14mdash5 is low and hence impedanceis largely determined by R1 As the humidity increasesR1 decreases however this translates into a reduction inimpedance only at relatively low frequencies where thevalue of R1 is comparable to or less than the capacitivereactance of Cn1(mdash) As the measurement frequency isreduced the reactance of Cn1(mdash) decreases hence the effectof R1 becomes more signi cant at lower humidity levels

32 Hydrogen Sensing with Titania Nanotubes

We recently reported62 the fabrication of self-organizedtitania nanotube arrays with an anodization techniqueAlthough the as-prepared nanotubes are amorphous theycrystallize on annealing at elevated temperatures and arestructurally stable to at least 600 C This stability ofstructure which is an essential criterion of a gas-sensingmaterial prompted us to study the gas-sensing behavior ofthese nanotubes for technologically important gases suchas oxygen carbon monoxide ammonia carbon dioxideand hydrogen

Titania has earned much attention for its oxygen-sensing capabilities110ndash115 Furthermore with propermanipulation of the microstructure or crystalline phaseandor the addition of the proper impurities or surfacefunctionalization titania can also be used as a reducinggas sensor116ndash127 The interaction of a gas with a metaloxide semiconductor is primarily a surface phenomenonand nanoporous metal oxides1221 1231 1281 129 offer the advan-tage of providing large sensing surface areas

Room-temperature metal oxide hydrogen sensors aregenerally based on Schottky barrier modulation of deviceslike PdTiO2 or PtTiO2 by hydrogen130ndash132 Elevated-temperature hydrogen sensors examine electrical resis-tance with hydrogen concentration for example Birke-feld et al133 observed that the resistance of the anatasephase of titania varies in the presence of carbon monoxideand hydrogen at temperatures above 500 C but on dop-ing with 10 alumina it becomes selective for hydrogen

287

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Barrier layer

Titanium Nano tube array

Platinum electrodes

Fig 19 Schematic representation of the hydrogen sensor geometry

Munuera et al134 used rhenium doping in the anatasephase to perform room-temperature hydrogen monitor-ing Recently Shimizu et al122 reported that anodizednanoporous titania lms with a Pd Schottky barrier aresensitive to hydrogen at 250 C

Titania nanotubes were grown from titanium foil(995 pure from Alfa Aesar Ward Hill MA) with athickness of 025 mm The anodization was performed inan electrolyte medium of 05 hydro uoric acid in waterwith the use of a platinum foil cathode Well-de nednanotube arrays were grown with anodizing potentialsranging from 12 V to 20 V Nanotube length increaseswith anodization time reaching 400 nm in approximately20 min and then remains constant For the present studythe samples were anodized for 25 min The samples werethen annealed at 500 C in a pure oxygen ambient for 6 hwith a heating and cooling rate of 1 Cmin

The electrode geometry of the titania nanotube sen-sors is shown in Figure 19 The sensor consists of a basetitanium metal foil with a nanotube array grown on topAn insulating barrier layer separates the nanotubes fromthe conducting titanium foil Electrical connections weremade by two parallel 10 mm pound 2 mm platinum pads of100-Œm thickness The test chamber consists of a 13-literquartz tube with stainless steel end caps placed insidea furnace (Thermolyne USA model 21100 tubular fur-nace) Gas ow through the test chamber was controlledvia a computer-controlled mass ow controller The elec-trical resistance of the titania sensors was measured witha computer-controlled Keithly digital multimeter Prior todata collection the test chamber was initially evacuatedwith a mechanical pump whereupon nitrogen (99999pure) was passed while the sensor under test was heatedto the desired temperature The test gases were mixed inappropriate ratios with nitrogen to create the necessarytest gas ambient

The surface morphology of nanotube arrays preparedwith an anodization potential of 20 V and annealed at500 C for 6 h in a pure oxygen ambient is shown inFigure 20a and b It can be seen from these images thatthe nanotubes are uniform over the surface The nano-tubes are approximately 400 nm in length and have abarrier layer62 thickness of ordm50 nm For the nanotubes

100 nm

(a)

(b)

1 mm (c)

100 nm Fig 20 The surface morphology of the titania nanotubes after anneal-ing at 500 C (a) High- and (b) low-magni cation images of a 20-Vsample (c) A high-magni cation image of a 12-V sample

fabricated with 20-V anodization the average pore diam-eter as determined from FESEM images is 76 nm (thestandard deviation 15 nm) with a wall thickness of 27 nm(standard deviation 6 nm) The sample anodized at 12 Vshown in Figure 2c was found to have an average porediameter of 46 nm (standard deviation 8 nm) with a wallthickness of 17 nm (standard deviation 2 nm) The porosi-ties of the 20-V and 12-V samples were calculated as 45and 61 respectively

Glancing-angle X-ray diffraction patterns of a 20-Vsample annealed at 500 C for 6 h in oxygen ambientis shown in Figure 21 It can be seen that both anataseand rutile phases of titania are present in the sampleA detailed study135 of these structures by high-resolutiontransmission electron microscope (HRTEM) showed that

288

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

20 30 40 50 60 70 80

2 q (deg)

(101

)

(110

)

A

R

TR

T

T

R R A TRR T R R T TT

Fig 21 Glancing-angle X-ray diffraction pattern of a 20-V sample(glancing angle D 2) annealed at 500 C in oxygen ambient A R andT represent respectively the re ections from anatase crystallites rutilecrystallites and the titanium substrate

the anatase crystallites were concentrated on the walls ofthe nanotubes and rutile on the barrier layer

Nanotubes annealed in a pure oxygen ambient werefound to be stable (intact) to temperatures of approxi-mately 580 C Above this temperature protrusions wereseen to be coming out through the nanotubes an effectthat spread with increasing temperature These protru-sions which are due to oxidation of the titanium substratecollapse the nanotubes

Figure 22 shows the response of the 20-V (76 nm) sam-ple as a function of ambient temperature as it is switchedfrom a nitrogen environment to one containing 1000 ppmhydrogen and then back to nitrogen The plot was madeusing 4Rg=R05

ƒ1 versus time where R0 is the base resis-tance of the sensor that is the sensor resistance before thetest gas was introduced and Rg is the measured resistancein the presence of test gas The sensor shows increasinghydrogen sensitivity with temperature with a change by3 orders of magnitude in resistance at temperatures aboveordm300 C At all of the temperatures the original resistancerecovered without hysteresis

The sensitivity S is de ned by the formula

S DR0

ƒ Rgs

Rgs

10 0

10 1

10 2

10 3

0 10 0 5 10 2 1 10 3 15 103 2 10 3 25 103 3 10 3 35 103 4 10 3

Time (Sec)

(Rg

R0)

-1

180degC

240degC

265degC

290degC

345degC

375degC

H2+N

2

N2

Fig 22 Variation of resistance Rg normalized with respect to baselineresistance R0 of a 20-V sample with time on exposure to 1000 ppmhydrogen at different temperatures It may be noted that the inverse ofRg=R0 was used in the plot to represent data in the positive y direction

10-1

100

101

102

103

150 200 250 300 350 400 450

Temperature (degC)

Sens

itivi

ty (

R0-

Rgs

)R

gs

Fig 23 The sensitivity temperature dependence of a 20-V sample to1000 ppm hydrogen

where R0 is the resistance of the sensor before the gas ispassed and Rgs is the resistance after gas is passed andreaches the saturation value The temperature-dependentsensitivity of a 20-V sample to 1000 ppm hydrogen isshown in Figure 23 Sensitivity is seen to increase withtemperature to approximately 380 C where the increasein sensitivity with temperature begins to saturate

The response time de ned as the time needed for thesensor to reach 90 of the nal signal for a given concen-tration of gas is plotted against temperature in Figure 24(the time includes that required for the gas to equili-brate inside the measurement chamber estimated to beordm30 s) The response time decreases exponentially withtemperature

To check the behavior of the sensor on repeated hydro-gen exposure the hydrogen concentration was varied indiscrete steps of 100 ppm from 0 to 500 ppm while thetemperature was kept constant at 290 C the chamber was ushed with nitrogen after each exposure to hydrogenThe response of the 20-V prepared nanotube sensorskept at 290 C is shown in Figure 25 The behaviorof the sensor is consistent recovering its original resis-tance after repeated exposure to varying hydrogen gasconcentrations The sensitivities of the 76-nm- 53-nm-and 22-nm-diameter nanotube sensors to hydrogen con-centrations ranging from 100 ppm to 1 are shown inFigure 26 The sensitivity of the 76-nm (20-V) sensor tolow concentrations is shown in Figure 27 there is a linearincrease in sensitivity at low concentrations The lower

10 1

10 2

10 3

10 4

150 200 250 300 350 400 450

Temperature (degC)

Res

pons

e T

ime

(Sec

)

Fig 24 Variation in response time for a 20-V sensor as a function oftemperature The dots represent measured data

289

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

10 2

0 4 4 4 4 4

10 3

0 10 1 10 2 10 3 10 4 10 5 10

Time (Sec)

Res

ista

nce

(W)

H2 (ppm)

200

300

400

500

400

300

200

100

100

Fig 25 Resistance of a 20-V sample when exposed to different con-centrations of hydrogen at 290 C The nitrogen-hydrogen mixture waspassed for 1500 s the chamber was then ushed with nitrogen for 3000 sbefore the nitrogen-hydrogen mixture was passed again

test limit of 20 ppm is determined by the experimentalapparatus and not the sensor

Figure 28 shows the change in resistance of the 76-nm-46-nm- and 22-nm-diameter nanotube sensors with expo-sure to 1000 ppm hydrogen at 290 C Whereas smallerdiameter nanotubes had greater sensitivity to hydrogenthe samples made at lower anodizing voltages tended tobe more brittle and harder to handle mechanically withoutbreaking

The 20-V 76-nm-diameter sample was exposed to oxy-gen carbon monoxide ammonia and carbon dioxide at290 C to study the cross-sensitivity The sensor wasfound to have no detectable variation in resistance onexposure to carbon dioxide The sensitivities of the tita-nia nanotubes to the other gases are shown in Figure 29The sensitivities of the nanotubes to carbon monoxide andammonia are negligible compared with that of hydrogenThe resistance of the nanotubes increased in the presenceof oxygen and did not regain their original electrical con-ductivity even after several hours in a 290 C nitrogenenvironment

1

10

100

1000

104

105

0 02 04 06 08 1

Hydrogen Concentration ()

Pore diameter = 76 nm

Pore diameter = 53 nm

Pore diameter = 22nm

Sen

sitiv

ity (

R0-

Rg)

Rg

Fig 26 The sensitivity variation of a samples with pore diameters of22 nm 53 nm and 76 nm versus time and hydrogen concentrations

0 100 200 300 400 5001

10

100

Hydrogen concentration (ppm)

Sen

sitiv

ity

(R0-

Rgs

)R

gs

Fig 27 Sensitivity of a 20-V (76 nm) sample at 290 C for low hydro-gen concentrations

Since the sensor measurements were made in atmo-spheres without oxygen the increase in conductivitycannot be due to hydrogen removing oxygen from thelattice132ndash134 or the removal of chemisorbed oxygen135ndash137

Hydrogen molecules can dissociate at defects on thetitania surface which can subsequently diffuse into thetitania lattice acting as electron donors1281 1381 139 How-ever a dissociation-driven process would result in aslow response and recovery times with complete recov-ery unlikely since the sensor completely regains its orig-inal resistance with hydrogen cycling it appears that thisis not the dominant mechanism behind high hydrogensensitivity We believe chemisorption of the dissociatedhydrogen on the titania surface is the underlying sensingmechanism137 During chemisorption hydrogen acts as asurface state and a partial charge transfer from hydrogento the titania conduction band takes place This createsan electron accumulation layer on the nanotube surfaceenhancing its electrical conductance Upon removal of thehydrogen ambient electron transfer back to the hydro-gen molecule takes place which subsequently desorbsrestoring the original electrical resistance of the mate-rial Another factor that may play a role in the hydrogensensitivity (and selectivity) is the platinum electrodes Atelevated temperatures hydrogen dissociation can occur onplatinum surfaces These dissociated hydrogen atoms mayspill1361 140 onto the nanotube surface where they diffuse

10-1

101

103

105

0 05 1 15

Time (Hour)

Pore size - 76 nm

Pore size - 53 nm

Pore size - 22 nm

(Rg

R0)

-1

Fig 28 A comparison of the variation in resistance of samples havingpore diameters of 22 nm 53 nm and 76 nm versus time upon exposureto 1000 ppm of hydrogen at 290 C

290

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

10-2

10-1

100

0 100 2 102 4 102 6 102 8 102 1 103

OxygenAmmoniaCarbon monoxide

Time (Sec)

(Rg

R0)

-1

Fig 29 Variation of resistance of a 20-V sample when exposed to1000 ppm carbon monoxide 5 ammonia and 1 oxygen at 290 C

into the material affecting its electrical properties Fromthe present study it was not clear how signi cant a rolethe platinum electrodes play

Anatase the polymorph of titania has been reported tohave high sensitivity for reducing gases like hydrogen andcarbon monoxide1211 1241 133 Our nanotube samples containanatase phase mainly on the walls and rutile in the barrierlayer As the diffusing hydrogen atoms go to the intersti-tial sites1331 142 and as the ca ratio of anatase is almostfour times that of rutile it appears that the anatase latticeaccommodates hydrogen easily and hence makes a highercontribution to hydrogen sensitivity

The effect of chemisorption can be neglected in theoxygen-sensing experiments As the recovery requiresseveral hours it appears that the nanotubes contain oxy-gen vacancies or titanium interstitial defects in the pres-ence of nitrogen On exposure of the sensor to oxygenambient the lattice reoxidizes and hence the conductiv-ity of the sensor decreases On the removal of oxygenthe reduction of the lattice will not immediately occurhence the sensor requires several hours to regain its orig-inal conductivity

4 CONCLUSIONS

This paper summarizes different routes of fabrication ofmetal oxide nanoarchitectures useful for sensing applica-tions Structures made via anodization of a starting metaland breath gure formation from a sol-gel are discussedand presented as are TiO2 nanowires made by lling ananoporous alumina membrane with subsequent removalof the supporting template Two test cases illustratingthe utility of these nanoporous metal oxide structures forsensing are presented highly ordered nanoporous aluminais used for humidity sensing and TiO2 nanotubes are usedfor hydrogen sensing These study test cases reveal sev-eral points worth brie y noting

In the rst sensing test case uniform nanoporous alu-mina sensors exhibit pore size and frequency-dependent

humidity-sensing characteristics The mechanism respon-sible for the sensing action of the nanoporous alumina sen-sors was studied by representing the impedance spectra atdifferent humidity levels in Cole-Cole plots This makes itpossible to identify the competing impedance-determiningprocesses occurring in the sensors namely charge transferand mass transport through the physisorbed water layeras dependent upon the humidity level and input signalfrequency Charge transfer was found to enhance the sen-sitivity by causing a large variation in impedance withrespect to humidity Mass transport leads to accumulationof ions at the electrodes which tends to decrease the mea-surement sensitivity by increasing the impedance Hencea proper choice of operating frequency high enough toavoid ion accumulation is essential for optimum sensorperformance

It has been inferred from our results that anions becomeincorporated into the alumina lm during the anodiza-tion process supporting the hypothesis of Khanna andNahar104ndash106 where they act as proton donors and facili-tate the adsorption of water molecules Hence liquid-likecharge transfer networks are formed at relatively lowerhumidity levels for the smaller pore-size sensors Thisprocess rather than capillary condensation is primarilyresponsible for the observed humidity-sensing behavior ofthe sensors Our studies show that a distribution of poresizes as suggested by Shimizu et al94 is not necessary toobtain sensitive performance over a wide humidity rangerather such optimal performance can be obtained with uni-form nanodimensional pores containing adsorbed anionicimpurities

In the second sensing test case titania nanotubes pre-pared by anodization and annealed in an oxygen atmo-sphere at a temperature of 500 C were found to behighly sensitive to hydrogen The nanotube sensors con-tained both anatase and rutile phases of titania and showedappreciable sensitivity toward hydrogen at temperaturesas low as 180 C The sensitivity increased drasticallywith temperature showing a variation of 3 orders in mag-nitude of resistance to 1000 ppm of hydrogen at 400 CThe response time decreased with increasing temperatureat 290 C full switching of the sensor took approximately3 min Results were highly reproducible with no indica-tion of hysteresis Our results showed that these sensorsare at least capable of monitoring hydrogen levels from20 ppm to 4 At 290 C nanotubes with smaller porediameters showed higher sensitivity to hydrogen com-pared with their larger pore counterparts The sensorsshowed high selectivity to hydrogen compared with car-bon monoxide ammonia and carbon dioxide Althoughthe sensor was sensitive to high concentrations of oxy-gen the response time was high and the sensor did notcompletely regain the original condition We believe thehydrogen sensitivity of the nanotubes is due to hydrogendiffusion into the titania lattice where they act as electrondonors

291

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Acknowledgments Support of this work by the NationalScience Foundation through grant ECS-0210033 is grate-fully acknowledged

References and Notes

1 K Kajihara K Tanaka K Horao and N Soga Jpn J ApplPhys 35 6110 (1996)

2 B O Regan and M Gratzel A low cost Nature 353 737 (1991)3 A Fujishima and K Honda Nature 238 37 (1972)4 K Sato A Tsuzuki H Taoda Y Torii T Kato and Y Butsugan

J Mater Sci 29 5911 (1994)5 K Sato A Tsuzuki Y Torii H Taoda T Kato and Y Butsugan

J Mater Sci 30 837 (1995)6 N Ozer and C M Lampert Sol Energy Mater Sol Cells 54 147

(1998)7 A Manz and H Becker editors Microsystem Technology in

Chemistry and Life Sciences Springer-Verlag Berlin 19988 E Kim Y Xia and G M Whitesides Nature 376 581 (1995)9 M Templin A Franck A Du Chesne A Leist A Zhang

R Ulrich V Schadler and U Weisner Science 278 1795 (1997)10 D Zhao J Feng Q Huo N Melosh G H Frederickson

B Chmelka and G D Stucky Science 279 548 (1998)11 O D Velev T A Jede R F Lobo and A M Lenhoff Nature

389 447 (1997)12 B T Holland C F Blanford and A Stein Science 281 538

(1998)13 S Matsushita T Miwa and A Fujishima Chem Lett 9 925

(1997)14 A Imhof and D J Pine Adv Mater 10 697 (1998)15 T Tatsuma A Ikezawa Y Ohko T Miwa T Matsue and

A Fujishima Adv Mater 12 643 (2000)16 Z Zhong Y Yin B Gates and Y Xia Adv Mater 12 206 (2000)17 O D Velev and A M Lenhoff Curr Opin Colloids Interface

Sci 5 56 (2000)18 O D Velev and E W Kaler Adv Mater 12 531 (2000)19 P Jiang J Cizeron J F Bertone and V L Colvin J Am Chem

Soc 121 7957 (1999)20 M E Turner T J Trentler and V L Colvin Adv Mater 13 180

(2001)21 K M Kulinowski P Jiang H Vaswani and V L Colvin Adv

Mater 12 833 (2000)22 P Jiang J F Bertone and V L Colvin Science 291 453 (2001)23 R Seshadri and F C Meldrum Adv Mater 12 1149 (2000)24 F C Meldrum and R Seshadri Chem Commun 29 (2000)25 Q B Meng Z Z Gu O Sato and A Fujishima Appl Phys

Lett 77 4313 (2000)26 G Widawski B Rawiso and B Francois Nature 369 3897

(1994)27 B Francois O Pitois and J Francois Adv Mater 7 1041 (1995)28 O Pitois and B Francois Eur Phys J B 8 225 (1999)29 S A Jenekhe and X L Chen Science 283 372 (1999)30 M Shimomuara and T Sawadaishi Curr Opin Colloid Interface

Sci 6 11 (2001)31 O Karthaus N Maruyama X Cieren M Shimomura

H Hasegawa and T Hashimoto Langmuir 16 6071 (2000)32 T Nishikawa J Nishida R Ookura S Nishimura S Wada

T Karino and M Shimomura Mater Sci Eng C 10 141 (1999)33 K Kajihara K Nakanishi K Tanaka K Hirao and N Soga

J Am Ceram Soc 81 2670 (1998)34 C A Grimes R S Singh E C Dickey and O K Varghese

J Mater Res 16 1686 (2001)35 M Srinivasrao D Collings A Phillips and S Patel Science 292

79 (2001)36 P G de Gennes Rev Mod Phys 57 827 (1985)

37 D Beysens A Steyer P Guenoun D Fritter and C M KnoblerPhase Transitions 31 219 (1991)

38 B J Briscoe and K P Galvin J Phys D 23 422 (1990)39 B J Briscoe and K P Galvin J Phys D 23 1265 (1990)40 D Beysens and C M Knobler Phys Rev Lett 57 1433 (1986)41 F Family and P Meakin Phys Rev Lett 61 428 (1988)42 A V Limaye R D Narhe A M Dhote and S B Ogale Phys

Rev Lett 76 3762 (1996)43 P Dellrsquo Aversana and G P Neitzel Phys Today 51 38 (1998)44 P Dellrsquo Aversana J R Banavar and J Koplik Phys Fluids 8 15

(1986)45 P Dellrsquo Aversana V Tontodonato and L Carotenuto Phys Fluids

9 2475 (1997)46 M Q Li S Q Xu and E Kumacheva Langmuir 16 7275 (2000)47 P Freymuth Rev Sci Instrum 64 1 (1993)48 P Ball The Self-Made Tapestry Pattern Formation in Nature

Oxford University Press New York 199949 M F Schatz S J VanHook W D McCormick J B Swift and

H L Swinney Phys Fluids 11 2577 (1999)50 S R Coriell M R Cordes and W J Boettinger J Cryst Growth

49 13 (1980)51 T S Sullivan Y M Liu and R E Ecke Phys Rev E 54 486

(1996)52 J L Rogers M F Schatz J L Bougie and J B Swift 84 87

(2000)53 K Nitschke 52 R5772 (1995)54 E Bodenschatz J R De Bruyn G Ahlers and D S Cannell

Phys Rev Lett 67 3078 (1991)55 C W Meyer D S Cannell G Ahlers J B Swift and P C

Hohenberg Phys Rev Lett 61 947 (1988)56 R F Ismagilov D Rosmarin D H Gracias A D Stroock and

G M Whitesides Appl Phys Lett 79 439 (2001)57 E A Barringer and H K Bowen Langmuir 1 420 (1985)58 M Tschapek C Wasowski and R M T Sanchez J Electroanal

Chem 74 167 (1976)59 H Masuda and K Fukuda Science 268 1466 (1995)60 H Masuda F Hasegwa and S Ono J Electrochem Soc 144

L 127 (1997)61 O Jessensky F Muller and U Gosele Appl Phys Lett 72 1173

(1998)62 D Gong C A Grimes O K Varghese W Hu R S Singh

Z Chen and E C Dickey J Mater Res 16 3331 (2001)63 B M Kulwicki J Am Ceram Soc 74 697 (1991)64 H Arai and T Seiyama in Sensors A Comprehensive Survey

edited by W Gopel J Hesse and J N Zemel VCH Weinheim(1992) Vol 3 pp 981ndash1012

65 E Traversa Sens Actuators B 23 135 (1995)66 N Yamazoe and Y Shimizu Sens Actuators 10 379 (1986)67 A Bearzotti I Fratoddi L Palummo S Petrocco A Furlani

C Lo Sterzo and M V Russo Sens Actuators B 76 316 (2001)68 Y Sakai Y Sadaoka and M Matsuguchi Sens Actuators B 35

85 (1996)69 C A Grimes and D Kouzoudis Sens Actuators A 84 205

(2000)70 K G Ong C A Grimes C L Robbins and R S Singh Sens

Actuators A 93 33 (2001)71 T Seiyama N Yamazoe and H Arai Sens Actuators 4 85

(1983)72 L Ketron Ceram Bull 68 860 (1989)73 E Traversa G Gusmano and A Montenero Eur J Solid State

Inorg Chem 32 719 (1995)74 M K Jain M C Bhatnagar and G L Sharma Sens Actuators

B 55 180 (1999)75 O K Varghese and L K Malhotra J Appl Phys 87 7457 (2000)76 G Sberveglieri R Murri and N Pinto Sens Actuators B 23

177 (1995)

292

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Fig 7 Schematic illustration of buildup of adsorbed water layersredrawn from Ref 96 building of a water layer upon an alumina surface

complex (Fig 6a and b) The negatively charged oxygenof the water molecule becomes attached to the positivelycharged cation site and one of the positively chargedhydrogen atoms becomes attached to the anionic site byelectrostatic attraction Eventually they are converted totwo chemisorbed hydroxyl (OH) groups (Fig 6c) Anincrease in humidity makes the water molecules physisorbto this chemisorbed hydroxyl layer The effectiveness ofphysisorption depends upon the cation charge complexes(from alumina or the surface impurities) and the hydroxylions on the surface of the alumina The physisorptionprocess is facilitated by higher surface charge densitiesAfter hydroxyl formation the next water molecule willbe physisorbed through hydrogen double bonds on thetwo neighbouring hydroxyl groups (Fig 6d) and a protonmay be transferred from a hydroxyl group to the watermolecule to form a H3OC ion At higher humidity levelseach water molecule will be singly bonded to a hydroxylgroup as illustrated in Figure 7 (redrawn from [96]) andproton hopping between adjacent water molecules in thecontinuous water layer takes place The conduction pro-cess in this case occurs by attachment of a proton to awater molecule thereby forming a hydronium ion thehydronium ion releases another proton to a second watermolecule that accepts this proton while releasing a thirdproton and so on through the liquid Hence the dominantcharge carrier in a high-moisture atmosphere is the HC

(proton) The concentration of HC increases with increas-ing humidity and it moves freely through the water-likelayer

In the case of porous alumina capillary condensationcan take place in pores with a radius up to rK at a partic-ular relative humidity and temperature which is given byKelvinrsquos relation

rKD 2ƒM

RT ln4Ps=P5

Here P is the water vapor pressure Ps is the water vaporpressure at saturation ƒ is the surface tension R is theuniversal gas constant T is the temperature in Kelvin

H O

H O

O

H

H +

H

H

O

H +H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H O

H O

O

H

H +

H

H

O

H +H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H O

H O

O

H

H +

H

H

O

H+H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H O

H O

O

H

H+

H

H

O

H+H

O

H

OH

O

O

HH

O

H+H

H+

H

OH

OH

O

O

HH

O

H+

HH

+H

O

H

O

H

O

O

HH

O

H+ HH+ H

O

H

O

H

O

O

HH

O

H+ HH+ H

O

Aluminium

AluminaCapillary Condensation

Capillary Condensation Adsorption

Adsorbed Water Layer

Fig 8 Schematic representation of water vapor interaction withnanoporous alumina illustrating some but not all97 of the multiple fac-tors inherent in determining the response

and and M are respectively the density and molecularweight of water

According this relation the condensation occurs insidesmall pores at lower humidity levels as illustrated inFigure 8 which is an illustrative drawing presentingthe different water-surface interaction mechanisms (NoteFigure 8 is not meant to be an exhaustive considera-tion of all possible effects for example the drawingdoes not address effects associated with capillary surfacecurvature97)

This condensed water reduces the impedance of alu-mina considerably No condensation occurs inside poresof radius greater than rK at a particular humidity level (seethe largest pore shown in Fig 8) In such pores as in thecase of a plane surface it is the number of physisorbedlayers that decides the impedance

Since capillary condensation enhances the sensingcapabilities of a material pore size distribution has beenwidely considered to be an important parameter in deter-mining the sensitivity in a particular humidity range941 98

However the results of our study on uniform nanoporousalumina lms presented below show that an easily builtand highly reproducible wide-range humidity sensor canbe achieved with nanodimensional pores of a narrow sizedistribution made by anodization of aluminum59ndash611 99

Starting from adhesive-backed aluminum tape (99pure) purchased from Tesa Tape Inc100 and subsequentlyanodized sensors SO50 SO30 and DO15 were preparedunder the conditions given in Table I Per sample nota-tion the rst letter denotes a single S or double D stepanodization process the second letter denotes the use of

Table I Anodization conditions used for making nanoporous alumina lms

Anodizing Duration of No ofvoltage anodization anodization

Sample (v) (min) stages

SO50 50 120 1SO30 30 120 1DO15 15 120 2

The electrolyte used in all cases was 2 wt oxalic acid

282

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

oxalic acid (O) in the anodization bath the two-digit num-ber denotes the anodization voltage It was observed thatsingle-step anodization results in disordered pore struc-tures at low voltages Hence a double-step process wasused to make sample DO15 These samples were initiallyanodized under the conditions given in Table I The sam-ples were then immersed in an aqueous solution contain-ing 18 wt chromic acid and 4 wt phosphoric acid for12 h to remove the initial nanoporous layer while keep-ing the nonporous bottom layer of alumina intact Thesamples were rinsed in distilled water and again anodizedunder the same conditions used for the rst anodizationstep Estimates of pore size distribution were determinedfrom FE-SEM images and are shown in Figure 9andashcas a size distribution histograph Samples SO50 SO30and DO15 have respectively average pore diameters of

0

10

20

30

375 40 425 45 475 50

o

f po

res

Pore diameter (nm)

0

10

20

30

25 325 375 425 475 525

o

f po

res

0

5

10

15

20

25

175 20 225 25 275 30 325 375

DO15

o

f po

res

(a)

Pore diameter (nm)

(c)

Pore diameter (nm)

(b)

Fig 9 Pore size distribution of the tested sensors (a) SO50 (b) SO30(c) DO15

Fig 10 Digital image of an impedance-based interdigital nanoporousalumina humidity sensor

452 nm 384 nm and 136 nm with a correspondingstandard deviation of 34 nm 78 nm and 49 nm Allsamples have a predominantly ordered pore structure

Immediately after anodization all samples were thor-oughly washed in distilled water and then dried in a nitro-gen atmosphere at 100 C An interdigital capacitor pat-tern with dimensions of 12 cm pound 12 cm was formed onthe surface of the alumina lms by the evaporation of gold(thickness ordm 250 nm) through a mask A digital cameraimage of a typical sensor is shown in Figure 10 To createthe necessary ambient humidity argon passed through amass ow controller (MFC) was bubbled through a bot-tle containing deionized water and then mixed with dryargon coming from another MFC in appropriate ratiosbefore being passed to the test chamber A constant total ow was maintained throughout the experiment Cham-ber humidity was monitored with the humidity probeof a Keithley 6517A electrometer and all measurementswere carried out at room temperature (23 C) Sensorimpedance was measured over the frequency range 5 Hzto 13 MHz with a computer-controlled Hewlett Packardimpedance analyzer (4192A) tted with an impedance test xture (Agilent 16034E) A signal amplitude of 90 mVwas used for all measurements Electrical contact wasmade between the sensor under test and the impedanceanalyzer test xture with 42-gauge 5-cm-long jumpersattached to the interdigital electrodes with silver pasteThe contacts were subsequently annealed at 100 C for2 h to cure the silver paste

Figure 11 shows the measured 5-kHz sensor impedanceof the different sensors as a function of humidity Thehumidity-sensitive region of DO15 is sup145 to 95 RHSensors SO30 and SO50 become sensitive to humidity atapproximately 65 and 75 RH respectively A smallerpore size increases the range of humidity values overwhich the sensor is responsive and increases the sensitiv-ity at lower humidity levels Therefore an optimal poresize can be selected for the operating region of interest

283

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

103

104

105

106

0 20 40 60 80 100

Z (SO30)

Z(SO50)

Z(DO15)

| Z| (

W)

RH ()

Fig 11 Measured 5-kHz impedance of sensors over different humiditylevels

Figure 12 shows the response of the different sensorsto humidity at different measurement frequencies It canbe seen from the gure that as the frequency decreasesthe width of the sensitive region increases and the point atwhich the sensor becomes responsive shifts to lower rela-tive humidity values For a given pore size the humidity-sensitive region of operation can be selected by frequencytuning

The response time de ned as the time needed to reach90 of the nal signal for a given relative humidityand the recovery time de ned as the time taken forthe signal to come to within 10 of the initial valuewere determined by alternately exposing the sensors toa 2 and a 45 RH ambient with impedance measuredat 5 kHz A typical responserecovery graph of sensorSO50 is shown in Figure 13 All sensors were com-pletely reversible regaining their original impedance val-ues even after repeated exposure to high humidity levelsThe response time and recovery times corresponding todifferent pore diameters are shown in Figure 14 The min-imum responserecovery times were observed for SO50the sample with the largest pore size

The impedance spectra of the sensors were taken overa range of humidity levels The total impedance wasresolved into real (Z0) and imaginary (Z00) parts and Cole-Cole impedance plots were constructed Figure 15andashcshows respectively the Cole-Cole plot of sensors SO50SO30 and DO15 The equivalent circuit models shown inFigure 16 were used to t the experimental data The t-ting was accomplished with the complex nonlinear least-square- tting program101 The dots in the plots repre-sent experimental data and lines represent the equiva-lent circuit model t The equivalent circuit denoted (a)or (b) with reference to Figure 16 used for tting isdenoted near the corresponding curves in Figure 15 Theequivalent circuit models are different from the empiri-cal model suggested by Falk et al102 for porous aluminasensors fabricated by anodization of aluminum thin lmsIn the equivalent circuit models R1 and R2 represent twofrequency-independent resistors in parallel with disper-sive frequency-dependent capacitors Cn14mdash5 and Cn24mdash5respectively103 These non-Debye capacitances can be

103

104

105

106

107

SO50

5kHz10kHz100kHz1MHz

0 20 40 60 80 100

0 20 40 60 80 100

0 20 40 60 80 100

| Z| (

W)

| Z| (

W)

| Z| (

W)

RH ()(a)

(c)

(b)

103

104

105

106

107

SO30

5kHz10kHz100kHz1MHz

RH ()

102

103

104

105

106

DO15

5kHz10kHz100kHz1MHz

RH ()

Fig 12 Variation of sensor impedance with humidity and measure-ment frequency (a) SO50 (b) SO30 (c) DO15

68

7

72

74

76

78

0 200 400 600 800

Time (Sec)

45 RH

2 RH

45 RH 45 RH

2 RH

| Z| (

x105

W)

Fig 13 Response of sensor SO50 upon repeated cycling between rel-ative humidity environments of 2 and 45

284

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

10

15

20

25

30

35

40

45

50

20 25 30 35 40 45 50

Res

pons

eR

ecov

ery

Tim

e (S

ec)

Pore diameter (nm)

DO15

SO30

SO50

Fig 14 Dependence of response time and recovery time on pore size

expressed in a general form as Cn4mdash5 D Bn4imdash5nƒ1 whereBn is a constant for a given set of experimental conditionsand n takes a value between 0 and 1 according to the localmicroscopic environment through which charge transporttakes place751 1021 103 Bn behaves as an ideal frequency-independent capacitor as n approaches 1 and as an idealconductor when the value of n approaches 0101

By examination of Figure 15a SO50 it can be seenthat at low humidity levels the Cole-Cole plot describesan arc with a very large radius of curvature As the humid-ity increases the radius decreases and at an RH of about80 the complete semicircle comes inside the measur-able impedance range However at the same humiditylevel a spur appears in the low-frequency region that dis-torts the semicircle The spur which can be viewed aspart of a second semicircle dominates at higher humid-ity levels while the semicircle described by the high-frequency impedance values diminishes in size Thisbehavior is similar for all sensors For sensors SO50SO30 and DO15 respectively the RH at which thehigh-frequency semicircle comes completely within themeasurable impedance range is ordm75 65 and 35these values roughly match the in ection points seen inFigure 11

The fundamental mechanism that enables ceramic sen-sors to sense humidity is the physisorption of watermolecules on an initially chemisorbed layer of hydroxylions651 96 The chemisorbed hydroxyl ions enhance theelectrical conductivity of the sensor either by donatingelectrons to the conduction band of the base materialor through proton hopping between adjacent hydroxylgroups upon the application of an electric eld Theprocess of chemisorption occurs at very low humiditylevels and is unaffected by further changes in humid-ity However an increase in humidity makes the watermolecules physisorb to this hydroxyl layer The effec-tiveness of physisorption depends upon the cation chargecomplexes (from the material or the impurities) and thehydroxyl ions (water molecules present at the surface ofthe base material)96 The physisorption process is facili-tated by higher surface charge densities During formationof the rst physisorbed layer a water molecule becomesattached to two neighboring hydroxyl groups through

0

02

04

06

08

1

12

0 02 04 06 08 1 12

Zrsquo (MW)

35 (a) 55 (a)

65 (a)

75 (b)

81 (b)

93 (b) SO50

0

02

04

06

08

1

12

14

0 02 04 06 08 1 12 14

Zrsquo (W)

Z

(W)

Z

(MW

)Z

(M

W)

35 (a)

55 (a)

65 (b)

75 (b)

81 (b)

93 (b)

SO30

0

2 105

4 105

6 105

8 105

1 106

12 106

0 2 105 4 105 6 105 8 105 1 106 12 106

17 (a)27 (a)

35 (b)45 (b)55 (b)

93 (b)

DO15

(a)

Zrsquo (MW)

(b)

(c)

Fig 15 Cole-Cole plot of the sensors at different humidity levels(a) SO50 (b) SO30 (c) DO15 The points represent the experimentaldata and lines represent the t with the equivalent circuit model givenin Figure 9 The RH values at which the data were collected and theequivalent circuit used for tting each curve (given as (a) or (b) denot-ing the circuit of Figure 17a and b respectively) are shown adjacent toeach curve

hydrogen double bonds and a proton may be transferredfrom a hydroxyl group to the water molecule to form aH3OC ion

When the physisorption occurs within less than a monolayer that is when clusters of physisorbed molecules

285

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

(b)

(a)

Cn1 (w)

R1

Cn2 (w) Cn1 (w)

R1 R2

Fig 16 Equivalent circuit used for tting the experimental data

are present at the surface it is thought that H3OC diffu-sion within hydroxyl groups and proton transfer betweenadjacent water molecules clusters take place96 HoweverKhanna and Nahar104ndash106 refute this theory They arguethat the proton diffusion is less probable in a surfacethat contains hydroxyl ions and clusters of physisorbedmolecules and hence cannot account for the conductiv-ity of porous alumina at an RH less than 40 that isbefore the rst monolayer is formed Instead they sug-gested a phonon-assisted electron tunneling mechanism atlow humidity levels in which electron tunneling occursfrom one water molecule to another as enhanced by thesurface anions of the alumina sensors

At higher humidity levels the number of physisorbedlayers increases allowing each water molecule to besingly bonded to a hydroxyl group and proton hop-ping between adjacent water molecules in the continu-ous water layer takes place The conduction process isthe same as that of pure water and is called the Grot-thuss chain reaction96 The dominant charge carrier ina high-humidity environment is therefore HC ions (pro-tons) The concentration of HC increases with increas-ing humidity and HC move freely through the water-likelayer According to the theory of Khanna and Nahar104

in porous alumina the protons for conduction are donatednot only by H2O but also by the Al(OH)3 formed asa result of water adsorption and the impurity anionsincorporated into the lm from the electrolyte duringanodization107ndash109 The water-like network consists ofsingly bonded water molecules which possess a highdielectric constant as they form dipoles and reorient freelyunder an externally applied electric eld The impedanceof the nanoporous alumina sensors includes this capaci-tive contribution as well

From the analysis of Figure 15 the high-frequencysemicircle that diminishes in size with increasing RHarises because of the effect of a physisorbed water layerat the surface Therefore R1 in the equivalent circuitof Figure 16 represents the charge transfer resistancethrough the adsorbed water layer The effect of electronic

102

104

106

108

SO50SO30

DO15

R1

(W)

0 20 40 60 80 100

RH ()

Fig 17 Humidity dependence of resistance R1 of different sensorsobtained by tting the equivalent circuit model with experimental data

conduction through the alumina bulk is parallel to thatof the surface conduction affected by the adsorbed waterlayer This bulk conduction process is dominant only atextremely low humidity values when there are only afew adsorbed molecules Hence the incomplete semicir-cle or arc in the low-humidity region is partially deter-mined from the base alumina Bn1 represents the capac-itance contribution from the adsorbed water layer Thelarge spurs seen in the low-frequency region are a con-sequence of ion migration in the adsorbed layer towardelectrodes1011 103 This migration leads to accumulation ofions at the electrodes Since the relaxation time of ionmigration is larger than the charge transfer process itis seen only at low frequencies The accumulated ionsin the adsorbed water layer give rise to R2 and Bn2From the equivalent circuits given in Figure 16 it canbe seen that the presence of ion accumulation repre-sented by the combination of R2 and Cn24mdash5 increasesthe overall impedance of the sensor Hence the reductionin impedance caused at high humidity due to the chargetransfer process is countered by ion accumulation

To better understand the fundamental sensing mecha-nisms R1 and Bn1 obtained through the numerical ttingprogram for the different sensors are plotted in Figures 17and 18 respectively It can be seen from Figure 17 thatR1 which is the resistance to charge transport in a staticelectric eld by the material containing an adsorbed waterlayer in the absence of ion migration is linear with RH

-11

10-10

10-9

10-8

10-7

50V30VDO15

Bn1

0 20 40 60 80 100

RH ()

Fig 18 Humidity dependence of Bn1 of different sensors obtained by tting the equivalent circuit model (Fig 17) with experimental data

286

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

in a semilog plot For sensor DO15 R1 decreases to afew hundred ohms at high RH levels indicating that theobserved conductivity is not due merely to the protonsprovided by the adsorbed water layer Since deionizedwater and argon gas are used to create the required humid-ity environment it is reasonable to assume that the sur-faces of the alumina lms contain anions presumablyfrom the electrolyte used for anodization which controlsconduction through the adsorbed layer108

With reference to Figure 18 Bn1 of the larger pore sam-ples SO50 and SO30 increases signi cantly only at thehigher humidity levels This result is in contrast to thatof the smaller pore size sensors which shows that Bn1

increases at lower humidity levels The relative humid-ity values where a signi cant increase in Bn1 takes placegenerally match the point seen in Figure 11 where theimpedance drops signi cantly which in turn is a func-tion of pore size These same relative humidity valuescorrelate with the point at which the spur appears inthe impedance diagrams given in Figure 15 The resultslead to the following conclusions Since hydrogen-bondedwater molecules possess a high dielectric constant theincrease in Bn1 indicates the presence of these bondson the surface which means that a liquid-like networkmore than one monolayer of physisorbed molecules hasformed at that humidity level This liquid-like network isresponsible for the sharp reduction in impedance seen inFigure 11 and the ion migration that creates the spur inthe low-frequency portion of the Cole-Cole plots as seenin Figure 15 According to Eq (1) capillary condensationdoes not take place in the alumina test sensors

In light of the above observations we propose thefollowing model to explain the behavior of nanoporousalumina in response to humidity The anodization pro-cess results in a certain amount of electrolyte anionstrapped within the pores1041 1071 108 Even in a low-humidityregime the presence of these anions on the sensor sur-face provides a high charge density for easy physisorp-tion of water molecules The result is the formationof a liquid-like network within the pores The impu-rity anions also act as proton donors by a dissociativemechanism104 joining the protons donated by the watermolecules The smaller the pore size the lower the RHlevel at which the liquid-like network is formed Hencefor the small pore sensors there is variation in Bn1 aswell as impedance magnitude in low-humidity regimesthe exponential dependence of R1 on RH also suggeststhe same phenomenon

In the case of DO15 the rst semicircle is highly dis-torted by the low-frequency spur originating with ionmigration A relatively large number of mobile ions mightbe present in these sensors leading to an enhancedaccumulation of protons at the electrode-sample contactAs mentioned earlier the dipoles in the singly bondedadsorbed water layer are free to rotate in an electric eld

Hence in the region close to the electrode-alumina inter-face the adsorbed water layer exhibits a net dipolar ori-entation and therefore it does not display its normaldielectric constant1011 103 Depending on ion size up toa monolayer of charge can exist at the electrode101 thediffuse (Gouy) layer of counter-ions spreads away fromthis inner (Stern) layer The number of ions available toform the Stern and Gouy layers at a particular frequencyincreases with increasing number of physisorbed layersHence as the number of physisorbed layers increases theStern and Gouy layers start forming at higher frequencies

With reference to Figure 12 the relative humiditylevel at which the sharp decrease in impedance occursshifts higher with increasing frequency This result can beexplained with the equivalent circuit shown in Figure 16athat is the circuit in the absence of ion accumulation Athigh frequencies the reactive impedance Xc of Cn14mdash5equal to Xc

D 1=jmdashCn14mdash5 is low and hence impedanceis largely determined by R1 As the humidity increasesR1 decreases however this translates into a reduction inimpedance only at relatively low frequencies where thevalue of R1 is comparable to or less than the capacitivereactance of Cn1(mdash) As the measurement frequency isreduced the reactance of Cn1(mdash) decreases hence the effectof R1 becomes more signi cant at lower humidity levels

32 Hydrogen Sensing with Titania Nanotubes

We recently reported62 the fabrication of self-organizedtitania nanotube arrays with an anodization techniqueAlthough the as-prepared nanotubes are amorphous theycrystallize on annealing at elevated temperatures and arestructurally stable to at least 600 C This stability ofstructure which is an essential criterion of a gas-sensingmaterial prompted us to study the gas-sensing behavior ofthese nanotubes for technologically important gases suchas oxygen carbon monoxide ammonia carbon dioxideand hydrogen

Titania has earned much attention for its oxygen-sensing capabilities110ndash115 Furthermore with propermanipulation of the microstructure or crystalline phaseandor the addition of the proper impurities or surfacefunctionalization titania can also be used as a reducinggas sensor116ndash127 The interaction of a gas with a metaloxide semiconductor is primarily a surface phenomenonand nanoporous metal oxides1221 1231 1281 129 offer the advan-tage of providing large sensing surface areas

Room-temperature metal oxide hydrogen sensors aregenerally based on Schottky barrier modulation of deviceslike PdTiO2 or PtTiO2 by hydrogen130ndash132 Elevated-temperature hydrogen sensors examine electrical resis-tance with hydrogen concentration for example Birke-feld et al133 observed that the resistance of the anatasephase of titania varies in the presence of carbon monoxideand hydrogen at temperatures above 500 C but on dop-ing with 10 alumina it becomes selective for hydrogen

287

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Barrier layer

Titanium Nano tube array

Platinum electrodes

Fig 19 Schematic representation of the hydrogen sensor geometry

Munuera et al134 used rhenium doping in the anatasephase to perform room-temperature hydrogen monitor-ing Recently Shimizu et al122 reported that anodizednanoporous titania lms with a Pd Schottky barrier aresensitive to hydrogen at 250 C

Titania nanotubes were grown from titanium foil(995 pure from Alfa Aesar Ward Hill MA) with athickness of 025 mm The anodization was performed inan electrolyte medium of 05 hydro uoric acid in waterwith the use of a platinum foil cathode Well-de nednanotube arrays were grown with anodizing potentialsranging from 12 V to 20 V Nanotube length increaseswith anodization time reaching 400 nm in approximately20 min and then remains constant For the present studythe samples were anodized for 25 min The samples werethen annealed at 500 C in a pure oxygen ambient for 6 hwith a heating and cooling rate of 1 Cmin

The electrode geometry of the titania nanotube sen-sors is shown in Figure 19 The sensor consists of a basetitanium metal foil with a nanotube array grown on topAn insulating barrier layer separates the nanotubes fromthe conducting titanium foil Electrical connections weremade by two parallel 10 mm pound 2 mm platinum pads of100-Œm thickness The test chamber consists of a 13-literquartz tube with stainless steel end caps placed insidea furnace (Thermolyne USA model 21100 tubular fur-nace) Gas ow through the test chamber was controlledvia a computer-controlled mass ow controller The elec-trical resistance of the titania sensors was measured witha computer-controlled Keithly digital multimeter Prior todata collection the test chamber was initially evacuatedwith a mechanical pump whereupon nitrogen (99999pure) was passed while the sensor under test was heatedto the desired temperature The test gases were mixed inappropriate ratios with nitrogen to create the necessarytest gas ambient

The surface morphology of nanotube arrays preparedwith an anodization potential of 20 V and annealed at500 C for 6 h in a pure oxygen ambient is shown inFigure 20a and b It can be seen from these images thatthe nanotubes are uniform over the surface The nano-tubes are approximately 400 nm in length and have abarrier layer62 thickness of ordm50 nm For the nanotubes

100 nm

(a)

(b)

1 mm (c)

100 nm Fig 20 The surface morphology of the titania nanotubes after anneal-ing at 500 C (a) High- and (b) low-magni cation images of a 20-Vsample (c) A high-magni cation image of a 12-V sample

fabricated with 20-V anodization the average pore diam-eter as determined from FESEM images is 76 nm (thestandard deviation 15 nm) with a wall thickness of 27 nm(standard deviation 6 nm) The sample anodized at 12 Vshown in Figure 2c was found to have an average porediameter of 46 nm (standard deviation 8 nm) with a wallthickness of 17 nm (standard deviation 2 nm) The porosi-ties of the 20-V and 12-V samples were calculated as 45and 61 respectively

Glancing-angle X-ray diffraction patterns of a 20-Vsample annealed at 500 C for 6 h in oxygen ambientis shown in Figure 21 It can be seen that both anataseand rutile phases of titania are present in the sampleA detailed study135 of these structures by high-resolutiontransmission electron microscope (HRTEM) showed that

288

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

20 30 40 50 60 70 80

2 q (deg)

(101

)

(110

)

A

R

TR

T

T

R R A TRR T R R T TT

Fig 21 Glancing-angle X-ray diffraction pattern of a 20-V sample(glancing angle D 2) annealed at 500 C in oxygen ambient A R andT represent respectively the re ections from anatase crystallites rutilecrystallites and the titanium substrate

the anatase crystallites were concentrated on the walls ofthe nanotubes and rutile on the barrier layer

Nanotubes annealed in a pure oxygen ambient werefound to be stable (intact) to temperatures of approxi-mately 580 C Above this temperature protrusions wereseen to be coming out through the nanotubes an effectthat spread with increasing temperature These protru-sions which are due to oxidation of the titanium substratecollapse the nanotubes

Figure 22 shows the response of the 20-V (76 nm) sam-ple as a function of ambient temperature as it is switchedfrom a nitrogen environment to one containing 1000 ppmhydrogen and then back to nitrogen The plot was madeusing 4Rg=R05

ƒ1 versus time where R0 is the base resis-tance of the sensor that is the sensor resistance before thetest gas was introduced and Rg is the measured resistancein the presence of test gas The sensor shows increasinghydrogen sensitivity with temperature with a change by3 orders of magnitude in resistance at temperatures aboveordm300 C At all of the temperatures the original resistancerecovered without hysteresis

The sensitivity S is de ned by the formula

S DR0

ƒ Rgs

Rgs

10 0

10 1

10 2

10 3

0 10 0 5 10 2 1 10 3 15 103 2 10 3 25 103 3 10 3 35 103 4 10 3

Time (Sec)

(Rg

R0)

-1

180degC

240degC

265degC

290degC

345degC

375degC

H2+N

2

N2

Fig 22 Variation of resistance Rg normalized with respect to baselineresistance R0 of a 20-V sample with time on exposure to 1000 ppmhydrogen at different temperatures It may be noted that the inverse ofRg=R0 was used in the plot to represent data in the positive y direction

10-1

100

101

102

103

150 200 250 300 350 400 450

Temperature (degC)

Sens

itivi

ty (

R0-

Rgs

)R

gs

Fig 23 The sensitivity temperature dependence of a 20-V sample to1000 ppm hydrogen

where R0 is the resistance of the sensor before the gas ispassed and Rgs is the resistance after gas is passed andreaches the saturation value The temperature-dependentsensitivity of a 20-V sample to 1000 ppm hydrogen isshown in Figure 23 Sensitivity is seen to increase withtemperature to approximately 380 C where the increasein sensitivity with temperature begins to saturate

The response time de ned as the time needed for thesensor to reach 90 of the nal signal for a given concen-tration of gas is plotted against temperature in Figure 24(the time includes that required for the gas to equili-brate inside the measurement chamber estimated to beordm30 s) The response time decreases exponentially withtemperature

To check the behavior of the sensor on repeated hydro-gen exposure the hydrogen concentration was varied indiscrete steps of 100 ppm from 0 to 500 ppm while thetemperature was kept constant at 290 C the chamber was ushed with nitrogen after each exposure to hydrogenThe response of the 20-V prepared nanotube sensorskept at 290 C is shown in Figure 25 The behaviorof the sensor is consistent recovering its original resis-tance after repeated exposure to varying hydrogen gasconcentrations The sensitivities of the 76-nm- 53-nm-and 22-nm-diameter nanotube sensors to hydrogen con-centrations ranging from 100 ppm to 1 are shown inFigure 26 The sensitivity of the 76-nm (20-V) sensor tolow concentrations is shown in Figure 27 there is a linearincrease in sensitivity at low concentrations The lower

10 1

10 2

10 3

10 4

150 200 250 300 350 400 450

Temperature (degC)

Res

pons

e T

ime

(Sec

)

Fig 24 Variation in response time for a 20-V sensor as a function oftemperature The dots represent measured data

289

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

10 2

0 4 4 4 4 4

10 3

0 10 1 10 2 10 3 10 4 10 5 10

Time (Sec)

Res

ista

nce

(W)

H2 (ppm)

200

300

400

500

400

300

200

100

100

Fig 25 Resistance of a 20-V sample when exposed to different con-centrations of hydrogen at 290 C The nitrogen-hydrogen mixture waspassed for 1500 s the chamber was then ushed with nitrogen for 3000 sbefore the nitrogen-hydrogen mixture was passed again

test limit of 20 ppm is determined by the experimentalapparatus and not the sensor

Figure 28 shows the change in resistance of the 76-nm-46-nm- and 22-nm-diameter nanotube sensors with expo-sure to 1000 ppm hydrogen at 290 C Whereas smallerdiameter nanotubes had greater sensitivity to hydrogenthe samples made at lower anodizing voltages tended tobe more brittle and harder to handle mechanically withoutbreaking

The 20-V 76-nm-diameter sample was exposed to oxy-gen carbon monoxide ammonia and carbon dioxide at290 C to study the cross-sensitivity The sensor wasfound to have no detectable variation in resistance onexposure to carbon dioxide The sensitivities of the tita-nia nanotubes to the other gases are shown in Figure 29The sensitivities of the nanotubes to carbon monoxide andammonia are negligible compared with that of hydrogenThe resistance of the nanotubes increased in the presenceof oxygen and did not regain their original electrical con-ductivity even after several hours in a 290 C nitrogenenvironment

1

10

100

1000

104

105

0 02 04 06 08 1

Hydrogen Concentration ()

Pore diameter = 76 nm

Pore diameter = 53 nm

Pore diameter = 22nm

Sen

sitiv

ity (

R0-

Rg)

Rg

Fig 26 The sensitivity variation of a samples with pore diameters of22 nm 53 nm and 76 nm versus time and hydrogen concentrations

0 100 200 300 400 5001

10

100

Hydrogen concentration (ppm)

Sen

sitiv

ity

(R0-

Rgs

)R

gs

Fig 27 Sensitivity of a 20-V (76 nm) sample at 290 C for low hydro-gen concentrations

Since the sensor measurements were made in atmo-spheres without oxygen the increase in conductivitycannot be due to hydrogen removing oxygen from thelattice132ndash134 or the removal of chemisorbed oxygen135ndash137

Hydrogen molecules can dissociate at defects on thetitania surface which can subsequently diffuse into thetitania lattice acting as electron donors1281 1381 139 How-ever a dissociation-driven process would result in aslow response and recovery times with complete recov-ery unlikely since the sensor completely regains its orig-inal resistance with hydrogen cycling it appears that thisis not the dominant mechanism behind high hydrogensensitivity We believe chemisorption of the dissociatedhydrogen on the titania surface is the underlying sensingmechanism137 During chemisorption hydrogen acts as asurface state and a partial charge transfer from hydrogento the titania conduction band takes place This createsan electron accumulation layer on the nanotube surfaceenhancing its electrical conductance Upon removal of thehydrogen ambient electron transfer back to the hydro-gen molecule takes place which subsequently desorbsrestoring the original electrical resistance of the mate-rial Another factor that may play a role in the hydrogensensitivity (and selectivity) is the platinum electrodes Atelevated temperatures hydrogen dissociation can occur onplatinum surfaces These dissociated hydrogen atoms mayspill1361 140 onto the nanotube surface where they diffuse

10-1

101

103

105

0 05 1 15

Time (Hour)

Pore size - 76 nm

Pore size - 53 nm

Pore size - 22 nm

(Rg

R0)

-1

Fig 28 A comparison of the variation in resistance of samples havingpore diameters of 22 nm 53 nm and 76 nm versus time upon exposureto 1000 ppm of hydrogen at 290 C

290

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

10-2

10-1

100

0 100 2 102 4 102 6 102 8 102 1 103

OxygenAmmoniaCarbon monoxide

Time (Sec)

(Rg

R0)

-1

Fig 29 Variation of resistance of a 20-V sample when exposed to1000 ppm carbon monoxide 5 ammonia and 1 oxygen at 290 C

into the material affecting its electrical properties Fromthe present study it was not clear how signi cant a rolethe platinum electrodes play

Anatase the polymorph of titania has been reported tohave high sensitivity for reducing gases like hydrogen andcarbon monoxide1211 1241 133 Our nanotube samples containanatase phase mainly on the walls and rutile in the barrierlayer As the diffusing hydrogen atoms go to the intersti-tial sites1331 142 and as the ca ratio of anatase is almostfour times that of rutile it appears that the anatase latticeaccommodates hydrogen easily and hence makes a highercontribution to hydrogen sensitivity

The effect of chemisorption can be neglected in theoxygen-sensing experiments As the recovery requiresseveral hours it appears that the nanotubes contain oxy-gen vacancies or titanium interstitial defects in the pres-ence of nitrogen On exposure of the sensor to oxygenambient the lattice reoxidizes and hence the conductiv-ity of the sensor decreases On the removal of oxygenthe reduction of the lattice will not immediately occurhence the sensor requires several hours to regain its orig-inal conductivity

4 CONCLUSIONS

This paper summarizes different routes of fabrication ofmetal oxide nanoarchitectures useful for sensing applica-tions Structures made via anodization of a starting metaland breath gure formation from a sol-gel are discussedand presented as are TiO2 nanowires made by lling ananoporous alumina membrane with subsequent removalof the supporting template Two test cases illustratingthe utility of these nanoporous metal oxide structures forsensing are presented highly ordered nanoporous aluminais used for humidity sensing and TiO2 nanotubes are usedfor hydrogen sensing These study test cases reveal sev-eral points worth brie y noting

In the rst sensing test case uniform nanoporous alu-mina sensors exhibit pore size and frequency-dependent

humidity-sensing characteristics The mechanism respon-sible for the sensing action of the nanoporous alumina sen-sors was studied by representing the impedance spectra atdifferent humidity levels in Cole-Cole plots This makes itpossible to identify the competing impedance-determiningprocesses occurring in the sensors namely charge transferand mass transport through the physisorbed water layeras dependent upon the humidity level and input signalfrequency Charge transfer was found to enhance the sen-sitivity by causing a large variation in impedance withrespect to humidity Mass transport leads to accumulationof ions at the electrodes which tends to decrease the mea-surement sensitivity by increasing the impedance Hencea proper choice of operating frequency high enough toavoid ion accumulation is essential for optimum sensorperformance

It has been inferred from our results that anions becomeincorporated into the alumina lm during the anodiza-tion process supporting the hypothesis of Khanna andNahar104ndash106 where they act as proton donors and facili-tate the adsorption of water molecules Hence liquid-likecharge transfer networks are formed at relatively lowerhumidity levels for the smaller pore-size sensors Thisprocess rather than capillary condensation is primarilyresponsible for the observed humidity-sensing behavior ofthe sensors Our studies show that a distribution of poresizes as suggested by Shimizu et al94 is not necessary toobtain sensitive performance over a wide humidity rangerather such optimal performance can be obtained with uni-form nanodimensional pores containing adsorbed anionicimpurities

In the second sensing test case titania nanotubes pre-pared by anodization and annealed in an oxygen atmo-sphere at a temperature of 500 C were found to behighly sensitive to hydrogen The nanotube sensors con-tained both anatase and rutile phases of titania and showedappreciable sensitivity toward hydrogen at temperaturesas low as 180 C The sensitivity increased drasticallywith temperature showing a variation of 3 orders in mag-nitude of resistance to 1000 ppm of hydrogen at 400 CThe response time decreased with increasing temperatureat 290 C full switching of the sensor took approximately3 min Results were highly reproducible with no indica-tion of hysteresis Our results showed that these sensorsare at least capable of monitoring hydrogen levels from20 ppm to 4 At 290 C nanotubes with smaller porediameters showed higher sensitivity to hydrogen com-pared with their larger pore counterparts The sensorsshowed high selectivity to hydrogen compared with car-bon monoxide ammonia and carbon dioxide Althoughthe sensor was sensitive to high concentrations of oxy-gen the response time was high and the sensor did notcompletely regain the original condition We believe thehydrogen sensitivity of the nanotubes is due to hydrogendiffusion into the titania lattice where they act as electrondonors

291

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Acknowledgments Support of this work by the NationalScience Foundation through grant ECS-0210033 is grate-fully acknowledged

References and Notes

1 K Kajihara K Tanaka K Horao and N Soga Jpn J ApplPhys 35 6110 (1996)

2 B O Regan and M Gratzel A low cost Nature 353 737 (1991)3 A Fujishima and K Honda Nature 238 37 (1972)4 K Sato A Tsuzuki H Taoda Y Torii T Kato and Y Butsugan

J Mater Sci 29 5911 (1994)5 K Sato A Tsuzuki Y Torii H Taoda T Kato and Y Butsugan

J Mater Sci 30 837 (1995)6 N Ozer and C M Lampert Sol Energy Mater Sol Cells 54 147

(1998)7 A Manz and H Becker editors Microsystem Technology in

Chemistry and Life Sciences Springer-Verlag Berlin 19988 E Kim Y Xia and G M Whitesides Nature 376 581 (1995)9 M Templin A Franck A Du Chesne A Leist A Zhang

R Ulrich V Schadler and U Weisner Science 278 1795 (1997)10 D Zhao J Feng Q Huo N Melosh G H Frederickson

B Chmelka and G D Stucky Science 279 548 (1998)11 O D Velev T A Jede R F Lobo and A M Lenhoff Nature

389 447 (1997)12 B T Holland C F Blanford and A Stein Science 281 538

(1998)13 S Matsushita T Miwa and A Fujishima Chem Lett 9 925

(1997)14 A Imhof and D J Pine Adv Mater 10 697 (1998)15 T Tatsuma A Ikezawa Y Ohko T Miwa T Matsue and

A Fujishima Adv Mater 12 643 (2000)16 Z Zhong Y Yin B Gates and Y Xia Adv Mater 12 206 (2000)17 O D Velev and A M Lenhoff Curr Opin Colloids Interface

Sci 5 56 (2000)18 O D Velev and E W Kaler Adv Mater 12 531 (2000)19 P Jiang J Cizeron J F Bertone and V L Colvin J Am Chem

Soc 121 7957 (1999)20 M E Turner T J Trentler and V L Colvin Adv Mater 13 180

(2001)21 K M Kulinowski P Jiang H Vaswani and V L Colvin Adv

Mater 12 833 (2000)22 P Jiang J F Bertone and V L Colvin Science 291 453 (2001)23 R Seshadri and F C Meldrum Adv Mater 12 1149 (2000)24 F C Meldrum and R Seshadri Chem Commun 29 (2000)25 Q B Meng Z Z Gu O Sato and A Fujishima Appl Phys

Lett 77 4313 (2000)26 G Widawski B Rawiso and B Francois Nature 369 3897

(1994)27 B Francois O Pitois and J Francois Adv Mater 7 1041 (1995)28 O Pitois and B Francois Eur Phys J B 8 225 (1999)29 S A Jenekhe and X L Chen Science 283 372 (1999)30 M Shimomuara and T Sawadaishi Curr Opin Colloid Interface

Sci 6 11 (2001)31 O Karthaus N Maruyama X Cieren M Shimomura

H Hasegawa and T Hashimoto Langmuir 16 6071 (2000)32 T Nishikawa J Nishida R Ookura S Nishimura S Wada

T Karino and M Shimomura Mater Sci Eng C 10 141 (1999)33 K Kajihara K Nakanishi K Tanaka K Hirao and N Soga

J Am Ceram Soc 81 2670 (1998)34 C A Grimes R S Singh E C Dickey and O K Varghese

J Mater Res 16 1686 (2001)35 M Srinivasrao D Collings A Phillips and S Patel Science 292

79 (2001)36 P G de Gennes Rev Mod Phys 57 827 (1985)

37 D Beysens A Steyer P Guenoun D Fritter and C M KnoblerPhase Transitions 31 219 (1991)

38 B J Briscoe and K P Galvin J Phys D 23 422 (1990)39 B J Briscoe and K P Galvin J Phys D 23 1265 (1990)40 D Beysens and C M Knobler Phys Rev Lett 57 1433 (1986)41 F Family and P Meakin Phys Rev Lett 61 428 (1988)42 A V Limaye R D Narhe A M Dhote and S B Ogale Phys

Rev Lett 76 3762 (1996)43 P Dellrsquo Aversana and G P Neitzel Phys Today 51 38 (1998)44 P Dellrsquo Aversana J R Banavar and J Koplik Phys Fluids 8 15

(1986)45 P Dellrsquo Aversana V Tontodonato and L Carotenuto Phys Fluids

9 2475 (1997)46 M Q Li S Q Xu and E Kumacheva Langmuir 16 7275 (2000)47 P Freymuth Rev Sci Instrum 64 1 (1993)48 P Ball The Self-Made Tapestry Pattern Formation in Nature

Oxford University Press New York 199949 M F Schatz S J VanHook W D McCormick J B Swift and

H L Swinney Phys Fluids 11 2577 (1999)50 S R Coriell M R Cordes and W J Boettinger J Cryst Growth

49 13 (1980)51 T S Sullivan Y M Liu and R E Ecke Phys Rev E 54 486

(1996)52 J L Rogers M F Schatz J L Bougie and J B Swift 84 87

(2000)53 K Nitschke 52 R5772 (1995)54 E Bodenschatz J R De Bruyn G Ahlers and D S Cannell

Phys Rev Lett 67 3078 (1991)55 C W Meyer D S Cannell G Ahlers J B Swift and P C

Hohenberg Phys Rev Lett 61 947 (1988)56 R F Ismagilov D Rosmarin D H Gracias A D Stroock and

G M Whitesides Appl Phys Lett 79 439 (2001)57 E A Barringer and H K Bowen Langmuir 1 420 (1985)58 M Tschapek C Wasowski and R M T Sanchez J Electroanal

Chem 74 167 (1976)59 H Masuda and K Fukuda Science 268 1466 (1995)60 H Masuda F Hasegwa and S Ono J Electrochem Soc 144

L 127 (1997)61 O Jessensky F Muller and U Gosele Appl Phys Lett 72 1173

(1998)62 D Gong C A Grimes O K Varghese W Hu R S Singh

Z Chen and E C Dickey J Mater Res 16 3331 (2001)63 B M Kulwicki J Am Ceram Soc 74 697 (1991)64 H Arai and T Seiyama in Sensors A Comprehensive Survey

edited by W Gopel J Hesse and J N Zemel VCH Weinheim(1992) Vol 3 pp 981ndash1012

65 E Traversa Sens Actuators B 23 135 (1995)66 N Yamazoe and Y Shimizu Sens Actuators 10 379 (1986)67 A Bearzotti I Fratoddi L Palummo S Petrocco A Furlani

C Lo Sterzo and M V Russo Sens Actuators B 76 316 (2001)68 Y Sakai Y Sadaoka and M Matsuguchi Sens Actuators B 35

85 (1996)69 C A Grimes and D Kouzoudis Sens Actuators A 84 205

(2000)70 K G Ong C A Grimes C L Robbins and R S Singh Sens

Actuators A 93 33 (2001)71 T Seiyama N Yamazoe and H Arai Sens Actuators 4 85

(1983)72 L Ketron Ceram Bull 68 860 (1989)73 E Traversa G Gusmano and A Montenero Eur J Solid State

Inorg Chem 32 719 (1995)74 M K Jain M C Bhatnagar and G L Sharma Sens Actuators

B 55 180 (1999)75 O K Varghese and L K Malhotra J Appl Phys 87 7457 (2000)76 G Sberveglieri R Murri and N Pinto Sens Actuators B 23

177 (1995)

292

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

oxalic acid (O) in the anodization bath the two-digit num-ber denotes the anodization voltage It was observed thatsingle-step anodization results in disordered pore struc-tures at low voltages Hence a double-step process wasused to make sample DO15 These samples were initiallyanodized under the conditions given in Table I The sam-ples were then immersed in an aqueous solution contain-ing 18 wt chromic acid and 4 wt phosphoric acid for12 h to remove the initial nanoporous layer while keep-ing the nonporous bottom layer of alumina intact Thesamples were rinsed in distilled water and again anodizedunder the same conditions used for the rst anodizationstep Estimates of pore size distribution were determinedfrom FE-SEM images and are shown in Figure 9andashcas a size distribution histograph Samples SO50 SO30and DO15 have respectively average pore diameters of

0

10

20

30

375 40 425 45 475 50

o

f po

res

Pore diameter (nm)

0

10

20

30

25 325 375 425 475 525

o

f po

res

0

5

10

15

20

25

175 20 225 25 275 30 325 375

DO15

o

f po

res

(a)

Pore diameter (nm)

(c)

Pore diameter (nm)

(b)

Fig 9 Pore size distribution of the tested sensors (a) SO50 (b) SO30(c) DO15

Fig 10 Digital image of an impedance-based interdigital nanoporousalumina humidity sensor

452 nm 384 nm and 136 nm with a correspondingstandard deviation of 34 nm 78 nm and 49 nm Allsamples have a predominantly ordered pore structure

Immediately after anodization all samples were thor-oughly washed in distilled water and then dried in a nitro-gen atmosphere at 100 C An interdigital capacitor pat-tern with dimensions of 12 cm pound 12 cm was formed onthe surface of the alumina lms by the evaporation of gold(thickness ordm 250 nm) through a mask A digital cameraimage of a typical sensor is shown in Figure 10 To createthe necessary ambient humidity argon passed through amass ow controller (MFC) was bubbled through a bot-tle containing deionized water and then mixed with dryargon coming from another MFC in appropriate ratiosbefore being passed to the test chamber A constant total ow was maintained throughout the experiment Cham-ber humidity was monitored with the humidity probeof a Keithley 6517A electrometer and all measurementswere carried out at room temperature (23 C) Sensorimpedance was measured over the frequency range 5 Hzto 13 MHz with a computer-controlled Hewlett Packardimpedance analyzer (4192A) tted with an impedance test xture (Agilent 16034E) A signal amplitude of 90 mVwas used for all measurements Electrical contact wasmade between the sensor under test and the impedanceanalyzer test xture with 42-gauge 5-cm-long jumpersattached to the interdigital electrodes with silver pasteThe contacts were subsequently annealed at 100 C for2 h to cure the silver paste

Figure 11 shows the measured 5-kHz sensor impedanceof the different sensors as a function of humidity Thehumidity-sensitive region of DO15 is sup145 to 95 RHSensors SO30 and SO50 become sensitive to humidity atapproximately 65 and 75 RH respectively A smallerpore size increases the range of humidity values overwhich the sensor is responsive and increases the sensitiv-ity at lower humidity levels Therefore an optimal poresize can be selected for the operating region of interest

283

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

103

104

105

106

0 20 40 60 80 100

Z (SO30)

Z(SO50)

Z(DO15)

| Z| (

W)

RH ()

Fig 11 Measured 5-kHz impedance of sensors over different humiditylevels

Figure 12 shows the response of the different sensorsto humidity at different measurement frequencies It canbe seen from the gure that as the frequency decreasesthe width of the sensitive region increases and the point atwhich the sensor becomes responsive shifts to lower rela-tive humidity values For a given pore size the humidity-sensitive region of operation can be selected by frequencytuning

The response time de ned as the time needed to reach90 of the nal signal for a given relative humidityand the recovery time de ned as the time taken forthe signal to come to within 10 of the initial valuewere determined by alternately exposing the sensors toa 2 and a 45 RH ambient with impedance measuredat 5 kHz A typical responserecovery graph of sensorSO50 is shown in Figure 13 All sensors were com-pletely reversible regaining their original impedance val-ues even after repeated exposure to high humidity levelsThe response time and recovery times corresponding todifferent pore diameters are shown in Figure 14 The min-imum responserecovery times were observed for SO50the sample with the largest pore size

The impedance spectra of the sensors were taken overa range of humidity levels The total impedance wasresolved into real (Z0) and imaginary (Z00) parts and Cole-Cole impedance plots were constructed Figure 15andashcshows respectively the Cole-Cole plot of sensors SO50SO30 and DO15 The equivalent circuit models shown inFigure 16 were used to t the experimental data The t-ting was accomplished with the complex nonlinear least-square- tting program101 The dots in the plots repre-sent experimental data and lines represent the equiva-lent circuit model t The equivalent circuit denoted (a)or (b) with reference to Figure 16 used for tting isdenoted near the corresponding curves in Figure 15 Theequivalent circuit models are different from the empiri-cal model suggested by Falk et al102 for porous aluminasensors fabricated by anodization of aluminum thin lmsIn the equivalent circuit models R1 and R2 represent twofrequency-independent resistors in parallel with disper-sive frequency-dependent capacitors Cn14mdash5 and Cn24mdash5respectively103 These non-Debye capacitances can be

103

104

105

106

107

SO50

5kHz10kHz100kHz1MHz

0 20 40 60 80 100

0 20 40 60 80 100

0 20 40 60 80 100

| Z| (

W)

| Z| (

W)

| Z| (

W)

RH ()(a)

(c)

(b)

103

104

105

106

107

SO30

5kHz10kHz100kHz1MHz

RH ()

102

103

104

105

106

DO15

5kHz10kHz100kHz1MHz

RH ()

Fig 12 Variation of sensor impedance with humidity and measure-ment frequency (a) SO50 (b) SO30 (c) DO15

68

7

72

74

76

78

0 200 400 600 800

Time (Sec)

45 RH

2 RH

45 RH 45 RH

2 RH

| Z| (

x105

W)

Fig 13 Response of sensor SO50 upon repeated cycling between rel-ative humidity environments of 2 and 45

284

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

10

15

20

25

30

35

40

45

50

20 25 30 35 40 45 50

Res

pons

eR

ecov

ery

Tim

e (S

ec)

Pore diameter (nm)

DO15

SO30

SO50

Fig 14 Dependence of response time and recovery time on pore size

expressed in a general form as Cn4mdash5 D Bn4imdash5nƒ1 whereBn is a constant for a given set of experimental conditionsand n takes a value between 0 and 1 according to the localmicroscopic environment through which charge transporttakes place751 1021 103 Bn behaves as an ideal frequency-independent capacitor as n approaches 1 and as an idealconductor when the value of n approaches 0101

By examination of Figure 15a SO50 it can be seenthat at low humidity levels the Cole-Cole plot describesan arc with a very large radius of curvature As the humid-ity increases the radius decreases and at an RH of about80 the complete semicircle comes inside the measur-able impedance range However at the same humiditylevel a spur appears in the low-frequency region that dis-torts the semicircle The spur which can be viewed aspart of a second semicircle dominates at higher humid-ity levels while the semicircle described by the high-frequency impedance values diminishes in size Thisbehavior is similar for all sensors For sensors SO50SO30 and DO15 respectively the RH at which thehigh-frequency semicircle comes completely within themeasurable impedance range is ordm75 65 and 35these values roughly match the in ection points seen inFigure 11

The fundamental mechanism that enables ceramic sen-sors to sense humidity is the physisorption of watermolecules on an initially chemisorbed layer of hydroxylions651 96 The chemisorbed hydroxyl ions enhance theelectrical conductivity of the sensor either by donatingelectrons to the conduction band of the base materialor through proton hopping between adjacent hydroxylgroups upon the application of an electric eld Theprocess of chemisorption occurs at very low humiditylevels and is unaffected by further changes in humid-ity However an increase in humidity makes the watermolecules physisorb to this hydroxyl layer The effec-tiveness of physisorption depends upon the cation chargecomplexes (from the material or the impurities) and thehydroxyl ions (water molecules present at the surface ofthe base material)96 The physisorption process is facili-tated by higher surface charge densities During formationof the rst physisorbed layer a water molecule becomesattached to two neighboring hydroxyl groups through

0

02

04

06

08

1

12

0 02 04 06 08 1 12

Zrsquo (MW)

35 (a) 55 (a)

65 (a)

75 (b)

81 (b)

93 (b) SO50

0

02

04

06

08

1

12

14

0 02 04 06 08 1 12 14

Zrsquo (W)

Z

(W)

Z

(MW

)Z

(M

W)

35 (a)

55 (a)

65 (b)

75 (b)

81 (b)

93 (b)

SO30

0

2 105

4 105

6 105

8 105

1 106

12 106

0 2 105 4 105 6 105 8 105 1 106 12 106

17 (a)27 (a)

35 (b)45 (b)55 (b)

93 (b)

DO15

(a)

Zrsquo (MW)

(b)

(c)

Fig 15 Cole-Cole plot of the sensors at different humidity levels(a) SO50 (b) SO30 (c) DO15 The points represent the experimentaldata and lines represent the t with the equivalent circuit model givenin Figure 9 The RH values at which the data were collected and theequivalent circuit used for tting each curve (given as (a) or (b) denot-ing the circuit of Figure 17a and b respectively) are shown adjacent toeach curve

hydrogen double bonds and a proton may be transferredfrom a hydroxyl group to the water molecule to form aH3OC ion

When the physisorption occurs within less than a monolayer that is when clusters of physisorbed molecules

285

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

(b)

(a)

Cn1 (w)

R1

Cn2 (w) Cn1 (w)

R1 R2

Fig 16 Equivalent circuit used for tting the experimental data

are present at the surface it is thought that H3OC diffu-sion within hydroxyl groups and proton transfer betweenadjacent water molecules clusters take place96 HoweverKhanna and Nahar104ndash106 refute this theory They arguethat the proton diffusion is less probable in a surfacethat contains hydroxyl ions and clusters of physisorbedmolecules and hence cannot account for the conductiv-ity of porous alumina at an RH less than 40 that isbefore the rst monolayer is formed Instead they sug-gested a phonon-assisted electron tunneling mechanism atlow humidity levels in which electron tunneling occursfrom one water molecule to another as enhanced by thesurface anions of the alumina sensors

At higher humidity levels the number of physisorbedlayers increases allowing each water molecule to besingly bonded to a hydroxyl group and proton hop-ping between adjacent water molecules in the continu-ous water layer takes place The conduction process isthe same as that of pure water and is called the Grot-thuss chain reaction96 The dominant charge carrier ina high-humidity environment is therefore HC ions (pro-tons) The concentration of HC increases with increas-ing humidity and HC move freely through the water-likelayer According to the theory of Khanna and Nahar104

in porous alumina the protons for conduction are donatednot only by H2O but also by the Al(OH)3 formed asa result of water adsorption and the impurity anionsincorporated into the lm from the electrolyte duringanodization107ndash109 The water-like network consists ofsingly bonded water molecules which possess a highdielectric constant as they form dipoles and reorient freelyunder an externally applied electric eld The impedanceof the nanoporous alumina sensors includes this capaci-tive contribution as well

From the analysis of Figure 15 the high-frequencysemicircle that diminishes in size with increasing RHarises because of the effect of a physisorbed water layerat the surface Therefore R1 in the equivalent circuitof Figure 16 represents the charge transfer resistancethrough the adsorbed water layer The effect of electronic

102

104

106

108

SO50SO30

DO15

R1

(W)

0 20 40 60 80 100

RH ()

Fig 17 Humidity dependence of resistance R1 of different sensorsobtained by tting the equivalent circuit model with experimental data

conduction through the alumina bulk is parallel to thatof the surface conduction affected by the adsorbed waterlayer This bulk conduction process is dominant only atextremely low humidity values when there are only afew adsorbed molecules Hence the incomplete semicir-cle or arc in the low-humidity region is partially deter-mined from the base alumina Bn1 represents the capac-itance contribution from the adsorbed water layer Thelarge spurs seen in the low-frequency region are a con-sequence of ion migration in the adsorbed layer towardelectrodes1011 103 This migration leads to accumulation ofions at the electrodes Since the relaxation time of ionmigration is larger than the charge transfer process itis seen only at low frequencies The accumulated ionsin the adsorbed water layer give rise to R2 and Bn2From the equivalent circuits given in Figure 16 it canbe seen that the presence of ion accumulation repre-sented by the combination of R2 and Cn24mdash5 increasesthe overall impedance of the sensor Hence the reductionin impedance caused at high humidity due to the chargetransfer process is countered by ion accumulation

To better understand the fundamental sensing mecha-nisms R1 and Bn1 obtained through the numerical ttingprogram for the different sensors are plotted in Figures 17and 18 respectively It can be seen from Figure 17 thatR1 which is the resistance to charge transport in a staticelectric eld by the material containing an adsorbed waterlayer in the absence of ion migration is linear with RH

-11

10-10

10-9

10-8

10-7

50V30VDO15

Bn1

0 20 40 60 80 100

RH ()

Fig 18 Humidity dependence of Bn1 of different sensors obtained by tting the equivalent circuit model (Fig 17) with experimental data

286

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

in a semilog plot For sensor DO15 R1 decreases to afew hundred ohms at high RH levels indicating that theobserved conductivity is not due merely to the protonsprovided by the adsorbed water layer Since deionizedwater and argon gas are used to create the required humid-ity environment it is reasonable to assume that the sur-faces of the alumina lms contain anions presumablyfrom the electrolyte used for anodization which controlsconduction through the adsorbed layer108

With reference to Figure 18 Bn1 of the larger pore sam-ples SO50 and SO30 increases signi cantly only at thehigher humidity levels This result is in contrast to thatof the smaller pore size sensors which shows that Bn1

increases at lower humidity levels The relative humid-ity values where a signi cant increase in Bn1 takes placegenerally match the point seen in Figure 11 where theimpedance drops signi cantly which in turn is a func-tion of pore size These same relative humidity valuescorrelate with the point at which the spur appears inthe impedance diagrams given in Figure 15 The resultslead to the following conclusions Since hydrogen-bondedwater molecules possess a high dielectric constant theincrease in Bn1 indicates the presence of these bondson the surface which means that a liquid-like networkmore than one monolayer of physisorbed molecules hasformed at that humidity level This liquid-like network isresponsible for the sharp reduction in impedance seen inFigure 11 and the ion migration that creates the spur inthe low-frequency portion of the Cole-Cole plots as seenin Figure 15 According to Eq (1) capillary condensationdoes not take place in the alumina test sensors

In light of the above observations we propose thefollowing model to explain the behavior of nanoporousalumina in response to humidity The anodization pro-cess results in a certain amount of electrolyte anionstrapped within the pores1041 1071 108 Even in a low-humidityregime the presence of these anions on the sensor sur-face provides a high charge density for easy physisorp-tion of water molecules The result is the formationof a liquid-like network within the pores The impu-rity anions also act as proton donors by a dissociativemechanism104 joining the protons donated by the watermolecules The smaller the pore size the lower the RHlevel at which the liquid-like network is formed Hencefor the small pore sensors there is variation in Bn1 aswell as impedance magnitude in low-humidity regimesthe exponential dependence of R1 on RH also suggeststhe same phenomenon

In the case of DO15 the rst semicircle is highly dis-torted by the low-frequency spur originating with ionmigration A relatively large number of mobile ions mightbe present in these sensors leading to an enhancedaccumulation of protons at the electrode-sample contactAs mentioned earlier the dipoles in the singly bondedadsorbed water layer are free to rotate in an electric eld

Hence in the region close to the electrode-alumina inter-face the adsorbed water layer exhibits a net dipolar ori-entation and therefore it does not display its normaldielectric constant1011 103 Depending on ion size up toa monolayer of charge can exist at the electrode101 thediffuse (Gouy) layer of counter-ions spreads away fromthis inner (Stern) layer The number of ions available toform the Stern and Gouy layers at a particular frequencyincreases with increasing number of physisorbed layersHence as the number of physisorbed layers increases theStern and Gouy layers start forming at higher frequencies

With reference to Figure 12 the relative humiditylevel at which the sharp decrease in impedance occursshifts higher with increasing frequency This result can beexplained with the equivalent circuit shown in Figure 16athat is the circuit in the absence of ion accumulation Athigh frequencies the reactive impedance Xc of Cn14mdash5equal to Xc

D 1=jmdashCn14mdash5 is low and hence impedanceis largely determined by R1 As the humidity increasesR1 decreases however this translates into a reduction inimpedance only at relatively low frequencies where thevalue of R1 is comparable to or less than the capacitivereactance of Cn1(mdash) As the measurement frequency isreduced the reactance of Cn1(mdash) decreases hence the effectof R1 becomes more signi cant at lower humidity levels

32 Hydrogen Sensing with Titania Nanotubes

We recently reported62 the fabrication of self-organizedtitania nanotube arrays with an anodization techniqueAlthough the as-prepared nanotubes are amorphous theycrystallize on annealing at elevated temperatures and arestructurally stable to at least 600 C This stability ofstructure which is an essential criterion of a gas-sensingmaterial prompted us to study the gas-sensing behavior ofthese nanotubes for technologically important gases suchas oxygen carbon monoxide ammonia carbon dioxideand hydrogen

Titania has earned much attention for its oxygen-sensing capabilities110ndash115 Furthermore with propermanipulation of the microstructure or crystalline phaseandor the addition of the proper impurities or surfacefunctionalization titania can also be used as a reducinggas sensor116ndash127 The interaction of a gas with a metaloxide semiconductor is primarily a surface phenomenonand nanoporous metal oxides1221 1231 1281 129 offer the advan-tage of providing large sensing surface areas

Room-temperature metal oxide hydrogen sensors aregenerally based on Schottky barrier modulation of deviceslike PdTiO2 or PtTiO2 by hydrogen130ndash132 Elevated-temperature hydrogen sensors examine electrical resis-tance with hydrogen concentration for example Birke-feld et al133 observed that the resistance of the anatasephase of titania varies in the presence of carbon monoxideand hydrogen at temperatures above 500 C but on dop-ing with 10 alumina it becomes selective for hydrogen

287

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Barrier layer

Titanium Nano tube array

Platinum electrodes

Fig 19 Schematic representation of the hydrogen sensor geometry

Munuera et al134 used rhenium doping in the anatasephase to perform room-temperature hydrogen monitor-ing Recently Shimizu et al122 reported that anodizednanoporous titania lms with a Pd Schottky barrier aresensitive to hydrogen at 250 C

Titania nanotubes were grown from titanium foil(995 pure from Alfa Aesar Ward Hill MA) with athickness of 025 mm The anodization was performed inan electrolyte medium of 05 hydro uoric acid in waterwith the use of a platinum foil cathode Well-de nednanotube arrays were grown with anodizing potentialsranging from 12 V to 20 V Nanotube length increaseswith anodization time reaching 400 nm in approximately20 min and then remains constant For the present studythe samples were anodized for 25 min The samples werethen annealed at 500 C in a pure oxygen ambient for 6 hwith a heating and cooling rate of 1 Cmin

The electrode geometry of the titania nanotube sen-sors is shown in Figure 19 The sensor consists of a basetitanium metal foil with a nanotube array grown on topAn insulating barrier layer separates the nanotubes fromthe conducting titanium foil Electrical connections weremade by two parallel 10 mm pound 2 mm platinum pads of100-Œm thickness The test chamber consists of a 13-literquartz tube with stainless steel end caps placed insidea furnace (Thermolyne USA model 21100 tubular fur-nace) Gas ow through the test chamber was controlledvia a computer-controlled mass ow controller The elec-trical resistance of the titania sensors was measured witha computer-controlled Keithly digital multimeter Prior todata collection the test chamber was initially evacuatedwith a mechanical pump whereupon nitrogen (99999pure) was passed while the sensor under test was heatedto the desired temperature The test gases were mixed inappropriate ratios with nitrogen to create the necessarytest gas ambient

The surface morphology of nanotube arrays preparedwith an anodization potential of 20 V and annealed at500 C for 6 h in a pure oxygen ambient is shown inFigure 20a and b It can be seen from these images thatthe nanotubes are uniform over the surface The nano-tubes are approximately 400 nm in length and have abarrier layer62 thickness of ordm50 nm For the nanotubes

100 nm

(a)

(b)

1 mm (c)

100 nm Fig 20 The surface morphology of the titania nanotubes after anneal-ing at 500 C (a) High- and (b) low-magni cation images of a 20-Vsample (c) A high-magni cation image of a 12-V sample

fabricated with 20-V anodization the average pore diam-eter as determined from FESEM images is 76 nm (thestandard deviation 15 nm) with a wall thickness of 27 nm(standard deviation 6 nm) The sample anodized at 12 Vshown in Figure 2c was found to have an average porediameter of 46 nm (standard deviation 8 nm) with a wallthickness of 17 nm (standard deviation 2 nm) The porosi-ties of the 20-V and 12-V samples were calculated as 45and 61 respectively

Glancing-angle X-ray diffraction patterns of a 20-Vsample annealed at 500 C for 6 h in oxygen ambientis shown in Figure 21 It can be seen that both anataseand rutile phases of titania are present in the sampleA detailed study135 of these structures by high-resolutiontransmission electron microscope (HRTEM) showed that

288

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

20 30 40 50 60 70 80

2 q (deg)

(101

)

(110

)

A

R

TR

T

T

R R A TRR T R R T TT

Fig 21 Glancing-angle X-ray diffraction pattern of a 20-V sample(glancing angle D 2) annealed at 500 C in oxygen ambient A R andT represent respectively the re ections from anatase crystallites rutilecrystallites and the titanium substrate

the anatase crystallites were concentrated on the walls ofthe nanotubes and rutile on the barrier layer

Nanotubes annealed in a pure oxygen ambient werefound to be stable (intact) to temperatures of approxi-mately 580 C Above this temperature protrusions wereseen to be coming out through the nanotubes an effectthat spread with increasing temperature These protru-sions which are due to oxidation of the titanium substratecollapse the nanotubes

Figure 22 shows the response of the 20-V (76 nm) sam-ple as a function of ambient temperature as it is switchedfrom a nitrogen environment to one containing 1000 ppmhydrogen and then back to nitrogen The plot was madeusing 4Rg=R05

ƒ1 versus time where R0 is the base resis-tance of the sensor that is the sensor resistance before thetest gas was introduced and Rg is the measured resistancein the presence of test gas The sensor shows increasinghydrogen sensitivity with temperature with a change by3 orders of magnitude in resistance at temperatures aboveordm300 C At all of the temperatures the original resistancerecovered without hysteresis

The sensitivity S is de ned by the formula

S DR0

ƒ Rgs

Rgs

10 0

10 1

10 2

10 3

0 10 0 5 10 2 1 10 3 15 103 2 10 3 25 103 3 10 3 35 103 4 10 3

Time (Sec)

(Rg

R0)

-1

180degC

240degC

265degC

290degC

345degC

375degC

H2+N

2

N2

Fig 22 Variation of resistance Rg normalized with respect to baselineresistance R0 of a 20-V sample with time on exposure to 1000 ppmhydrogen at different temperatures It may be noted that the inverse ofRg=R0 was used in the plot to represent data in the positive y direction

10-1

100

101

102

103

150 200 250 300 350 400 450

Temperature (degC)

Sens

itivi

ty (

R0-

Rgs

)R

gs

Fig 23 The sensitivity temperature dependence of a 20-V sample to1000 ppm hydrogen

where R0 is the resistance of the sensor before the gas ispassed and Rgs is the resistance after gas is passed andreaches the saturation value The temperature-dependentsensitivity of a 20-V sample to 1000 ppm hydrogen isshown in Figure 23 Sensitivity is seen to increase withtemperature to approximately 380 C where the increasein sensitivity with temperature begins to saturate

The response time de ned as the time needed for thesensor to reach 90 of the nal signal for a given concen-tration of gas is plotted against temperature in Figure 24(the time includes that required for the gas to equili-brate inside the measurement chamber estimated to beordm30 s) The response time decreases exponentially withtemperature

To check the behavior of the sensor on repeated hydro-gen exposure the hydrogen concentration was varied indiscrete steps of 100 ppm from 0 to 500 ppm while thetemperature was kept constant at 290 C the chamber was ushed with nitrogen after each exposure to hydrogenThe response of the 20-V prepared nanotube sensorskept at 290 C is shown in Figure 25 The behaviorof the sensor is consistent recovering its original resis-tance after repeated exposure to varying hydrogen gasconcentrations The sensitivities of the 76-nm- 53-nm-and 22-nm-diameter nanotube sensors to hydrogen con-centrations ranging from 100 ppm to 1 are shown inFigure 26 The sensitivity of the 76-nm (20-V) sensor tolow concentrations is shown in Figure 27 there is a linearincrease in sensitivity at low concentrations The lower

10 1

10 2

10 3

10 4

150 200 250 300 350 400 450

Temperature (degC)

Res

pons

e T

ime

(Sec

)

Fig 24 Variation in response time for a 20-V sensor as a function oftemperature The dots represent measured data

289

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

10 2

0 4 4 4 4 4

10 3

0 10 1 10 2 10 3 10 4 10 5 10

Time (Sec)

Res

ista

nce

(W)

H2 (ppm)

200

300

400

500

400

300

200

100

100

Fig 25 Resistance of a 20-V sample when exposed to different con-centrations of hydrogen at 290 C The nitrogen-hydrogen mixture waspassed for 1500 s the chamber was then ushed with nitrogen for 3000 sbefore the nitrogen-hydrogen mixture was passed again

test limit of 20 ppm is determined by the experimentalapparatus and not the sensor

Figure 28 shows the change in resistance of the 76-nm-46-nm- and 22-nm-diameter nanotube sensors with expo-sure to 1000 ppm hydrogen at 290 C Whereas smallerdiameter nanotubes had greater sensitivity to hydrogenthe samples made at lower anodizing voltages tended tobe more brittle and harder to handle mechanically withoutbreaking

The 20-V 76-nm-diameter sample was exposed to oxy-gen carbon monoxide ammonia and carbon dioxide at290 C to study the cross-sensitivity The sensor wasfound to have no detectable variation in resistance onexposure to carbon dioxide The sensitivities of the tita-nia nanotubes to the other gases are shown in Figure 29The sensitivities of the nanotubes to carbon monoxide andammonia are negligible compared with that of hydrogenThe resistance of the nanotubes increased in the presenceof oxygen and did not regain their original electrical con-ductivity even after several hours in a 290 C nitrogenenvironment

1

10

100

1000

104

105

0 02 04 06 08 1

Hydrogen Concentration ()

Pore diameter = 76 nm

Pore diameter = 53 nm

Pore diameter = 22nm

Sen

sitiv

ity (

R0-

Rg)

Rg

Fig 26 The sensitivity variation of a samples with pore diameters of22 nm 53 nm and 76 nm versus time and hydrogen concentrations

0 100 200 300 400 5001

10

100

Hydrogen concentration (ppm)

Sen

sitiv

ity

(R0-

Rgs

)R

gs

Fig 27 Sensitivity of a 20-V (76 nm) sample at 290 C for low hydro-gen concentrations

Since the sensor measurements were made in atmo-spheres without oxygen the increase in conductivitycannot be due to hydrogen removing oxygen from thelattice132ndash134 or the removal of chemisorbed oxygen135ndash137

Hydrogen molecules can dissociate at defects on thetitania surface which can subsequently diffuse into thetitania lattice acting as electron donors1281 1381 139 How-ever a dissociation-driven process would result in aslow response and recovery times with complete recov-ery unlikely since the sensor completely regains its orig-inal resistance with hydrogen cycling it appears that thisis not the dominant mechanism behind high hydrogensensitivity We believe chemisorption of the dissociatedhydrogen on the titania surface is the underlying sensingmechanism137 During chemisorption hydrogen acts as asurface state and a partial charge transfer from hydrogento the titania conduction band takes place This createsan electron accumulation layer on the nanotube surfaceenhancing its electrical conductance Upon removal of thehydrogen ambient electron transfer back to the hydro-gen molecule takes place which subsequently desorbsrestoring the original electrical resistance of the mate-rial Another factor that may play a role in the hydrogensensitivity (and selectivity) is the platinum electrodes Atelevated temperatures hydrogen dissociation can occur onplatinum surfaces These dissociated hydrogen atoms mayspill1361 140 onto the nanotube surface where they diffuse

10-1

101

103

105

0 05 1 15

Time (Hour)

Pore size - 76 nm

Pore size - 53 nm

Pore size - 22 nm

(Rg

R0)

-1

Fig 28 A comparison of the variation in resistance of samples havingpore diameters of 22 nm 53 nm and 76 nm versus time upon exposureto 1000 ppm of hydrogen at 290 C

290

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

10-2

10-1

100

0 100 2 102 4 102 6 102 8 102 1 103

OxygenAmmoniaCarbon monoxide

Time (Sec)

(Rg

R0)

-1

Fig 29 Variation of resistance of a 20-V sample when exposed to1000 ppm carbon monoxide 5 ammonia and 1 oxygen at 290 C

into the material affecting its electrical properties Fromthe present study it was not clear how signi cant a rolethe platinum electrodes play

Anatase the polymorph of titania has been reported tohave high sensitivity for reducing gases like hydrogen andcarbon monoxide1211 1241 133 Our nanotube samples containanatase phase mainly on the walls and rutile in the barrierlayer As the diffusing hydrogen atoms go to the intersti-tial sites1331 142 and as the ca ratio of anatase is almostfour times that of rutile it appears that the anatase latticeaccommodates hydrogen easily and hence makes a highercontribution to hydrogen sensitivity

The effect of chemisorption can be neglected in theoxygen-sensing experiments As the recovery requiresseveral hours it appears that the nanotubes contain oxy-gen vacancies or titanium interstitial defects in the pres-ence of nitrogen On exposure of the sensor to oxygenambient the lattice reoxidizes and hence the conductiv-ity of the sensor decreases On the removal of oxygenthe reduction of the lattice will not immediately occurhence the sensor requires several hours to regain its orig-inal conductivity

4 CONCLUSIONS

This paper summarizes different routes of fabrication ofmetal oxide nanoarchitectures useful for sensing applica-tions Structures made via anodization of a starting metaland breath gure formation from a sol-gel are discussedand presented as are TiO2 nanowires made by lling ananoporous alumina membrane with subsequent removalof the supporting template Two test cases illustratingthe utility of these nanoporous metal oxide structures forsensing are presented highly ordered nanoporous aluminais used for humidity sensing and TiO2 nanotubes are usedfor hydrogen sensing These study test cases reveal sev-eral points worth brie y noting

In the rst sensing test case uniform nanoporous alu-mina sensors exhibit pore size and frequency-dependent

humidity-sensing characteristics The mechanism respon-sible for the sensing action of the nanoporous alumina sen-sors was studied by representing the impedance spectra atdifferent humidity levels in Cole-Cole plots This makes itpossible to identify the competing impedance-determiningprocesses occurring in the sensors namely charge transferand mass transport through the physisorbed water layeras dependent upon the humidity level and input signalfrequency Charge transfer was found to enhance the sen-sitivity by causing a large variation in impedance withrespect to humidity Mass transport leads to accumulationof ions at the electrodes which tends to decrease the mea-surement sensitivity by increasing the impedance Hencea proper choice of operating frequency high enough toavoid ion accumulation is essential for optimum sensorperformance

It has been inferred from our results that anions becomeincorporated into the alumina lm during the anodiza-tion process supporting the hypothesis of Khanna andNahar104ndash106 where they act as proton donors and facili-tate the adsorption of water molecules Hence liquid-likecharge transfer networks are formed at relatively lowerhumidity levels for the smaller pore-size sensors Thisprocess rather than capillary condensation is primarilyresponsible for the observed humidity-sensing behavior ofthe sensors Our studies show that a distribution of poresizes as suggested by Shimizu et al94 is not necessary toobtain sensitive performance over a wide humidity rangerather such optimal performance can be obtained with uni-form nanodimensional pores containing adsorbed anionicimpurities

In the second sensing test case titania nanotubes pre-pared by anodization and annealed in an oxygen atmo-sphere at a temperature of 500 C were found to behighly sensitive to hydrogen The nanotube sensors con-tained both anatase and rutile phases of titania and showedappreciable sensitivity toward hydrogen at temperaturesas low as 180 C The sensitivity increased drasticallywith temperature showing a variation of 3 orders in mag-nitude of resistance to 1000 ppm of hydrogen at 400 CThe response time decreased with increasing temperatureat 290 C full switching of the sensor took approximately3 min Results were highly reproducible with no indica-tion of hysteresis Our results showed that these sensorsare at least capable of monitoring hydrogen levels from20 ppm to 4 At 290 C nanotubes with smaller porediameters showed higher sensitivity to hydrogen com-pared with their larger pore counterparts The sensorsshowed high selectivity to hydrogen compared with car-bon monoxide ammonia and carbon dioxide Althoughthe sensor was sensitive to high concentrations of oxy-gen the response time was high and the sensor did notcompletely regain the original condition We believe thehydrogen sensitivity of the nanotubes is due to hydrogendiffusion into the titania lattice where they act as electrondonors

291

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Acknowledgments Support of this work by the NationalScience Foundation through grant ECS-0210033 is grate-fully acknowledged

References and Notes

1 K Kajihara K Tanaka K Horao and N Soga Jpn J ApplPhys 35 6110 (1996)

2 B O Regan and M Gratzel A low cost Nature 353 737 (1991)3 A Fujishima and K Honda Nature 238 37 (1972)4 K Sato A Tsuzuki H Taoda Y Torii T Kato and Y Butsugan

J Mater Sci 29 5911 (1994)5 K Sato A Tsuzuki Y Torii H Taoda T Kato and Y Butsugan

J Mater Sci 30 837 (1995)6 N Ozer and C M Lampert Sol Energy Mater Sol Cells 54 147

(1998)7 A Manz and H Becker editors Microsystem Technology in

Chemistry and Life Sciences Springer-Verlag Berlin 19988 E Kim Y Xia and G M Whitesides Nature 376 581 (1995)9 M Templin A Franck A Du Chesne A Leist A Zhang

R Ulrich V Schadler and U Weisner Science 278 1795 (1997)10 D Zhao J Feng Q Huo N Melosh G H Frederickson

B Chmelka and G D Stucky Science 279 548 (1998)11 O D Velev T A Jede R F Lobo and A M Lenhoff Nature

389 447 (1997)12 B T Holland C F Blanford and A Stein Science 281 538

(1998)13 S Matsushita T Miwa and A Fujishima Chem Lett 9 925

(1997)14 A Imhof and D J Pine Adv Mater 10 697 (1998)15 T Tatsuma A Ikezawa Y Ohko T Miwa T Matsue and

A Fujishima Adv Mater 12 643 (2000)16 Z Zhong Y Yin B Gates and Y Xia Adv Mater 12 206 (2000)17 O D Velev and A M Lenhoff Curr Opin Colloids Interface

Sci 5 56 (2000)18 O D Velev and E W Kaler Adv Mater 12 531 (2000)19 P Jiang J Cizeron J F Bertone and V L Colvin J Am Chem

Soc 121 7957 (1999)20 M E Turner T J Trentler and V L Colvin Adv Mater 13 180

(2001)21 K M Kulinowski P Jiang H Vaswani and V L Colvin Adv

Mater 12 833 (2000)22 P Jiang J F Bertone and V L Colvin Science 291 453 (2001)23 R Seshadri and F C Meldrum Adv Mater 12 1149 (2000)24 F C Meldrum and R Seshadri Chem Commun 29 (2000)25 Q B Meng Z Z Gu O Sato and A Fujishima Appl Phys

Lett 77 4313 (2000)26 G Widawski B Rawiso and B Francois Nature 369 3897

(1994)27 B Francois O Pitois and J Francois Adv Mater 7 1041 (1995)28 O Pitois and B Francois Eur Phys J B 8 225 (1999)29 S A Jenekhe and X L Chen Science 283 372 (1999)30 M Shimomuara and T Sawadaishi Curr Opin Colloid Interface

Sci 6 11 (2001)31 O Karthaus N Maruyama X Cieren M Shimomura

H Hasegawa and T Hashimoto Langmuir 16 6071 (2000)32 T Nishikawa J Nishida R Ookura S Nishimura S Wada

T Karino and M Shimomura Mater Sci Eng C 10 141 (1999)33 K Kajihara K Nakanishi K Tanaka K Hirao and N Soga

J Am Ceram Soc 81 2670 (1998)34 C A Grimes R S Singh E C Dickey and O K Varghese

J Mater Res 16 1686 (2001)35 M Srinivasrao D Collings A Phillips and S Patel Science 292

79 (2001)36 P G de Gennes Rev Mod Phys 57 827 (1985)

37 D Beysens A Steyer P Guenoun D Fritter and C M KnoblerPhase Transitions 31 219 (1991)

38 B J Briscoe and K P Galvin J Phys D 23 422 (1990)39 B J Briscoe and K P Galvin J Phys D 23 1265 (1990)40 D Beysens and C M Knobler Phys Rev Lett 57 1433 (1986)41 F Family and P Meakin Phys Rev Lett 61 428 (1988)42 A V Limaye R D Narhe A M Dhote and S B Ogale Phys

Rev Lett 76 3762 (1996)43 P Dellrsquo Aversana and G P Neitzel Phys Today 51 38 (1998)44 P Dellrsquo Aversana J R Banavar and J Koplik Phys Fluids 8 15

(1986)45 P Dellrsquo Aversana V Tontodonato and L Carotenuto Phys Fluids

9 2475 (1997)46 M Q Li S Q Xu and E Kumacheva Langmuir 16 7275 (2000)47 P Freymuth Rev Sci Instrum 64 1 (1993)48 P Ball The Self-Made Tapestry Pattern Formation in Nature

Oxford University Press New York 199949 M F Schatz S J VanHook W D McCormick J B Swift and

H L Swinney Phys Fluids 11 2577 (1999)50 S R Coriell M R Cordes and W J Boettinger J Cryst Growth

49 13 (1980)51 T S Sullivan Y M Liu and R E Ecke Phys Rev E 54 486

(1996)52 J L Rogers M F Schatz J L Bougie and J B Swift 84 87

(2000)53 K Nitschke 52 R5772 (1995)54 E Bodenschatz J R De Bruyn G Ahlers and D S Cannell

Phys Rev Lett 67 3078 (1991)55 C W Meyer D S Cannell G Ahlers J B Swift and P C

Hohenberg Phys Rev Lett 61 947 (1988)56 R F Ismagilov D Rosmarin D H Gracias A D Stroock and

G M Whitesides Appl Phys Lett 79 439 (2001)57 E A Barringer and H K Bowen Langmuir 1 420 (1985)58 M Tschapek C Wasowski and R M T Sanchez J Electroanal

Chem 74 167 (1976)59 H Masuda and K Fukuda Science 268 1466 (1995)60 H Masuda F Hasegwa and S Ono J Electrochem Soc 144

L 127 (1997)61 O Jessensky F Muller and U Gosele Appl Phys Lett 72 1173

(1998)62 D Gong C A Grimes O K Varghese W Hu R S Singh

Z Chen and E C Dickey J Mater Res 16 3331 (2001)63 B M Kulwicki J Am Ceram Soc 74 697 (1991)64 H Arai and T Seiyama in Sensors A Comprehensive Survey

edited by W Gopel J Hesse and J N Zemel VCH Weinheim(1992) Vol 3 pp 981ndash1012

65 E Traversa Sens Actuators B 23 135 (1995)66 N Yamazoe and Y Shimizu Sens Actuators 10 379 (1986)67 A Bearzotti I Fratoddi L Palummo S Petrocco A Furlani

C Lo Sterzo and M V Russo Sens Actuators B 76 316 (2001)68 Y Sakai Y Sadaoka and M Matsuguchi Sens Actuators B 35

85 (1996)69 C A Grimes and D Kouzoudis Sens Actuators A 84 205

(2000)70 K G Ong C A Grimes C L Robbins and R S Singh Sens

Actuators A 93 33 (2001)71 T Seiyama N Yamazoe and H Arai Sens Actuators 4 85

(1983)72 L Ketron Ceram Bull 68 860 (1989)73 E Traversa G Gusmano and A Montenero Eur J Solid State

Inorg Chem 32 719 (1995)74 M K Jain M C Bhatnagar and G L Sharma Sens Actuators

B 55 180 (1999)75 O K Varghese and L K Malhotra J Appl Phys 87 7457 (2000)76 G Sberveglieri R Murri and N Pinto Sens Actuators B 23

177 (1995)

292

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

103

104

105

106

0 20 40 60 80 100

Z (SO30)

Z(SO50)

Z(DO15)

| Z| (

W)

RH ()

Fig 11 Measured 5-kHz impedance of sensors over different humiditylevels

Figure 12 shows the response of the different sensorsto humidity at different measurement frequencies It canbe seen from the gure that as the frequency decreasesthe width of the sensitive region increases and the point atwhich the sensor becomes responsive shifts to lower rela-tive humidity values For a given pore size the humidity-sensitive region of operation can be selected by frequencytuning

The response time de ned as the time needed to reach90 of the nal signal for a given relative humidityand the recovery time de ned as the time taken forthe signal to come to within 10 of the initial valuewere determined by alternately exposing the sensors toa 2 and a 45 RH ambient with impedance measuredat 5 kHz A typical responserecovery graph of sensorSO50 is shown in Figure 13 All sensors were com-pletely reversible regaining their original impedance val-ues even after repeated exposure to high humidity levelsThe response time and recovery times corresponding todifferent pore diameters are shown in Figure 14 The min-imum responserecovery times were observed for SO50the sample with the largest pore size

The impedance spectra of the sensors were taken overa range of humidity levels The total impedance wasresolved into real (Z0) and imaginary (Z00) parts and Cole-Cole impedance plots were constructed Figure 15andashcshows respectively the Cole-Cole plot of sensors SO50SO30 and DO15 The equivalent circuit models shown inFigure 16 were used to t the experimental data The t-ting was accomplished with the complex nonlinear least-square- tting program101 The dots in the plots repre-sent experimental data and lines represent the equiva-lent circuit model t The equivalent circuit denoted (a)or (b) with reference to Figure 16 used for tting isdenoted near the corresponding curves in Figure 15 Theequivalent circuit models are different from the empiri-cal model suggested by Falk et al102 for porous aluminasensors fabricated by anodization of aluminum thin lmsIn the equivalent circuit models R1 and R2 represent twofrequency-independent resistors in parallel with disper-sive frequency-dependent capacitors Cn14mdash5 and Cn24mdash5respectively103 These non-Debye capacitances can be

103

104

105

106

107

SO50

5kHz10kHz100kHz1MHz

0 20 40 60 80 100

0 20 40 60 80 100

0 20 40 60 80 100

| Z| (

W)

| Z| (

W)

| Z| (

W)

RH ()(a)

(c)

(b)

103

104

105

106

107

SO30

5kHz10kHz100kHz1MHz

RH ()

102

103

104

105

106

DO15

5kHz10kHz100kHz1MHz

RH ()

Fig 12 Variation of sensor impedance with humidity and measure-ment frequency (a) SO50 (b) SO30 (c) DO15

68

7

72

74

76

78

0 200 400 600 800

Time (Sec)

45 RH

2 RH

45 RH 45 RH

2 RH

| Z| (

x105

W)

Fig 13 Response of sensor SO50 upon repeated cycling between rel-ative humidity environments of 2 and 45

284

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

10

15

20

25

30

35

40

45

50

20 25 30 35 40 45 50

Res

pons

eR

ecov

ery

Tim

e (S

ec)

Pore diameter (nm)

DO15

SO30

SO50

Fig 14 Dependence of response time and recovery time on pore size

expressed in a general form as Cn4mdash5 D Bn4imdash5nƒ1 whereBn is a constant for a given set of experimental conditionsand n takes a value between 0 and 1 according to the localmicroscopic environment through which charge transporttakes place751 1021 103 Bn behaves as an ideal frequency-independent capacitor as n approaches 1 and as an idealconductor when the value of n approaches 0101

By examination of Figure 15a SO50 it can be seenthat at low humidity levels the Cole-Cole plot describesan arc with a very large radius of curvature As the humid-ity increases the radius decreases and at an RH of about80 the complete semicircle comes inside the measur-able impedance range However at the same humiditylevel a spur appears in the low-frequency region that dis-torts the semicircle The spur which can be viewed aspart of a second semicircle dominates at higher humid-ity levels while the semicircle described by the high-frequency impedance values diminishes in size Thisbehavior is similar for all sensors For sensors SO50SO30 and DO15 respectively the RH at which thehigh-frequency semicircle comes completely within themeasurable impedance range is ordm75 65 and 35these values roughly match the in ection points seen inFigure 11

The fundamental mechanism that enables ceramic sen-sors to sense humidity is the physisorption of watermolecules on an initially chemisorbed layer of hydroxylions651 96 The chemisorbed hydroxyl ions enhance theelectrical conductivity of the sensor either by donatingelectrons to the conduction band of the base materialor through proton hopping between adjacent hydroxylgroups upon the application of an electric eld Theprocess of chemisorption occurs at very low humiditylevels and is unaffected by further changes in humid-ity However an increase in humidity makes the watermolecules physisorb to this hydroxyl layer The effec-tiveness of physisorption depends upon the cation chargecomplexes (from the material or the impurities) and thehydroxyl ions (water molecules present at the surface ofthe base material)96 The physisorption process is facili-tated by higher surface charge densities During formationof the rst physisorbed layer a water molecule becomesattached to two neighboring hydroxyl groups through

0

02

04

06

08

1

12

0 02 04 06 08 1 12

Zrsquo (MW)

35 (a) 55 (a)

65 (a)

75 (b)

81 (b)

93 (b) SO50

0

02

04

06

08

1

12

14

0 02 04 06 08 1 12 14

Zrsquo (W)

Z

(W)

Z

(MW

)Z

(M

W)

35 (a)

55 (a)

65 (b)

75 (b)

81 (b)

93 (b)

SO30

0

2 105

4 105

6 105

8 105

1 106

12 106

0 2 105 4 105 6 105 8 105 1 106 12 106

17 (a)27 (a)

35 (b)45 (b)55 (b)

93 (b)

DO15

(a)

Zrsquo (MW)

(b)

(c)

Fig 15 Cole-Cole plot of the sensors at different humidity levels(a) SO50 (b) SO30 (c) DO15 The points represent the experimentaldata and lines represent the t with the equivalent circuit model givenin Figure 9 The RH values at which the data were collected and theequivalent circuit used for tting each curve (given as (a) or (b) denot-ing the circuit of Figure 17a and b respectively) are shown adjacent toeach curve

hydrogen double bonds and a proton may be transferredfrom a hydroxyl group to the water molecule to form aH3OC ion

When the physisorption occurs within less than a monolayer that is when clusters of physisorbed molecules

285

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

(b)

(a)

Cn1 (w)

R1

Cn2 (w) Cn1 (w)

R1 R2

Fig 16 Equivalent circuit used for tting the experimental data

are present at the surface it is thought that H3OC diffu-sion within hydroxyl groups and proton transfer betweenadjacent water molecules clusters take place96 HoweverKhanna and Nahar104ndash106 refute this theory They arguethat the proton diffusion is less probable in a surfacethat contains hydroxyl ions and clusters of physisorbedmolecules and hence cannot account for the conductiv-ity of porous alumina at an RH less than 40 that isbefore the rst monolayer is formed Instead they sug-gested a phonon-assisted electron tunneling mechanism atlow humidity levels in which electron tunneling occursfrom one water molecule to another as enhanced by thesurface anions of the alumina sensors

At higher humidity levels the number of physisorbedlayers increases allowing each water molecule to besingly bonded to a hydroxyl group and proton hop-ping between adjacent water molecules in the continu-ous water layer takes place The conduction process isthe same as that of pure water and is called the Grot-thuss chain reaction96 The dominant charge carrier ina high-humidity environment is therefore HC ions (pro-tons) The concentration of HC increases with increas-ing humidity and HC move freely through the water-likelayer According to the theory of Khanna and Nahar104

in porous alumina the protons for conduction are donatednot only by H2O but also by the Al(OH)3 formed asa result of water adsorption and the impurity anionsincorporated into the lm from the electrolyte duringanodization107ndash109 The water-like network consists ofsingly bonded water molecules which possess a highdielectric constant as they form dipoles and reorient freelyunder an externally applied electric eld The impedanceof the nanoporous alumina sensors includes this capaci-tive contribution as well

From the analysis of Figure 15 the high-frequencysemicircle that diminishes in size with increasing RHarises because of the effect of a physisorbed water layerat the surface Therefore R1 in the equivalent circuitof Figure 16 represents the charge transfer resistancethrough the adsorbed water layer The effect of electronic

102

104

106

108

SO50SO30

DO15

R1

(W)

0 20 40 60 80 100

RH ()

Fig 17 Humidity dependence of resistance R1 of different sensorsobtained by tting the equivalent circuit model with experimental data

conduction through the alumina bulk is parallel to thatof the surface conduction affected by the adsorbed waterlayer This bulk conduction process is dominant only atextremely low humidity values when there are only afew adsorbed molecules Hence the incomplete semicir-cle or arc in the low-humidity region is partially deter-mined from the base alumina Bn1 represents the capac-itance contribution from the adsorbed water layer Thelarge spurs seen in the low-frequency region are a con-sequence of ion migration in the adsorbed layer towardelectrodes1011 103 This migration leads to accumulation ofions at the electrodes Since the relaxation time of ionmigration is larger than the charge transfer process itis seen only at low frequencies The accumulated ionsin the adsorbed water layer give rise to R2 and Bn2From the equivalent circuits given in Figure 16 it canbe seen that the presence of ion accumulation repre-sented by the combination of R2 and Cn24mdash5 increasesthe overall impedance of the sensor Hence the reductionin impedance caused at high humidity due to the chargetransfer process is countered by ion accumulation

To better understand the fundamental sensing mecha-nisms R1 and Bn1 obtained through the numerical ttingprogram for the different sensors are plotted in Figures 17and 18 respectively It can be seen from Figure 17 thatR1 which is the resistance to charge transport in a staticelectric eld by the material containing an adsorbed waterlayer in the absence of ion migration is linear with RH

-11

10-10

10-9

10-8

10-7

50V30VDO15

Bn1

0 20 40 60 80 100

RH ()

Fig 18 Humidity dependence of Bn1 of different sensors obtained by tting the equivalent circuit model (Fig 17) with experimental data

286

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

in a semilog plot For sensor DO15 R1 decreases to afew hundred ohms at high RH levels indicating that theobserved conductivity is not due merely to the protonsprovided by the adsorbed water layer Since deionizedwater and argon gas are used to create the required humid-ity environment it is reasonable to assume that the sur-faces of the alumina lms contain anions presumablyfrom the electrolyte used for anodization which controlsconduction through the adsorbed layer108

With reference to Figure 18 Bn1 of the larger pore sam-ples SO50 and SO30 increases signi cantly only at thehigher humidity levels This result is in contrast to thatof the smaller pore size sensors which shows that Bn1

increases at lower humidity levels The relative humid-ity values where a signi cant increase in Bn1 takes placegenerally match the point seen in Figure 11 where theimpedance drops signi cantly which in turn is a func-tion of pore size These same relative humidity valuescorrelate with the point at which the spur appears inthe impedance diagrams given in Figure 15 The resultslead to the following conclusions Since hydrogen-bondedwater molecules possess a high dielectric constant theincrease in Bn1 indicates the presence of these bondson the surface which means that a liquid-like networkmore than one monolayer of physisorbed molecules hasformed at that humidity level This liquid-like network isresponsible for the sharp reduction in impedance seen inFigure 11 and the ion migration that creates the spur inthe low-frequency portion of the Cole-Cole plots as seenin Figure 15 According to Eq (1) capillary condensationdoes not take place in the alumina test sensors

In light of the above observations we propose thefollowing model to explain the behavior of nanoporousalumina in response to humidity The anodization pro-cess results in a certain amount of electrolyte anionstrapped within the pores1041 1071 108 Even in a low-humidityregime the presence of these anions on the sensor sur-face provides a high charge density for easy physisorp-tion of water molecules The result is the formationof a liquid-like network within the pores The impu-rity anions also act as proton donors by a dissociativemechanism104 joining the protons donated by the watermolecules The smaller the pore size the lower the RHlevel at which the liquid-like network is formed Hencefor the small pore sensors there is variation in Bn1 aswell as impedance magnitude in low-humidity regimesthe exponential dependence of R1 on RH also suggeststhe same phenomenon

In the case of DO15 the rst semicircle is highly dis-torted by the low-frequency spur originating with ionmigration A relatively large number of mobile ions mightbe present in these sensors leading to an enhancedaccumulation of protons at the electrode-sample contactAs mentioned earlier the dipoles in the singly bondedadsorbed water layer are free to rotate in an electric eld

Hence in the region close to the electrode-alumina inter-face the adsorbed water layer exhibits a net dipolar ori-entation and therefore it does not display its normaldielectric constant1011 103 Depending on ion size up toa monolayer of charge can exist at the electrode101 thediffuse (Gouy) layer of counter-ions spreads away fromthis inner (Stern) layer The number of ions available toform the Stern and Gouy layers at a particular frequencyincreases with increasing number of physisorbed layersHence as the number of physisorbed layers increases theStern and Gouy layers start forming at higher frequencies

With reference to Figure 12 the relative humiditylevel at which the sharp decrease in impedance occursshifts higher with increasing frequency This result can beexplained with the equivalent circuit shown in Figure 16athat is the circuit in the absence of ion accumulation Athigh frequencies the reactive impedance Xc of Cn14mdash5equal to Xc

D 1=jmdashCn14mdash5 is low and hence impedanceis largely determined by R1 As the humidity increasesR1 decreases however this translates into a reduction inimpedance only at relatively low frequencies where thevalue of R1 is comparable to or less than the capacitivereactance of Cn1(mdash) As the measurement frequency isreduced the reactance of Cn1(mdash) decreases hence the effectof R1 becomes more signi cant at lower humidity levels

32 Hydrogen Sensing with Titania Nanotubes

We recently reported62 the fabrication of self-organizedtitania nanotube arrays with an anodization techniqueAlthough the as-prepared nanotubes are amorphous theycrystallize on annealing at elevated temperatures and arestructurally stable to at least 600 C This stability ofstructure which is an essential criterion of a gas-sensingmaterial prompted us to study the gas-sensing behavior ofthese nanotubes for technologically important gases suchas oxygen carbon monoxide ammonia carbon dioxideand hydrogen

Titania has earned much attention for its oxygen-sensing capabilities110ndash115 Furthermore with propermanipulation of the microstructure or crystalline phaseandor the addition of the proper impurities or surfacefunctionalization titania can also be used as a reducinggas sensor116ndash127 The interaction of a gas with a metaloxide semiconductor is primarily a surface phenomenonand nanoporous metal oxides1221 1231 1281 129 offer the advan-tage of providing large sensing surface areas

Room-temperature metal oxide hydrogen sensors aregenerally based on Schottky barrier modulation of deviceslike PdTiO2 or PtTiO2 by hydrogen130ndash132 Elevated-temperature hydrogen sensors examine electrical resis-tance with hydrogen concentration for example Birke-feld et al133 observed that the resistance of the anatasephase of titania varies in the presence of carbon monoxideand hydrogen at temperatures above 500 C but on dop-ing with 10 alumina it becomes selective for hydrogen

287

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Barrier layer

Titanium Nano tube array

Platinum electrodes

Fig 19 Schematic representation of the hydrogen sensor geometry

Munuera et al134 used rhenium doping in the anatasephase to perform room-temperature hydrogen monitor-ing Recently Shimizu et al122 reported that anodizednanoporous titania lms with a Pd Schottky barrier aresensitive to hydrogen at 250 C

Titania nanotubes were grown from titanium foil(995 pure from Alfa Aesar Ward Hill MA) with athickness of 025 mm The anodization was performed inan electrolyte medium of 05 hydro uoric acid in waterwith the use of a platinum foil cathode Well-de nednanotube arrays were grown with anodizing potentialsranging from 12 V to 20 V Nanotube length increaseswith anodization time reaching 400 nm in approximately20 min and then remains constant For the present studythe samples were anodized for 25 min The samples werethen annealed at 500 C in a pure oxygen ambient for 6 hwith a heating and cooling rate of 1 Cmin

The electrode geometry of the titania nanotube sen-sors is shown in Figure 19 The sensor consists of a basetitanium metal foil with a nanotube array grown on topAn insulating barrier layer separates the nanotubes fromthe conducting titanium foil Electrical connections weremade by two parallel 10 mm pound 2 mm platinum pads of100-Œm thickness The test chamber consists of a 13-literquartz tube with stainless steel end caps placed insidea furnace (Thermolyne USA model 21100 tubular fur-nace) Gas ow through the test chamber was controlledvia a computer-controlled mass ow controller The elec-trical resistance of the titania sensors was measured witha computer-controlled Keithly digital multimeter Prior todata collection the test chamber was initially evacuatedwith a mechanical pump whereupon nitrogen (99999pure) was passed while the sensor under test was heatedto the desired temperature The test gases were mixed inappropriate ratios with nitrogen to create the necessarytest gas ambient

The surface morphology of nanotube arrays preparedwith an anodization potential of 20 V and annealed at500 C for 6 h in a pure oxygen ambient is shown inFigure 20a and b It can be seen from these images thatthe nanotubes are uniform over the surface The nano-tubes are approximately 400 nm in length and have abarrier layer62 thickness of ordm50 nm For the nanotubes

100 nm

(a)

(b)

1 mm (c)

100 nm Fig 20 The surface morphology of the titania nanotubes after anneal-ing at 500 C (a) High- and (b) low-magni cation images of a 20-Vsample (c) A high-magni cation image of a 12-V sample

fabricated with 20-V anodization the average pore diam-eter as determined from FESEM images is 76 nm (thestandard deviation 15 nm) with a wall thickness of 27 nm(standard deviation 6 nm) The sample anodized at 12 Vshown in Figure 2c was found to have an average porediameter of 46 nm (standard deviation 8 nm) with a wallthickness of 17 nm (standard deviation 2 nm) The porosi-ties of the 20-V and 12-V samples were calculated as 45and 61 respectively

Glancing-angle X-ray diffraction patterns of a 20-Vsample annealed at 500 C for 6 h in oxygen ambientis shown in Figure 21 It can be seen that both anataseand rutile phases of titania are present in the sampleA detailed study135 of these structures by high-resolutiontransmission electron microscope (HRTEM) showed that

288

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

20 30 40 50 60 70 80

2 q (deg)

(101

)

(110

)

A

R

TR

T

T

R R A TRR T R R T TT

Fig 21 Glancing-angle X-ray diffraction pattern of a 20-V sample(glancing angle D 2) annealed at 500 C in oxygen ambient A R andT represent respectively the re ections from anatase crystallites rutilecrystallites and the titanium substrate

the anatase crystallites were concentrated on the walls ofthe nanotubes and rutile on the barrier layer

Nanotubes annealed in a pure oxygen ambient werefound to be stable (intact) to temperatures of approxi-mately 580 C Above this temperature protrusions wereseen to be coming out through the nanotubes an effectthat spread with increasing temperature These protru-sions which are due to oxidation of the titanium substratecollapse the nanotubes

Figure 22 shows the response of the 20-V (76 nm) sam-ple as a function of ambient temperature as it is switchedfrom a nitrogen environment to one containing 1000 ppmhydrogen and then back to nitrogen The plot was madeusing 4Rg=R05

ƒ1 versus time where R0 is the base resis-tance of the sensor that is the sensor resistance before thetest gas was introduced and Rg is the measured resistancein the presence of test gas The sensor shows increasinghydrogen sensitivity with temperature with a change by3 orders of magnitude in resistance at temperatures aboveordm300 C At all of the temperatures the original resistancerecovered without hysteresis

The sensitivity S is de ned by the formula

S DR0

ƒ Rgs

Rgs

10 0

10 1

10 2

10 3

0 10 0 5 10 2 1 10 3 15 103 2 10 3 25 103 3 10 3 35 103 4 10 3

Time (Sec)

(Rg

R0)

-1

180degC

240degC

265degC

290degC

345degC

375degC

H2+N

2

N2

Fig 22 Variation of resistance Rg normalized with respect to baselineresistance R0 of a 20-V sample with time on exposure to 1000 ppmhydrogen at different temperatures It may be noted that the inverse ofRg=R0 was used in the plot to represent data in the positive y direction

10-1

100

101

102

103

150 200 250 300 350 400 450

Temperature (degC)

Sens

itivi

ty (

R0-

Rgs

)R

gs

Fig 23 The sensitivity temperature dependence of a 20-V sample to1000 ppm hydrogen

where R0 is the resistance of the sensor before the gas ispassed and Rgs is the resistance after gas is passed andreaches the saturation value The temperature-dependentsensitivity of a 20-V sample to 1000 ppm hydrogen isshown in Figure 23 Sensitivity is seen to increase withtemperature to approximately 380 C where the increasein sensitivity with temperature begins to saturate

The response time de ned as the time needed for thesensor to reach 90 of the nal signal for a given concen-tration of gas is plotted against temperature in Figure 24(the time includes that required for the gas to equili-brate inside the measurement chamber estimated to beordm30 s) The response time decreases exponentially withtemperature

To check the behavior of the sensor on repeated hydro-gen exposure the hydrogen concentration was varied indiscrete steps of 100 ppm from 0 to 500 ppm while thetemperature was kept constant at 290 C the chamber was ushed with nitrogen after each exposure to hydrogenThe response of the 20-V prepared nanotube sensorskept at 290 C is shown in Figure 25 The behaviorof the sensor is consistent recovering its original resis-tance after repeated exposure to varying hydrogen gasconcentrations The sensitivities of the 76-nm- 53-nm-and 22-nm-diameter nanotube sensors to hydrogen con-centrations ranging from 100 ppm to 1 are shown inFigure 26 The sensitivity of the 76-nm (20-V) sensor tolow concentrations is shown in Figure 27 there is a linearincrease in sensitivity at low concentrations The lower

10 1

10 2

10 3

10 4

150 200 250 300 350 400 450

Temperature (degC)

Res

pons

e T

ime

(Sec

)

Fig 24 Variation in response time for a 20-V sensor as a function oftemperature The dots represent measured data

289

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

10 2

0 4 4 4 4 4

10 3

0 10 1 10 2 10 3 10 4 10 5 10

Time (Sec)

Res

ista

nce

(W)

H2 (ppm)

200

300

400

500

400

300

200

100

100

Fig 25 Resistance of a 20-V sample when exposed to different con-centrations of hydrogen at 290 C The nitrogen-hydrogen mixture waspassed for 1500 s the chamber was then ushed with nitrogen for 3000 sbefore the nitrogen-hydrogen mixture was passed again

test limit of 20 ppm is determined by the experimentalapparatus and not the sensor

Figure 28 shows the change in resistance of the 76-nm-46-nm- and 22-nm-diameter nanotube sensors with expo-sure to 1000 ppm hydrogen at 290 C Whereas smallerdiameter nanotubes had greater sensitivity to hydrogenthe samples made at lower anodizing voltages tended tobe more brittle and harder to handle mechanically withoutbreaking

The 20-V 76-nm-diameter sample was exposed to oxy-gen carbon monoxide ammonia and carbon dioxide at290 C to study the cross-sensitivity The sensor wasfound to have no detectable variation in resistance onexposure to carbon dioxide The sensitivities of the tita-nia nanotubes to the other gases are shown in Figure 29The sensitivities of the nanotubes to carbon monoxide andammonia are negligible compared with that of hydrogenThe resistance of the nanotubes increased in the presenceof oxygen and did not regain their original electrical con-ductivity even after several hours in a 290 C nitrogenenvironment

1

10

100

1000

104

105

0 02 04 06 08 1

Hydrogen Concentration ()

Pore diameter = 76 nm

Pore diameter = 53 nm

Pore diameter = 22nm

Sen

sitiv

ity (

R0-

Rg)

Rg

Fig 26 The sensitivity variation of a samples with pore diameters of22 nm 53 nm and 76 nm versus time and hydrogen concentrations

0 100 200 300 400 5001

10

100

Hydrogen concentration (ppm)

Sen

sitiv

ity

(R0-

Rgs

)R

gs

Fig 27 Sensitivity of a 20-V (76 nm) sample at 290 C for low hydro-gen concentrations

Since the sensor measurements were made in atmo-spheres without oxygen the increase in conductivitycannot be due to hydrogen removing oxygen from thelattice132ndash134 or the removal of chemisorbed oxygen135ndash137

Hydrogen molecules can dissociate at defects on thetitania surface which can subsequently diffuse into thetitania lattice acting as electron donors1281 1381 139 How-ever a dissociation-driven process would result in aslow response and recovery times with complete recov-ery unlikely since the sensor completely regains its orig-inal resistance with hydrogen cycling it appears that thisis not the dominant mechanism behind high hydrogensensitivity We believe chemisorption of the dissociatedhydrogen on the titania surface is the underlying sensingmechanism137 During chemisorption hydrogen acts as asurface state and a partial charge transfer from hydrogento the titania conduction band takes place This createsan electron accumulation layer on the nanotube surfaceenhancing its electrical conductance Upon removal of thehydrogen ambient electron transfer back to the hydro-gen molecule takes place which subsequently desorbsrestoring the original electrical resistance of the mate-rial Another factor that may play a role in the hydrogensensitivity (and selectivity) is the platinum electrodes Atelevated temperatures hydrogen dissociation can occur onplatinum surfaces These dissociated hydrogen atoms mayspill1361 140 onto the nanotube surface where they diffuse

10-1

101

103

105

0 05 1 15

Time (Hour)

Pore size - 76 nm

Pore size - 53 nm

Pore size - 22 nm

(Rg

R0)

-1

Fig 28 A comparison of the variation in resistance of samples havingpore diameters of 22 nm 53 nm and 76 nm versus time upon exposureto 1000 ppm of hydrogen at 290 C

290

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

10-2

10-1

100

0 100 2 102 4 102 6 102 8 102 1 103

OxygenAmmoniaCarbon monoxide

Time (Sec)

(Rg

R0)

-1

Fig 29 Variation of resistance of a 20-V sample when exposed to1000 ppm carbon monoxide 5 ammonia and 1 oxygen at 290 C

into the material affecting its electrical properties Fromthe present study it was not clear how signi cant a rolethe platinum electrodes play

Anatase the polymorph of titania has been reported tohave high sensitivity for reducing gases like hydrogen andcarbon monoxide1211 1241 133 Our nanotube samples containanatase phase mainly on the walls and rutile in the barrierlayer As the diffusing hydrogen atoms go to the intersti-tial sites1331 142 and as the ca ratio of anatase is almostfour times that of rutile it appears that the anatase latticeaccommodates hydrogen easily and hence makes a highercontribution to hydrogen sensitivity

The effect of chemisorption can be neglected in theoxygen-sensing experiments As the recovery requiresseveral hours it appears that the nanotubes contain oxy-gen vacancies or titanium interstitial defects in the pres-ence of nitrogen On exposure of the sensor to oxygenambient the lattice reoxidizes and hence the conductiv-ity of the sensor decreases On the removal of oxygenthe reduction of the lattice will not immediately occurhence the sensor requires several hours to regain its orig-inal conductivity

4 CONCLUSIONS

This paper summarizes different routes of fabrication ofmetal oxide nanoarchitectures useful for sensing applica-tions Structures made via anodization of a starting metaland breath gure formation from a sol-gel are discussedand presented as are TiO2 nanowires made by lling ananoporous alumina membrane with subsequent removalof the supporting template Two test cases illustratingthe utility of these nanoporous metal oxide structures forsensing are presented highly ordered nanoporous aluminais used for humidity sensing and TiO2 nanotubes are usedfor hydrogen sensing These study test cases reveal sev-eral points worth brie y noting

In the rst sensing test case uniform nanoporous alu-mina sensors exhibit pore size and frequency-dependent

humidity-sensing characteristics The mechanism respon-sible for the sensing action of the nanoporous alumina sen-sors was studied by representing the impedance spectra atdifferent humidity levels in Cole-Cole plots This makes itpossible to identify the competing impedance-determiningprocesses occurring in the sensors namely charge transferand mass transport through the physisorbed water layeras dependent upon the humidity level and input signalfrequency Charge transfer was found to enhance the sen-sitivity by causing a large variation in impedance withrespect to humidity Mass transport leads to accumulationof ions at the electrodes which tends to decrease the mea-surement sensitivity by increasing the impedance Hencea proper choice of operating frequency high enough toavoid ion accumulation is essential for optimum sensorperformance

It has been inferred from our results that anions becomeincorporated into the alumina lm during the anodiza-tion process supporting the hypothesis of Khanna andNahar104ndash106 where they act as proton donors and facili-tate the adsorption of water molecules Hence liquid-likecharge transfer networks are formed at relatively lowerhumidity levels for the smaller pore-size sensors Thisprocess rather than capillary condensation is primarilyresponsible for the observed humidity-sensing behavior ofthe sensors Our studies show that a distribution of poresizes as suggested by Shimizu et al94 is not necessary toobtain sensitive performance over a wide humidity rangerather such optimal performance can be obtained with uni-form nanodimensional pores containing adsorbed anionicimpurities

In the second sensing test case titania nanotubes pre-pared by anodization and annealed in an oxygen atmo-sphere at a temperature of 500 C were found to behighly sensitive to hydrogen The nanotube sensors con-tained both anatase and rutile phases of titania and showedappreciable sensitivity toward hydrogen at temperaturesas low as 180 C The sensitivity increased drasticallywith temperature showing a variation of 3 orders in mag-nitude of resistance to 1000 ppm of hydrogen at 400 CThe response time decreased with increasing temperatureat 290 C full switching of the sensor took approximately3 min Results were highly reproducible with no indica-tion of hysteresis Our results showed that these sensorsare at least capable of monitoring hydrogen levels from20 ppm to 4 At 290 C nanotubes with smaller porediameters showed higher sensitivity to hydrogen com-pared with their larger pore counterparts The sensorsshowed high selectivity to hydrogen compared with car-bon monoxide ammonia and carbon dioxide Althoughthe sensor was sensitive to high concentrations of oxy-gen the response time was high and the sensor did notcompletely regain the original condition We believe thehydrogen sensitivity of the nanotubes is due to hydrogendiffusion into the titania lattice where they act as electrondonors

291

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Acknowledgments Support of this work by the NationalScience Foundation through grant ECS-0210033 is grate-fully acknowledged

References and Notes

1 K Kajihara K Tanaka K Horao and N Soga Jpn J ApplPhys 35 6110 (1996)

2 B O Regan and M Gratzel A low cost Nature 353 737 (1991)3 A Fujishima and K Honda Nature 238 37 (1972)4 K Sato A Tsuzuki H Taoda Y Torii T Kato and Y Butsugan

J Mater Sci 29 5911 (1994)5 K Sato A Tsuzuki Y Torii H Taoda T Kato and Y Butsugan

J Mater Sci 30 837 (1995)6 N Ozer and C M Lampert Sol Energy Mater Sol Cells 54 147

(1998)7 A Manz and H Becker editors Microsystem Technology in

Chemistry and Life Sciences Springer-Verlag Berlin 19988 E Kim Y Xia and G M Whitesides Nature 376 581 (1995)9 M Templin A Franck A Du Chesne A Leist A Zhang

R Ulrich V Schadler and U Weisner Science 278 1795 (1997)10 D Zhao J Feng Q Huo N Melosh G H Frederickson

B Chmelka and G D Stucky Science 279 548 (1998)11 O D Velev T A Jede R F Lobo and A M Lenhoff Nature

389 447 (1997)12 B T Holland C F Blanford and A Stein Science 281 538

(1998)13 S Matsushita T Miwa and A Fujishima Chem Lett 9 925

(1997)14 A Imhof and D J Pine Adv Mater 10 697 (1998)15 T Tatsuma A Ikezawa Y Ohko T Miwa T Matsue and

A Fujishima Adv Mater 12 643 (2000)16 Z Zhong Y Yin B Gates and Y Xia Adv Mater 12 206 (2000)17 O D Velev and A M Lenhoff Curr Opin Colloids Interface

Sci 5 56 (2000)18 O D Velev and E W Kaler Adv Mater 12 531 (2000)19 P Jiang J Cizeron J F Bertone and V L Colvin J Am Chem

Soc 121 7957 (1999)20 M E Turner T J Trentler and V L Colvin Adv Mater 13 180

(2001)21 K M Kulinowski P Jiang H Vaswani and V L Colvin Adv

Mater 12 833 (2000)22 P Jiang J F Bertone and V L Colvin Science 291 453 (2001)23 R Seshadri and F C Meldrum Adv Mater 12 1149 (2000)24 F C Meldrum and R Seshadri Chem Commun 29 (2000)25 Q B Meng Z Z Gu O Sato and A Fujishima Appl Phys

Lett 77 4313 (2000)26 G Widawski B Rawiso and B Francois Nature 369 3897

(1994)27 B Francois O Pitois and J Francois Adv Mater 7 1041 (1995)28 O Pitois and B Francois Eur Phys J B 8 225 (1999)29 S A Jenekhe and X L Chen Science 283 372 (1999)30 M Shimomuara and T Sawadaishi Curr Opin Colloid Interface

Sci 6 11 (2001)31 O Karthaus N Maruyama X Cieren M Shimomura

H Hasegawa and T Hashimoto Langmuir 16 6071 (2000)32 T Nishikawa J Nishida R Ookura S Nishimura S Wada

T Karino and M Shimomura Mater Sci Eng C 10 141 (1999)33 K Kajihara K Nakanishi K Tanaka K Hirao and N Soga

J Am Ceram Soc 81 2670 (1998)34 C A Grimes R S Singh E C Dickey and O K Varghese

J Mater Res 16 1686 (2001)35 M Srinivasrao D Collings A Phillips and S Patel Science 292

79 (2001)36 P G de Gennes Rev Mod Phys 57 827 (1985)

37 D Beysens A Steyer P Guenoun D Fritter and C M KnoblerPhase Transitions 31 219 (1991)

38 B J Briscoe and K P Galvin J Phys D 23 422 (1990)39 B J Briscoe and K P Galvin J Phys D 23 1265 (1990)40 D Beysens and C M Knobler Phys Rev Lett 57 1433 (1986)41 F Family and P Meakin Phys Rev Lett 61 428 (1988)42 A V Limaye R D Narhe A M Dhote and S B Ogale Phys

Rev Lett 76 3762 (1996)43 P Dellrsquo Aversana and G P Neitzel Phys Today 51 38 (1998)44 P Dellrsquo Aversana J R Banavar and J Koplik Phys Fluids 8 15

(1986)45 P Dellrsquo Aversana V Tontodonato and L Carotenuto Phys Fluids

9 2475 (1997)46 M Q Li S Q Xu and E Kumacheva Langmuir 16 7275 (2000)47 P Freymuth Rev Sci Instrum 64 1 (1993)48 P Ball The Self-Made Tapestry Pattern Formation in Nature

Oxford University Press New York 199949 M F Schatz S J VanHook W D McCormick J B Swift and

H L Swinney Phys Fluids 11 2577 (1999)50 S R Coriell M R Cordes and W J Boettinger J Cryst Growth

49 13 (1980)51 T S Sullivan Y M Liu and R E Ecke Phys Rev E 54 486

(1996)52 J L Rogers M F Schatz J L Bougie and J B Swift 84 87

(2000)53 K Nitschke 52 R5772 (1995)54 E Bodenschatz J R De Bruyn G Ahlers and D S Cannell

Phys Rev Lett 67 3078 (1991)55 C W Meyer D S Cannell G Ahlers J B Swift and P C

Hohenberg Phys Rev Lett 61 947 (1988)56 R F Ismagilov D Rosmarin D H Gracias A D Stroock and

G M Whitesides Appl Phys Lett 79 439 (2001)57 E A Barringer and H K Bowen Langmuir 1 420 (1985)58 M Tschapek C Wasowski and R M T Sanchez J Electroanal

Chem 74 167 (1976)59 H Masuda and K Fukuda Science 268 1466 (1995)60 H Masuda F Hasegwa and S Ono J Electrochem Soc 144

L 127 (1997)61 O Jessensky F Muller and U Gosele Appl Phys Lett 72 1173

(1998)62 D Gong C A Grimes O K Varghese W Hu R S Singh

Z Chen and E C Dickey J Mater Res 16 3331 (2001)63 B M Kulwicki J Am Ceram Soc 74 697 (1991)64 H Arai and T Seiyama in Sensors A Comprehensive Survey

edited by W Gopel J Hesse and J N Zemel VCH Weinheim(1992) Vol 3 pp 981ndash1012

65 E Traversa Sens Actuators B 23 135 (1995)66 N Yamazoe and Y Shimizu Sens Actuators 10 379 (1986)67 A Bearzotti I Fratoddi L Palummo S Petrocco A Furlani

C Lo Sterzo and M V Russo Sens Actuators B 76 316 (2001)68 Y Sakai Y Sadaoka and M Matsuguchi Sens Actuators B 35

85 (1996)69 C A Grimes and D Kouzoudis Sens Actuators A 84 205

(2000)70 K G Ong C A Grimes C L Robbins and R S Singh Sens

Actuators A 93 33 (2001)71 T Seiyama N Yamazoe and H Arai Sens Actuators 4 85

(1983)72 L Ketron Ceram Bull 68 860 (1989)73 E Traversa G Gusmano and A Montenero Eur J Solid State

Inorg Chem 32 719 (1995)74 M K Jain M C Bhatnagar and G L Sharma Sens Actuators

B 55 180 (1999)75 O K Varghese and L K Malhotra J Appl Phys 87 7457 (2000)76 G Sberveglieri R Murri and N Pinto Sens Actuators B 23

177 (1995)

292

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

10

15

20

25

30

35

40

45

50

20 25 30 35 40 45 50

Res

pons

eR

ecov

ery

Tim

e (S

ec)

Pore diameter (nm)

DO15

SO30

SO50

Fig 14 Dependence of response time and recovery time on pore size

expressed in a general form as Cn4mdash5 D Bn4imdash5nƒ1 whereBn is a constant for a given set of experimental conditionsand n takes a value between 0 and 1 according to the localmicroscopic environment through which charge transporttakes place751 1021 103 Bn behaves as an ideal frequency-independent capacitor as n approaches 1 and as an idealconductor when the value of n approaches 0101

By examination of Figure 15a SO50 it can be seenthat at low humidity levels the Cole-Cole plot describesan arc with a very large radius of curvature As the humid-ity increases the radius decreases and at an RH of about80 the complete semicircle comes inside the measur-able impedance range However at the same humiditylevel a spur appears in the low-frequency region that dis-torts the semicircle The spur which can be viewed aspart of a second semicircle dominates at higher humid-ity levels while the semicircle described by the high-frequency impedance values diminishes in size Thisbehavior is similar for all sensors For sensors SO50SO30 and DO15 respectively the RH at which thehigh-frequency semicircle comes completely within themeasurable impedance range is ordm75 65 and 35these values roughly match the in ection points seen inFigure 11

The fundamental mechanism that enables ceramic sen-sors to sense humidity is the physisorption of watermolecules on an initially chemisorbed layer of hydroxylions651 96 The chemisorbed hydroxyl ions enhance theelectrical conductivity of the sensor either by donatingelectrons to the conduction band of the base materialor through proton hopping between adjacent hydroxylgroups upon the application of an electric eld Theprocess of chemisorption occurs at very low humiditylevels and is unaffected by further changes in humid-ity However an increase in humidity makes the watermolecules physisorb to this hydroxyl layer The effec-tiveness of physisorption depends upon the cation chargecomplexes (from the material or the impurities) and thehydroxyl ions (water molecules present at the surface ofthe base material)96 The physisorption process is facili-tated by higher surface charge densities During formationof the rst physisorbed layer a water molecule becomesattached to two neighboring hydroxyl groups through

0

02

04

06

08

1

12

0 02 04 06 08 1 12

Zrsquo (MW)

35 (a) 55 (a)

65 (a)

75 (b)

81 (b)

93 (b) SO50

0

02

04

06

08

1

12

14

0 02 04 06 08 1 12 14

Zrsquo (W)

Z

(W)

Z

(MW

)Z

(M

W)

35 (a)

55 (a)

65 (b)

75 (b)

81 (b)

93 (b)

SO30

0

2 105

4 105

6 105

8 105

1 106

12 106

0 2 105 4 105 6 105 8 105 1 106 12 106

17 (a)27 (a)

35 (b)45 (b)55 (b)

93 (b)

DO15

(a)

Zrsquo (MW)

(b)

(c)

Fig 15 Cole-Cole plot of the sensors at different humidity levels(a) SO50 (b) SO30 (c) DO15 The points represent the experimentaldata and lines represent the t with the equivalent circuit model givenin Figure 9 The RH values at which the data were collected and theequivalent circuit used for tting each curve (given as (a) or (b) denot-ing the circuit of Figure 17a and b respectively) are shown adjacent toeach curve

hydrogen double bonds and a proton may be transferredfrom a hydroxyl group to the water molecule to form aH3OC ion

When the physisorption occurs within less than a monolayer that is when clusters of physisorbed molecules

285

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

(b)

(a)

Cn1 (w)

R1

Cn2 (w) Cn1 (w)

R1 R2

Fig 16 Equivalent circuit used for tting the experimental data

are present at the surface it is thought that H3OC diffu-sion within hydroxyl groups and proton transfer betweenadjacent water molecules clusters take place96 HoweverKhanna and Nahar104ndash106 refute this theory They arguethat the proton diffusion is less probable in a surfacethat contains hydroxyl ions and clusters of physisorbedmolecules and hence cannot account for the conductiv-ity of porous alumina at an RH less than 40 that isbefore the rst monolayer is formed Instead they sug-gested a phonon-assisted electron tunneling mechanism atlow humidity levels in which electron tunneling occursfrom one water molecule to another as enhanced by thesurface anions of the alumina sensors

At higher humidity levels the number of physisorbedlayers increases allowing each water molecule to besingly bonded to a hydroxyl group and proton hop-ping between adjacent water molecules in the continu-ous water layer takes place The conduction process isthe same as that of pure water and is called the Grot-thuss chain reaction96 The dominant charge carrier ina high-humidity environment is therefore HC ions (pro-tons) The concentration of HC increases with increas-ing humidity and HC move freely through the water-likelayer According to the theory of Khanna and Nahar104

in porous alumina the protons for conduction are donatednot only by H2O but also by the Al(OH)3 formed asa result of water adsorption and the impurity anionsincorporated into the lm from the electrolyte duringanodization107ndash109 The water-like network consists ofsingly bonded water molecules which possess a highdielectric constant as they form dipoles and reorient freelyunder an externally applied electric eld The impedanceof the nanoporous alumina sensors includes this capaci-tive contribution as well

From the analysis of Figure 15 the high-frequencysemicircle that diminishes in size with increasing RHarises because of the effect of a physisorbed water layerat the surface Therefore R1 in the equivalent circuitof Figure 16 represents the charge transfer resistancethrough the adsorbed water layer The effect of electronic

102

104

106

108

SO50SO30

DO15

R1

(W)

0 20 40 60 80 100

RH ()

Fig 17 Humidity dependence of resistance R1 of different sensorsobtained by tting the equivalent circuit model with experimental data

conduction through the alumina bulk is parallel to thatof the surface conduction affected by the adsorbed waterlayer This bulk conduction process is dominant only atextremely low humidity values when there are only afew adsorbed molecules Hence the incomplete semicir-cle or arc in the low-humidity region is partially deter-mined from the base alumina Bn1 represents the capac-itance contribution from the adsorbed water layer Thelarge spurs seen in the low-frequency region are a con-sequence of ion migration in the adsorbed layer towardelectrodes1011 103 This migration leads to accumulation ofions at the electrodes Since the relaxation time of ionmigration is larger than the charge transfer process itis seen only at low frequencies The accumulated ionsin the adsorbed water layer give rise to R2 and Bn2From the equivalent circuits given in Figure 16 it canbe seen that the presence of ion accumulation repre-sented by the combination of R2 and Cn24mdash5 increasesthe overall impedance of the sensor Hence the reductionin impedance caused at high humidity due to the chargetransfer process is countered by ion accumulation

To better understand the fundamental sensing mecha-nisms R1 and Bn1 obtained through the numerical ttingprogram for the different sensors are plotted in Figures 17and 18 respectively It can be seen from Figure 17 thatR1 which is the resistance to charge transport in a staticelectric eld by the material containing an adsorbed waterlayer in the absence of ion migration is linear with RH

-11

10-10

10-9

10-8

10-7

50V30VDO15

Bn1

0 20 40 60 80 100

RH ()

Fig 18 Humidity dependence of Bn1 of different sensors obtained by tting the equivalent circuit model (Fig 17) with experimental data

286

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

in a semilog plot For sensor DO15 R1 decreases to afew hundred ohms at high RH levels indicating that theobserved conductivity is not due merely to the protonsprovided by the adsorbed water layer Since deionizedwater and argon gas are used to create the required humid-ity environment it is reasonable to assume that the sur-faces of the alumina lms contain anions presumablyfrom the electrolyte used for anodization which controlsconduction through the adsorbed layer108

With reference to Figure 18 Bn1 of the larger pore sam-ples SO50 and SO30 increases signi cantly only at thehigher humidity levels This result is in contrast to thatof the smaller pore size sensors which shows that Bn1

increases at lower humidity levels The relative humid-ity values where a signi cant increase in Bn1 takes placegenerally match the point seen in Figure 11 where theimpedance drops signi cantly which in turn is a func-tion of pore size These same relative humidity valuescorrelate with the point at which the spur appears inthe impedance diagrams given in Figure 15 The resultslead to the following conclusions Since hydrogen-bondedwater molecules possess a high dielectric constant theincrease in Bn1 indicates the presence of these bondson the surface which means that a liquid-like networkmore than one monolayer of physisorbed molecules hasformed at that humidity level This liquid-like network isresponsible for the sharp reduction in impedance seen inFigure 11 and the ion migration that creates the spur inthe low-frequency portion of the Cole-Cole plots as seenin Figure 15 According to Eq (1) capillary condensationdoes not take place in the alumina test sensors

In light of the above observations we propose thefollowing model to explain the behavior of nanoporousalumina in response to humidity The anodization pro-cess results in a certain amount of electrolyte anionstrapped within the pores1041 1071 108 Even in a low-humidityregime the presence of these anions on the sensor sur-face provides a high charge density for easy physisorp-tion of water molecules The result is the formationof a liquid-like network within the pores The impu-rity anions also act as proton donors by a dissociativemechanism104 joining the protons donated by the watermolecules The smaller the pore size the lower the RHlevel at which the liquid-like network is formed Hencefor the small pore sensors there is variation in Bn1 aswell as impedance magnitude in low-humidity regimesthe exponential dependence of R1 on RH also suggeststhe same phenomenon

In the case of DO15 the rst semicircle is highly dis-torted by the low-frequency spur originating with ionmigration A relatively large number of mobile ions mightbe present in these sensors leading to an enhancedaccumulation of protons at the electrode-sample contactAs mentioned earlier the dipoles in the singly bondedadsorbed water layer are free to rotate in an electric eld

Hence in the region close to the electrode-alumina inter-face the adsorbed water layer exhibits a net dipolar ori-entation and therefore it does not display its normaldielectric constant1011 103 Depending on ion size up toa monolayer of charge can exist at the electrode101 thediffuse (Gouy) layer of counter-ions spreads away fromthis inner (Stern) layer The number of ions available toform the Stern and Gouy layers at a particular frequencyincreases with increasing number of physisorbed layersHence as the number of physisorbed layers increases theStern and Gouy layers start forming at higher frequencies

With reference to Figure 12 the relative humiditylevel at which the sharp decrease in impedance occursshifts higher with increasing frequency This result can beexplained with the equivalent circuit shown in Figure 16athat is the circuit in the absence of ion accumulation Athigh frequencies the reactive impedance Xc of Cn14mdash5equal to Xc

D 1=jmdashCn14mdash5 is low and hence impedanceis largely determined by R1 As the humidity increasesR1 decreases however this translates into a reduction inimpedance only at relatively low frequencies where thevalue of R1 is comparable to or less than the capacitivereactance of Cn1(mdash) As the measurement frequency isreduced the reactance of Cn1(mdash) decreases hence the effectof R1 becomes more signi cant at lower humidity levels

32 Hydrogen Sensing with Titania Nanotubes

We recently reported62 the fabrication of self-organizedtitania nanotube arrays with an anodization techniqueAlthough the as-prepared nanotubes are amorphous theycrystallize on annealing at elevated temperatures and arestructurally stable to at least 600 C This stability ofstructure which is an essential criterion of a gas-sensingmaterial prompted us to study the gas-sensing behavior ofthese nanotubes for technologically important gases suchas oxygen carbon monoxide ammonia carbon dioxideand hydrogen

Titania has earned much attention for its oxygen-sensing capabilities110ndash115 Furthermore with propermanipulation of the microstructure or crystalline phaseandor the addition of the proper impurities or surfacefunctionalization titania can also be used as a reducinggas sensor116ndash127 The interaction of a gas with a metaloxide semiconductor is primarily a surface phenomenonand nanoporous metal oxides1221 1231 1281 129 offer the advan-tage of providing large sensing surface areas

Room-temperature metal oxide hydrogen sensors aregenerally based on Schottky barrier modulation of deviceslike PdTiO2 or PtTiO2 by hydrogen130ndash132 Elevated-temperature hydrogen sensors examine electrical resis-tance with hydrogen concentration for example Birke-feld et al133 observed that the resistance of the anatasephase of titania varies in the presence of carbon monoxideand hydrogen at temperatures above 500 C but on dop-ing with 10 alumina it becomes selective for hydrogen

287

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Barrier layer

Titanium Nano tube array

Platinum electrodes

Fig 19 Schematic representation of the hydrogen sensor geometry

Munuera et al134 used rhenium doping in the anatasephase to perform room-temperature hydrogen monitor-ing Recently Shimizu et al122 reported that anodizednanoporous titania lms with a Pd Schottky barrier aresensitive to hydrogen at 250 C

Titania nanotubes were grown from titanium foil(995 pure from Alfa Aesar Ward Hill MA) with athickness of 025 mm The anodization was performed inan electrolyte medium of 05 hydro uoric acid in waterwith the use of a platinum foil cathode Well-de nednanotube arrays were grown with anodizing potentialsranging from 12 V to 20 V Nanotube length increaseswith anodization time reaching 400 nm in approximately20 min and then remains constant For the present studythe samples were anodized for 25 min The samples werethen annealed at 500 C in a pure oxygen ambient for 6 hwith a heating and cooling rate of 1 Cmin

The electrode geometry of the titania nanotube sen-sors is shown in Figure 19 The sensor consists of a basetitanium metal foil with a nanotube array grown on topAn insulating barrier layer separates the nanotubes fromthe conducting titanium foil Electrical connections weremade by two parallel 10 mm pound 2 mm platinum pads of100-Œm thickness The test chamber consists of a 13-literquartz tube with stainless steel end caps placed insidea furnace (Thermolyne USA model 21100 tubular fur-nace) Gas ow through the test chamber was controlledvia a computer-controlled mass ow controller The elec-trical resistance of the titania sensors was measured witha computer-controlled Keithly digital multimeter Prior todata collection the test chamber was initially evacuatedwith a mechanical pump whereupon nitrogen (99999pure) was passed while the sensor under test was heatedto the desired temperature The test gases were mixed inappropriate ratios with nitrogen to create the necessarytest gas ambient

The surface morphology of nanotube arrays preparedwith an anodization potential of 20 V and annealed at500 C for 6 h in a pure oxygen ambient is shown inFigure 20a and b It can be seen from these images thatthe nanotubes are uniform over the surface The nano-tubes are approximately 400 nm in length and have abarrier layer62 thickness of ordm50 nm For the nanotubes

100 nm

(a)

(b)

1 mm (c)

100 nm Fig 20 The surface morphology of the titania nanotubes after anneal-ing at 500 C (a) High- and (b) low-magni cation images of a 20-Vsample (c) A high-magni cation image of a 12-V sample

fabricated with 20-V anodization the average pore diam-eter as determined from FESEM images is 76 nm (thestandard deviation 15 nm) with a wall thickness of 27 nm(standard deviation 6 nm) The sample anodized at 12 Vshown in Figure 2c was found to have an average porediameter of 46 nm (standard deviation 8 nm) with a wallthickness of 17 nm (standard deviation 2 nm) The porosi-ties of the 20-V and 12-V samples were calculated as 45and 61 respectively

Glancing-angle X-ray diffraction patterns of a 20-Vsample annealed at 500 C for 6 h in oxygen ambientis shown in Figure 21 It can be seen that both anataseand rutile phases of titania are present in the sampleA detailed study135 of these structures by high-resolutiontransmission electron microscope (HRTEM) showed that

288

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

20 30 40 50 60 70 80

2 q (deg)

(101

)

(110

)

A

R

TR

T

T

R R A TRR T R R T TT

Fig 21 Glancing-angle X-ray diffraction pattern of a 20-V sample(glancing angle D 2) annealed at 500 C in oxygen ambient A R andT represent respectively the re ections from anatase crystallites rutilecrystallites and the titanium substrate

the anatase crystallites were concentrated on the walls ofthe nanotubes and rutile on the barrier layer

Nanotubes annealed in a pure oxygen ambient werefound to be stable (intact) to temperatures of approxi-mately 580 C Above this temperature protrusions wereseen to be coming out through the nanotubes an effectthat spread with increasing temperature These protru-sions which are due to oxidation of the titanium substratecollapse the nanotubes

Figure 22 shows the response of the 20-V (76 nm) sam-ple as a function of ambient temperature as it is switchedfrom a nitrogen environment to one containing 1000 ppmhydrogen and then back to nitrogen The plot was madeusing 4Rg=R05

ƒ1 versus time where R0 is the base resis-tance of the sensor that is the sensor resistance before thetest gas was introduced and Rg is the measured resistancein the presence of test gas The sensor shows increasinghydrogen sensitivity with temperature with a change by3 orders of magnitude in resistance at temperatures aboveordm300 C At all of the temperatures the original resistancerecovered without hysteresis

The sensitivity S is de ned by the formula

S DR0

ƒ Rgs

Rgs

10 0

10 1

10 2

10 3

0 10 0 5 10 2 1 10 3 15 103 2 10 3 25 103 3 10 3 35 103 4 10 3

Time (Sec)

(Rg

R0)

-1

180degC

240degC

265degC

290degC

345degC

375degC

H2+N

2

N2

Fig 22 Variation of resistance Rg normalized with respect to baselineresistance R0 of a 20-V sample with time on exposure to 1000 ppmhydrogen at different temperatures It may be noted that the inverse ofRg=R0 was used in the plot to represent data in the positive y direction

10-1

100

101

102

103

150 200 250 300 350 400 450

Temperature (degC)

Sens

itivi

ty (

R0-

Rgs

)R

gs

Fig 23 The sensitivity temperature dependence of a 20-V sample to1000 ppm hydrogen

where R0 is the resistance of the sensor before the gas ispassed and Rgs is the resistance after gas is passed andreaches the saturation value The temperature-dependentsensitivity of a 20-V sample to 1000 ppm hydrogen isshown in Figure 23 Sensitivity is seen to increase withtemperature to approximately 380 C where the increasein sensitivity with temperature begins to saturate

The response time de ned as the time needed for thesensor to reach 90 of the nal signal for a given concen-tration of gas is plotted against temperature in Figure 24(the time includes that required for the gas to equili-brate inside the measurement chamber estimated to beordm30 s) The response time decreases exponentially withtemperature

To check the behavior of the sensor on repeated hydro-gen exposure the hydrogen concentration was varied indiscrete steps of 100 ppm from 0 to 500 ppm while thetemperature was kept constant at 290 C the chamber was ushed with nitrogen after each exposure to hydrogenThe response of the 20-V prepared nanotube sensorskept at 290 C is shown in Figure 25 The behaviorof the sensor is consistent recovering its original resis-tance after repeated exposure to varying hydrogen gasconcentrations The sensitivities of the 76-nm- 53-nm-and 22-nm-diameter nanotube sensors to hydrogen con-centrations ranging from 100 ppm to 1 are shown inFigure 26 The sensitivity of the 76-nm (20-V) sensor tolow concentrations is shown in Figure 27 there is a linearincrease in sensitivity at low concentrations The lower

10 1

10 2

10 3

10 4

150 200 250 300 350 400 450

Temperature (degC)

Res

pons

e T

ime

(Sec

)

Fig 24 Variation in response time for a 20-V sensor as a function oftemperature The dots represent measured data

289

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

10 2

0 4 4 4 4 4

10 3

0 10 1 10 2 10 3 10 4 10 5 10

Time (Sec)

Res

ista

nce

(W)

H2 (ppm)

200

300

400

500

400

300

200

100

100

Fig 25 Resistance of a 20-V sample when exposed to different con-centrations of hydrogen at 290 C The nitrogen-hydrogen mixture waspassed for 1500 s the chamber was then ushed with nitrogen for 3000 sbefore the nitrogen-hydrogen mixture was passed again

test limit of 20 ppm is determined by the experimentalapparatus and not the sensor

Figure 28 shows the change in resistance of the 76-nm-46-nm- and 22-nm-diameter nanotube sensors with expo-sure to 1000 ppm hydrogen at 290 C Whereas smallerdiameter nanotubes had greater sensitivity to hydrogenthe samples made at lower anodizing voltages tended tobe more brittle and harder to handle mechanically withoutbreaking

The 20-V 76-nm-diameter sample was exposed to oxy-gen carbon monoxide ammonia and carbon dioxide at290 C to study the cross-sensitivity The sensor wasfound to have no detectable variation in resistance onexposure to carbon dioxide The sensitivities of the tita-nia nanotubes to the other gases are shown in Figure 29The sensitivities of the nanotubes to carbon monoxide andammonia are negligible compared with that of hydrogenThe resistance of the nanotubes increased in the presenceof oxygen and did not regain their original electrical con-ductivity even after several hours in a 290 C nitrogenenvironment

1

10

100

1000

104

105

0 02 04 06 08 1

Hydrogen Concentration ()

Pore diameter = 76 nm

Pore diameter = 53 nm

Pore diameter = 22nm

Sen

sitiv

ity (

R0-

Rg)

Rg

Fig 26 The sensitivity variation of a samples with pore diameters of22 nm 53 nm and 76 nm versus time and hydrogen concentrations

0 100 200 300 400 5001

10

100

Hydrogen concentration (ppm)

Sen

sitiv

ity

(R0-

Rgs

)R

gs

Fig 27 Sensitivity of a 20-V (76 nm) sample at 290 C for low hydro-gen concentrations

Since the sensor measurements were made in atmo-spheres without oxygen the increase in conductivitycannot be due to hydrogen removing oxygen from thelattice132ndash134 or the removal of chemisorbed oxygen135ndash137

Hydrogen molecules can dissociate at defects on thetitania surface which can subsequently diffuse into thetitania lattice acting as electron donors1281 1381 139 How-ever a dissociation-driven process would result in aslow response and recovery times with complete recov-ery unlikely since the sensor completely regains its orig-inal resistance with hydrogen cycling it appears that thisis not the dominant mechanism behind high hydrogensensitivity We believe chemisorption of the dissociatedhydrogen on the titania surface is the underlying sensingmechanism137 During chemisorption hydrogen acts as asurface state and a partial charge transfer from hydrogento the titania conduction band takes place This createsan electron accumulation layer on the nanotube surfaceenhancing its electrical conductance Upon removal of thehydrogen ambient electron transfer back to the hydro-gen molecule takes place which subsequently desorbsrestoring the original electrical resistance of the mate-rial Another factor that may play a role in the hydrogensensitivity (and selectivity) is the platinum electrodes Atelevated temperatures hydrogen dissociation can occur onplatinum surfaces These dissociated hydrogen atoms mayspill1361 140 onto the nanotube surface where they diffuse

10-1

101

103

105

0 05 1 15

Time (Hour)

Pore size - 76 nm

Pore size - 53 nm

Pore size - 22 nm

(Rg

R0)

-1

Fig 28 A comparison of the variation in resistance of samples havingpore diameters of 22 nm 53 nm and 76 nm versus time upon exposureto 1000 ppm of hydrogen at 290 C

290

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

10-2

10-1

100

0 100 2 102 4 102 6 102 8 102 1 103

OxygenAmmoniaCarbon monoxide

Time (Sec)

(Rg

R0)

-1

Fig 29 Variation of resistance of a 20-V sample when exposed to1000 ppm carbon monoxide 5 ammonia and 1 oxygen at 290 C

into the material affecting its electrical properties Fromthe present study it was not clear how signi cant a rolethe platinum electrodes play

Anatase the polymorph of titania has been reported tohave high sensitivity for reducing gases like hydrogen andcarbon monoxide1211 1241 133 Our nanotube samples containanatase phase mainly on the walls and rutile in the barrierlayer As the diffusing hydrogen atoms go to the intersti-tial sites1331 142 and as the ca ratio of anatase is almostfour times that of rutile it appears that the anatase latticeaccommodates hydrogen easily and hence makes a highercontribution to hydrogen sensitivity

The effect of chemisorption can be neglected in theoxygen-sensing experiments As the recovery requiresseveral hours it appears that the nanotubes contain oxy-gen vacancies or titanium interstitial defects in the pres-ence of nitrogen On exposure of the sensor to oxygenambient the lattice reoxidizes and hence the conductiv-ity of the sensor decreases On the removal of oxygenthe reduction of the lattice will not immediately occurhence the sensor requires several hours to regain its orig-inal conductivity

4 CONCLUSIONS

This paper summarizes different routes of fabrication ofmetal oxide nanoarchitectures useful for sensing applica-tions Structures made via anodization of a starting metaland breath gure formation from a sol-gel are discussedand presented as are TiO2 nanowires made by lling ananoporous alumina membrane with subsequent removalof the supporting template Two test cases illustratingthe utility of these nanoporous metal oxide structures forsensing are presented highly ordered nanoporous aluminais used for humidity sensing and TiO2 nanotubes are usedfor hydrogen sensing These study test cases reveal sev-eral points worth brie y noting

In the rst sensing test case uniform nanoporous alu-mina sensors exhibit pore size and frequency-dependent

humidity-sensing characteristics The mechanism respon-sible for the sensing action of the nanoporous alumina sen-sors was studied by representing the impedance spectra atdifferent humidity levels in Cole-Cole plots This makes itpossible to identify the competing impedance-determiningprocesses occurring in the sensors namely charge transferand mass transport through the physisorbed water layeras dependent upon the humidity level and input signalfrequency Charge transfer was found to enhance the sen-sitivity by causing a large variation in impedance withrespect to humidity Mass transport leads to accumulationof ions at the electrodes which tends to decrease the mea-surement sensitivity by increasing the impedance Hencea proper choice of operating frequency high enough toavoid ion accumulation is essential for optimum sensorperformance

It has been inferred from our results that anions becomeincorporated into the alumina lm during the anodiza-tion process supporting the hypothesis of Khanna andNahar104ndash106 where they act as proton donors and facili-tate the adsorption of water molecules Hence liquid-likecharge transfer networks are formed at relatively lowerhumidity levels for the smaller pore-size sensors Thisprocess rather than capillary condensation is primarilyresponsible for the observed humidity-sensing behavior ofthe sensors Our studies show that a distribution of poresizes as suggested by Shimizu et al94 is not necessary toobtain sensitive performance over a wide humidity rangerather such optimal performance can be obtained with uni-form nanodimensional pores containing adsorbed anionicimpurities

In the second sensing test case titania nanotubes pre-pared by anodization and annealed in an oxygen atmo-sphere at a temperature of 500 C were found to behighly sensitive to hydrogen The nanotube sensors con-tained both anatase and rutile phases of titania and showedappreciable sensitivity toward hydrogen at temperaturesas low as 180 C The sensitivity increased drasticallywith temperature showing a variation of 3 orders in mag-nitude of resistance to 1000 ppm of hydrogen at 400 CThe response time decreased with increasing temperatureat 290 C full switching of the sensor took approximately3 min Results were highly reproducible with no indica-tion of hysteresis Our results showed that these sensorsare at least capable of monitoring hydrogen levels from20 ppm to 4 At 290 C nanotubes with smaller porediameters showed higher sensitivity to hydrogen com-pared with their larger pore counterparts The sensorsshowed high selectivity to hydrogen compared with car-bon monoxide ammonia and carbon dioxide Althoughthe sensor was sensitive to high concentrations of oxy-gen the response time was high and the sensor did notcompletely regain the original condition We believe thehydrogen sensitivity of the nanotubes is due to hydrogendiffusion into the titania lattice where they act as electrondonors

291

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Acknowledgments Support of this work by the NationalScience Foundation through grant ECS-0210033 is grate-fully acknowledged

References and Notes

1 K Kajihara K Tanaka K Horao and N Soga Jpn J ApplPhys 35 6110 (1996)

2 B O Regan and M Gratzel A low cost Nature 353 737 (1991)3 A Fujishima and K Honda Nature 238 37 (1972)4 K Sato A Tsuzuki H Taoda Y Torii T Kato and Y Butsugan

J Mater Sci 29 5911 (1994)5 K Sato A Tsuzuki Y Torii H Taoda T Kato and Y Butsugan

J Mater Sci 30 837 (1995)6 N Ozer and C M Lampert Sol Energy Mater Sol Cells 54 147

(1998)7 A Manz and H Becker editors Microsystem Technology in

Chemistry and Life Sciences Springer-Verlag Berlin 19988 E Kim Y Xia and G M Whitesides Nature 376 581 (1995)9 M Templin A Franck A Du Chesne A Leist A Zhang

R Ulrich V Schadler and U Weisner Science 278 1795 (1997)10 D Zhao J Feng Q Huo N Melosh G H Frederickson

B Chmelka and G D Stucky Science 279 548 (1998)11 O D Velev T A Jede R F Lobo and A M Lenhoff Nature

389 447 (1997)12 B T Holland C F Blanford and A Stein Science 281 538

(1998)13 S Matsushita T Miwa and A Fujishima Chem Lett 9 925

(1997)14 A Imhof and D J Pine Adv Mater 10 697 (1998)15 T Tatsuma A Ikezawa Y Ohko T Miwa T Matsue and

A Fujishima Adv Mater 12 643 (2000)16 Z Zhong Y Yin B Gates and Y Xia Adv Mater 12 206 (2000)17 O D Velev and A M Lenhoff Curr Opin Colloids Interface

Sci 5 56 (2000)18 O D Velev and E W Kaler Adv Mater 12 531 (2000)19 P Jiang J Cizeron J F Bertone and V L Colvin J Am Chem

Soc 121 7957 (1999)20 M E Turner T J Trentler and V L Colvin Adv Mater 13 180

(2001)21 K M Kulinowski P Jiang H Vaswani and V L Colvin Adv

Mater 12 833 (2000)22 P Jiang J F Bertone and V L Colvin Science 291 453 (2001)23 R Seshadri and F C Meldrum Adv Mater 12 1149 (2000)24 F C Meldrum and R Seshadri Chem Commun 29 (2000)25 Q B Meng Z Z Gu O Sato and A Fujishima Appl Phys

Lett 77 4313 (2000)26 G Widawski B Rawiso and B Francois Nature 369 3897

(1994)27 B Francois O Pitois and J Francois Adv Mater 7 1041 (1995)28 O Pitois and B Francois Eur Phys J B 8 225 (1999)29 S A Jenekhe and X L Chen Science 283 372 (1999)30 M Shimomuara and T Sawadaishi Curr Opin Colloid Interface

Sci 6 11 (2001)31 O Karthaus N Maruyama X Cieren M Shimomura

H Hasegawa and T Hashimoto Langmuir 16 6071 (2000)32 T Nishikawa J Nishida R Ookura S Nishimura S Wada

T Karino and M Shimomura Mater Sci Eng C 10 141 (1999)33 K Kajihara K Nakanishi K Tanaka K Hirao and N Soga

J Am Ceram Soc 81 2670 (1998)34 C A Grimes R S Singh E C Dickey and O K Varghese

J Mater Res 16 1686 (2001)35 M Srinivasrao D Collings A Phillips and S Patel Science 292

79 (2001)36 P G de Gennes Rev Mod Phys 57 827 (1985)

37 D Beysens A Steyer P Guenoun D Fritter and C M KnoblerPhase Transitions 31 219 (1991)

38 B J Briscoe and K P Galvin J Phys D 23 422 (1990)39 B J Briscoe and K P Galvin J Phys D 23 1265 (1990)40 D Beysens and C M Knobler Phys Rev Lett 57 1433 (1986)41 F Family and P Meakin Phys Rev Lett 61 428 (1988)42 A V Limaye R D Narhe A M Dhote and S B Ogale Phys

Rev Lett 76 3762 (1996)43 P Dellrsquo Aversana and G P Neitzel Phys Today 51 38 (1998)44 P Dellrsquo Aversana J R Banavar and J Koplik Phys Fluids 8 15

(1986)45 P Dellrsquo Aversana V Tontodonato and L Carotenuto Phys Fluids

9 2475 (1997)46 M Q Li S Q Xu and E Kumacheva Langmuir 16 7275 (2000)47 P Freymuth Rev Sci Instrum 64 1 (1993)48 P Ball The Self-Made Tapestry Pattern Formation in Nature

Oxford University Press New York 199949 M F Schatz S J VanHook W D McCormick J B Swift and

H L Swinney Phys Fluids 11 2577 (1999)50 S R Coriell M R Cordes and W J Boettinger J Cryst Growth

49 13 (1980)51 T S Sullivan Y M Liu and R E Ecke Phys Rev E 54 486

(1996)52 J L Rogers M F Schatz J L Bougie and J B Swift 84 87

(2000)53 K Nitschke 52 R5772 (1995)54 E Bodenschatz J R De Bruyn G Ahlers and D S Cannell

Phys Rev Lett 67 3078 (1991)55 C W Meyer D S Cannell G Ahlers J B Swift and P C

Hohenberg Phys Rev Lett 61 947 (1988)56 R F Ismagilov D Rosmarin D H Gracias A D Stroock and

G M Whitesides Appl Phys Lett 79 439 (2001)57 E A Barringer and H K Bowen Langmuir 1 420 (1985)58 M Tschapek C Wasowski and R M T Sanchez J Electroanal

Chem 74 167 (1976)59 H Masuda and K Fukuda Science 268 1466 (1995)60 H Masuda F Hasegwa and S Ono J Electrochem Soc 144

L 127 (1997)61 O Jessensky F Muller and U Gosele Appl Phys Lett 72 1173

(1998)62 D Gong C A Grimes O K Varghese W Hu R S Singh

Z Chen and E C Dickey J Mater Res 16 3331 (2001)63 B M Kulwicki J Am Ceram Soc 74 697 (1991)64 H Arai and T Seiyama in Sensors A Comprehensive Survey

edited by W Gopel J Hesse and J N Zemel VCH Weinheim(1992) Vol 3 pp 981ndash1012

65 E Traversa Sens Actuators B 23 135 (1995)66 N Yamazoe and Y Shimizu Sens Actuators 10 379 (1986)67 A Bearzotti I Fratoddi L Palummo S Petrocco A Furlani

C Lo Sterzo and M V Russo Sens Actuators B 76 316 (2001)68 Y Sakai Y Sadaoka and M Matsuguchi Sens Actuators B 35

85 (1996)69 C A Grimes and D Kouzoudis Sens Actuators A 84 205

(2000)70 K G Ong C A Grimes C L Robbins and R S Singh Sens

Actuators A 93 33 (2001)71 T Seiyama N Yamazoe and H Arai Sens Actuators 4 85

(1983)72 L Ketron Ceram Bull 68 860 (1989)73 E Traversa G Gusmano and A Montenero Eur J Solid State

Inorg Chem 32 719 (1995)74 M K Jain M C Bhatnagar and G L Sharma Sens Actuators

B 55 180 (1999)75 O K Varghese and L K Malhotra J Appl Phys 87 7457 (2000)76 G Sberveglieri R Murri and N Pinto Sens Actuators B 23

177 (1995)

292

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

(b)

(a)

Cn1 (w)

R1

Cn2 (w) Cn1 (w)

R1 R2

Fig 16 Equivalent circuit used for tting the experimental data

are present at the surface it is thought that H3OC diffu-sion within hydroxyl groups and proton transfer betweenadjacent water molecules clusters take place96 HoweverKhanna and Nahar104ndash106 refute this theory They arguethat the proton diffusion is less probable in a surfacethat contains hydroxyl ions and clusters of physisorbedmolecules and hence cannot account for the conductiv-ity of porous alumina at an RH less than 40 that isbefore the rst monolayer is formed Instead they sug-gested a phonon-assisted electron tunneling mechanism atlow humidity levels in which electron tunneling occursfrom one water molecule to another as enhanced by thesurface anions of the alumina sensors

At higher humidity levels the number of physisorbedlayers increases allowing each water molecule to besingly bonded to a hydroxyl group and proton hop-ping between adjacent water molecules in the continu-ous water layer takes place The conduction process isthe same as that of pure water and is called the Grot-thuss chain reaction96 The dominant charge carrier ina high-humidity environment is therefore HC ions (pro-tons) The concentration of HC increases with increas-ing humidity and HC move freely through the water-likelayer According to the theory of Khanna and Nahar104

in porous alumina the protons for conduction are donatednot only by H2O but also by the Al(OH)3 formed asa result of water adsorption and the impurity anionsincorporated into the lm from the electrolyte duringanodization107ndash109 The water-like network consists ofsingly bonded water molecules which possess a highdielectric constant as they form dipoles and reorient freelyunder an externally applied electric eld The impedanceof the nanoporous alumina sensors includes this capaci-tive contribution as well

From the analysis of Figure 15 the high-frequencysemicircle that diminishes in size with increasing RHarises because of the effect of a physisorbed water layerat the surface Therefore R1 in the equivalent circuitof Figure 16 represents the charge transfer resistancethrough the adsorbed water layer The effect of electronic

102

104

106

108

SO50SO30

DO15

R1

(W)

0 20 40 60 80 100

RH ()

Fig 17 Humidity dependence of resistance R1 of different sensorsobtained by tting the equivalent circuit model with experimental data

conduction through the alumina bulk is parallel to thatof the surface conduction affected by the adsorbed waterlayer This bulk conduction process is dominant only atextremely low humidity values when there are only afew adsorbed molecules Hence the incomplete semicir-cle or arc in the low-humidity region is partially deter-mined from the base alumina Bn1 represents the capac-itance contribution from the adsorbed water layer Thelarge spurs seen in the low-frequency region are a con-sequence of ion migration in the adsorbed layer towardelectrodes1011 103 This migration leads to accumulation ofions at the electrodes Since the relaxation time of ionmigration is larger than the charge transfer process itis seen only at low frequencies The accumulated ionsin the adsorbed water layer give rise to R2 and Bn2From the equivalent circuits given in Figure 16 it canbe seen that the presence of ion accumulation repre-sented by the combination of R2 and Cn24mdash5 increasesthe overall impedance of the sensor Hence the reductionin impedance caused at high humidity due to the chargetransfer process is countered by ion accumulation

To better understand the fundamental sensing mecha-nisms R1 and Bn1 obtained through the numerical ttingprogram for the different sensors are plotted in Figures 17and 18 respectively It can be seen from Figure 17 thatR1 which is the resistance to charge transport in a staticelectric eld by the material containing an adsorbed waterlayer in the absence of ion migration is linear with RH

-11

10-10

10-9

10-8

10-7

50V30VDO15

Bn1

0 20 40 60 80 100

RH ()

Fig 18 Humidity dependence of Bn1 of different sensors obtained by tting the equivalent circuit model (Fig 17) with experimental data

286

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

in a semilog plot For sensor DO15 R1 decreases to afew hundred ohms at high RH levels indicating that theobserved conductivity is not due merely to the protonsprovided by the adsorbed water layer Since deionizedwater and argon gas are used to create the required humid-ity environment it is reasonable to assume that the sur-faces of the alumina lms contain anions presumablyfrom the electrolyte used for anodization which controlsconduction through the adsorbed layer108

With reference to Figure 18 Bn1 of the larger pore sam-ples SO50 and SO30 increases signi cantly only at thehigher humidity levels This result is in contrast to thatof the smaller pore size sensors which shows that Bn1

increases at lower humidity levels The relative humid-ity values where a signi cant increase in Bn1 takes placegenerally match the point seen in Figure 11 where theimpedance drops signi cantly which in turn is a func-tion of pore size These same relative humidity valuescorrelate with the point at which the spur appears inthe impedance diagrams given in Figure 15 The resultslead to the following conclusions Since hydrogen-bondedwater molecules possess a high dielectric constant theincrease in Bn1 indicates the presence of these bondson the surface which means that a liquid-like networkmore than one monolayer of physisorbed molecules hasformed at that humidity level This liquid-like network isresponsible for the sharp reduction in impedance seen inFigure 11 and the ion migration that creates the spur inthe low-frequency portion of the Cole-Cole plots as seenin Figure 15 According to Eq (1) capillary condensationdoes not take place in the alumina test sensors

In light of the above observations we propose thefollowing model to explain the behavior of nanoporousalumina in response to humidity The anodization pro-cess results in a certain amount of electrolyte anionstrapped within the pores1041 1071 108 Even in a low-humidityregime the presence of these anions on the sensor sur-face provides a high charge density for easy physisorp-tion of water molecules The result is the formationof a liquid-like network within the pores The impu-rity anions also act as proton donors by a dissociativemechanism104 joining the protons donated by the watermolecules The smaller the pore size the lower the RHlevel at which the liquid-like network is formed Hencefor the small pore sensors there is variation in Bn1 aswell as impedance magnitude in low-humidity regimesthe exponential dependence of R1 on RH also suggeststhe same phenomenon

In the case of DO15 the rst semicircle is highly dis-torted by the low-frequency spur originating with ionmigration A relatively large number of mobile ions mightbe present in these sensors leading to an enhancedaccumulation of protons at the electrode-sample contactAs mentioned earlier the dipoles in the singly bondedadsorbed water layer are free to rotate in an electric eld

Hence in the region close to the electrode-alumina inter-face the adsorbed water layer exhibits a net dipolar ori-entation and therefore it does not display its normaldielectric constant1011 103 Depending on ion size up toa monolayer of charge can exist at the electrode101 thediffuse (Gouy) layer of counter-ions spreads away fromthis inner (Stern) layer The number of ions available toform the Stern and Gouy layers at a particular frequencyincreases with increasing number of physisorbed layersHence as the number of physisorbed layers increases theStern and Gouy layers start forming at higher frequencies

With reference to Figure 12 the relative humiditylevel at which the sharp decrease in impedance occursshifts higher with increasing frequency This result can beexplained with the equivalent circuit shown in Figure 16athat is the circuit in the absence of ion accumulation Athigh frequencies the reactive impedance Xc of Cn14mdash5equal to Xc

D 1=jmdashCn14mdash5 is low and hence impedanceis largely determined by R1 As the humidity increasesR1 decreases however this translates into a reduction inimpedance only at relatively low frequencies where thevalue of R1 is comparable to or less than the capacitivereactance of Cn1(mdash) As the measurement frequency isreduced the reactance of Cn1(mdash) decreases hence the effectof R1 becomes more signi cant at lower humidity levels

32 Hydrogen Sensing with Titania Nanotubes

We recently reported62 the fabrication of self-organizedtitania nanotube arrays with an anodization techniqueAlthough the as-prepared nanotubes are amorphous theycrystallize on annealing at elevated temperatures and arestructurally stable to at least 600 C This stability ofstructure which is an essential criterion of a gas-sensingmaterial prompted us to study the gas-sensing behavior ofthese nanotubes for technologically important gases suchas oxygen carbon monoxide ammonia carbon dioxideand hydrogen

Titania has earned much attention for its oxygen-sensing capabilities110ndash115 Furthermore with propermanipulation of the microstructure or crystalline phaseandor the addition of the proper impurities or surfacefunctionalization titania can also be used as a reducinggas sensor116ndash127 The interaction of a gas with a metaloxide semiconductor is primarily a surface phenomenonand nanoporous metal oxides1221 1231 1281 129 offer the advan-tage of providing large sensing surface areas

Room-temperature metal oxide hydrogen sensors aregenerally based on Schottky barrier modulation of deviceslike PdTiO2 or PtTiO2 by hydrogen130ndash132 Elevated-temperature hydrogen sensors examine electrical resis-tance with hydrogen concentration for example Birke-feld et al133 observed that the resistance of the anatasephase of titania varies in the presence of carbon monoxideand hydrogen at temperatures above 500 C but on dop-ing with 10 alumina it becomes selective for hydrogen

287

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Barrier layer

Titanium Nano tube array

Platinum electrodes

Fig 19 Schematic representation of the hydrogen sensor geometry

Munuera et al134 used rhenium doping in the anatasephase to perform room-temperature hydrogen monitor-ing Recently Shimizu et al122 reported that anodizednanoporous titania lms with a Pd Schottky barrier aresensitive to hydrogen at 250 C

Titania nanotubes were grown from titanium foil(995 pure from Alfa Aesar Ward Hill MA) with athickness of 025 mm The anodization was performed inan electrolyte medium of 05 hydro uoric acid in waterwith the use of a platinum foil cathode Well-de nednanotube arrays were grown with anodizing potentialsranging from 12 V to 20 V Nanotube length increaseswith anodization time reaching 400 nm in approximately20 min and then remains constant For the present studythe samples were anodized for 25 min The samples werethen annealed at 500 C in a pure oxygen ambient for 6 hwith a heating and cooling rate of 1 Cmin

The electrode geometry of the titania nanotube sen-sors is shown in Figure 19 The sensor consists of a basetitanium metal foil with a nanotube array grown on topAn insulating barrier layer separates the nanotubes fromthe conducting titanium foil Electrical connections weremade by two parallel 10 mm pound 2 mm platinum pads of100-Œm thickness The test chamber consists of a 13-literquartz tube with stainless steel end caps placed insidea furnace (Thermolyne USA model 21100 tubular fur-nace) Gas ow through the test chamber was controlledvia a computer-controlled mass ow controller The elec-trical resistance of the titania sensors was measured witha computer-controlled Keithly digital multimeter Prior todata collection the test chamber was initially evacuatedwith a mechanical pump whereupon nitrogen (99999pure) was passed while the sensor under test was heatedto the desired temperature The test gases were mixed inappropriate ratios with nitrogen to create the necessarytest gas ambient

The surface morphology of nanotube arrays preparedwith an anodization potential of 20 V and annealed at500 C for 6 h in a pure oxygen ambient is shown inFigure 20a and b It can be seen from these images thatthe nanotubes are uniform over the surface The nano-tubes are approximately 400 nm in length and have abarrier layer62 thickness of ordm50 nm For the nanotubes

100 nm

(a)

(b)

1 mm (c)

100 nm Fig 20 The surface morphology of the titania nanotubes after anneal-ing at 500 C (a) High- and (b) low-magni cation images of a 20-Vsample (c) A high-magni cation image of a 12-V sample

fabricated with 20-V anodization the average pore diam-eter as determined from FESEM images is 76 nm (thestandard deviation 15 nm) with a wall thickness of 27 nm(standard deviation 6 nm) The sample anodized at 12 Vshown in Figure 2c was found to have an average porediameter of 46 nm (standard deviation 8 nm) with a wallthickness of 17 nm (standard deviation 2 nm) The porosi-ties of the 20-V and 12-V samples were calculated as 45and 61 respectively

Glancing-angle X-ray diffraction patterns of a 20-Vsample annealed at 500 C for 6 h in oxygen ambientis shown in Figure 21 It can be seen that both anataseand rutile phases of titania are present in the sampleA detailed study135 of these structures by high-resolutiontransmission electron microscope (HRTEM) showed that

288

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

20 30 40 50 60 70 80

2 q (deg)

(101

)

(110

)

A

R

TR

T

T

R R A TRR T R R T TT

Fig 21 Glancing-angle X-ray diffraction pattern of a 20-V sample(glancing angle D 2) annealed at 500 C in oxygen ambient A R andT represent respectively the re ections from anatase crystallites rutilecrystallites and the titanium substrate

the anatase crystallites were concentrated on the walls ofthe nanotubes and rutile on the barrier layer

Nanotubes annealed in a pure oxygen ambient werefound to be stable (intact) to temperatures of approxi-mately 580 C Above this temperature protrusions wereseen to be coming out through the nanotubes an effectthat spread with increasing temperature These protru-sions which are due to oxidation of the titanium substratecollapse the nanotubes

Figure 22 shows the response of the 20-V (76 nm) sam-ple as a function of ambient temperature as it is switchedfrom a nitrogen environment to one containing 1000 ppmhydrogen and then back to nitrogen The plot was madeusing 4Rg=R05

ƒ1 versus time where R0 is the base resis-tance of the sensor that is the sensor resistance before thetest gas was introduced and Rg is the measured resistancein the presence of test gas The sensor shows increasinghydrogen sensitivity with temperature with a change by3 orders of magnitude in resistance at temperatures aboveordm300 C At all of the temperatures the original resistancerecovered without hysteresis

The sensitivity S is de ned by the formula

S DR0

ƒ Rgs

Rgs

10 0

10 1

10 2

10 3

0 10 0 5 10 2 1 10 3 15 103 2 10 3 25 103 3 10 3 35 103 4 10 3

Time (Sec)

(Rg

R0)

-1

180degC

240degC

265degC

290degC

345degC

375degC

H2+N

2

N2

Fig 22 Variation of resistance Rg normalized with respect to baselineresistance R0 of a 20-V sample with time on exposure to 1000 ppmhydrogen at different temperatures It may be noted that the inverse ofRg=R0 was used in the plot to represent data in the positive y direction

10-1

100

101

102

103

150 200 250 300 350 400 450

Temperature (degC)

Sens

itivi

ty (

R0-

Rgs

)R

gs

Fig 23 The sensitivity temperature dependence of a 20-V sample to1000 ppm hydrogen

where R0 is the resistance of the sensor before the gas ispassed and Rgs is the resistance after gas is passed andreaches the saturation value The temperature-dependentsensitivity of a 20-V sample to 1000 ppm hydrogen isshown in Figure 23 Sensitivity is seen to increase withtemperature to approximately 380 C where the increasein sensitivity with temperature begins to saturate

The response time de ned as the time needed for thesensor to reach 90 of the nal signal for a given concen-tration of gas is plotted against temperature in Figure 24(the time includes that required for the gas to equili-brate inside the measurement chamber estimated to beordm30 s) The response time decreases exponentially withtemperature

To check the behavior of the sensor on repeated hydro-gen exposure the hydrogen concentration was varied indiscrete steps of 100 ppm from 0 to 500 ppm while thetemperature was kept constant at 290 C the chamber was ushed with nitrogen after each exposure to hydrogenThe response of the 20-V prepared nanotube sensorskept at 290 C is shown in Figure 25 The behaviorof the sensor is consistent recovering its original resis-tance after repeated exposure to varying hydrogen gasconcentrations The sensitivities of the 76-nm- 53-nm-and 22-nm-diameter nanotube sensors to hydrogen con-centrations ranging from 100 ppm to 1 are shown inFigure 26 The sensitivity of the 76-nm (20-V) sensor tolow concentrations is shown in Figure 27 there is a linearincrease in sensitivity at low concentrations The lower

10 1

10 2

10 3

10 4

150 200 250 300 350 400 450

Temperature (degC)

Res

pons

e T

ime

(Sec

)

Fig 24 Variation in response time for a 20-V sensor as a function oftemperature The dots represent measured data

289

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

10 2

0 4 4 4 4 4

10 3

0 10 1 10 2 10 3 10 4 10 5 10

Time (Sec)

Res

ista

nce

(W)

H2 (ppm)

200

300

400

500

400

300

200

100

100

Fig 25 Resistance of a 20-V sample when exposed to different con-centrations of hydrogen at 290 C The nitrogen-hydrogen mixture waspassed for 1500 s the chamber was then ushed with nitrogen for 3000 sbefore the nitrogen-hydrogen mixture was passed again

test limit of 20 ppm is determined by the experimentalapparatus and not the sensor

Figure 28 shows the change in resistance of the 76-nm-46-nm- and 22-nm-diameter nanotube sensors with expo-sure to 1000 ppm hydrogen at 290 C Whereas smallerdiameter nanotubes had greater sensitivity to hydrogenthe samples made at lower anodizing voltages tended tobe more brittle and harder to handle mechanically withoutbreaking

The 20-V 76-nm-diameter sample was exposed to oxy-gen carbon monoxide ammonia and carbon dioxide at290 C to study the cross-sensitivity The sensor wasfound to have no detectable variation in resistance onexposure to carbon dioxide The sensitivities of the tita-nia nanotubes to the other gases are shown in Figure 29The sensitivities of the nanotubes to carbon monoxide andammonia are negligible compared with that of hydrogenThe resistance of the nanotubes increased in the presenceof oxygen and did not regain their original electrical con-ductivity even after several hours in a 290 C nitrogenenvironment

1

10

100

1000

104

105

0 02 04 06 08 1

Hydrogen Concentration ()

Pore diameter = 76 nm

Pore diameter = 53 nm

Pore diameter = 22nm

Sen

sitiv

ity (

R0-

Rg)

Rg

Fig 26 The sensitivity variation of a samples with pore diameters of22 nm 53 nm and 76 nm versus time and hydrogen concentrations

0 100 200 300 400 5001

10

100

Hydrogen concentration (ppm)

Sen

sitiv

ity

(R0-

Rgs

)R

gs

Fig 27 Sensitivity of a 20-V (76 nm) sample at 290 C for low hydro-gen concentrations

Since the sensor measurements were made in atmo-spheres without oxygen the increase in conductivitycannot be due to hydrogen removing oxygen from thelattice132ndash134 or the removal of chemisorbed oxygen135ndash137

Hydrogen molecules can dissociate at defects on thetitania surface which can subsequently diffuse into thetitania lattice acting as electron donors1281 1381 139 How-ever a dissociation-driven process would result in aslow response and recovery times with complete recov-ery unlikely since the sensor completely regains its orig-inal resistance with hydrogen cycling it appears that thisis not the dominant mechanism behind high hydrogensensitivity We believe chemisorption of the dissociatedhydrogen on the titania surface is the underlying sensingmechanism137 During chemisorption hydrogen acts as asurface state and a partial charge transfer from hydrogento the titania conduction band takes place This createsan electron accumulation layer on the nanotube surfaceenhancing its electrical conductance Upon removal of thehydrogen ambient electron transfer back to the hydro-gen molecule takes place which subsequently desorbsrestoring the original electrical resistance of the mate-rial Another factor that may play a role in the hydrogensensitivity (and selectivity) is the platinum electrodes Atelevated temperatures hydrogen dissociation can occur onplatinum surfaces These dissociated hydrogen atoms mayspill1361 140 onto the nanotube surface where they diffuse

10-1

101

103

105

0 05 1 15

Time (Hour)

Pore size - 76 nm

Pore size - 53 nm

Pore size - 22 nm

(Rg

R0)

-1

Fig 28 A comparison of the variation in resistance of samples havingpore diameters of 22 nm 53 nm and 76 nm versus time upon exposureto 1000 ppm of hydrogen at 290 C

290

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

10-2

10-1

100

0 100 2 102 4 102 6 102 8 102 1 103

OxygenAmmoniaCarbon monoxide

Time (Sec)

(Rg

R0)

-1

Fig 29 Variation of resistance of a 20-V sample when exposed to1000 ppm carbon monoxide 5 ammonia and 1 oxygen at 290 C

into the material affecting its electrical properties Fromthe present study it was not clear how signi cant a rolethe platinum electrodes play

Anatase the polymorph of titania has been reported tohave high sensitivity for reducing gases like hydrogen andcarbon monoxide1211 1241 133 Our nanotube samples containanatase phase mainly on the walls and rutile in the barrierlayer As the diffusing hydrogen atoms go to the intersti-tial sites1331 142 and as the ca ratio of anatase is almostfour times that of rutile it appears that the anatase latticeaccommodates hydrogen easily and hence makes a highercontribution to hydrogen sensitivity

The effect of chemisorption can be neglected in theoxygen-sensing experiments As the recovery requiresseveral hours it appears that the nanotubes contain oxy-gen vacancies or titanium interstitial defects in the pres-ence of nitrogen On exposure of the sensor to oxygenambient the lattice reoxidizes and hence the conductiv-ity of the sensor decreases On the removal of oxygenthe reduction of the lattice will not immediately occurhence the sensor requires several hours to regain its orig-inal conductivity

4 CONCLUSIONS

This paper summarizes different routes of fabrication ofmetal oxide nanoarchitectures useful for sensing applica-tions Structures made via anodization of a starting metaland breath gure formation from a sol-gel are discussedand presented as are TiO2 nanowires made by lling ananoporous alumina membrane with subsequent removalof the supporting template Two test cases illustratingthe utility of these nanoporous metal oxide structures forsensing are presented highly ordered nanoporous aluminais used for humidity sensing and TiO2 nanotubes are usedfor hydrogen sensing These study test cases reveal sev-eral points worth brie y noting

In the rst sensing test case uniform nanoporous alu-mina sensors exhibit pore size and frequency-dependent

humidity-sensing characteristics The mechanism respon-sible for the sensing action of the nanoporous alumina sen-sors was studied by representing the impedance spectra atdifferent humidity levels in Cole-Cole plots This makes itpossible to identify the competing impedance-determiningprocesses occurring in the sensors namely charge transferand mass transport through the physisorbed water layeras dependent upon the humidity level and input signalfrequency Charge transfer was found to enhance the sen-sitivity by causing a large variation in impedance withrespect to humidity Mass transport leads to accumulationof ions at the electrodes which tends to decrease the mea-surement sensitivity by increasing the impedance Hencea proper choice of operating frequency high enough toavoid ion accumulation is essential for optimum sensorperformance

It has been inferred from our results that anions becomeincorporated into the alumina lm during the anodiza-tion process supporting the hypothesis of Khanna andNahar104ndash106 where they act as proton donors and facili-tate the adsorption of water molecules Hence liquid-likecharge transfer networks are formed at relatively lowerhumidity levels for the smaller pore-size sensors Thisprocess rather than capillary condensation is primarilyresponsible for the observed humidity-sensing behavior ofthe sensors Our studies show that a distribution of poresizes as suggested by Shimizu et al94 is not necessary toobtain sensitive performance over a wide humidity rangerather such optimal performance can be obtained with uni-form nanodimensional pores containing adsorbed anionicimpurities

In the second sensing test case titania nanotubes pre-pared by anodization and annealed in an oxygen atmo-sphere at a temperature of 500 C were found to behighly sensitive to hydrogen The nanotube sensors con-tained both anatase and rutile phases of titania and showedappreciable sensitivity toward hydrogen at temperaturesas low as 180 C The sensitivity increased drasticallywith temperature showing a variation of 3 orders in mag-nitude of resistance to 1000 ppm of hydrogen at 400 CThe response time decreased with increasing temperatureat 290 C full switching of the sensor took approximately3 min Results were highly reproducible with no indica-tion of hysteresis Our results showed that these sensorsare at least capable of monitoring hydrogen levels from20 ppm to 4 At 290 C nanotubes with smaller porediameters showed higher sensitivity to hydrogen com-pared with their larger pore counterparts The sensorsshowed high selectivity to hydrogen compared with car-bon monoxide ammonia and carbon dioxide Althoughthe sensor was sensitive to high concentrations of oxy-gen the response time was high and the sensor did notcompletely regain the original condition We believe thehydrogen sensitivity of the nanotubes is due to hydrogendiffusion into the titania lattice where they act as electrondonors

291

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Acknowledgments Support of this work by the NationalScience Foundation through grant ECS-0210033 is grate-fully acknowledged

References and Notes

1 K Kajihara K Tanaka K Horao and N Soga Jpn J ApplPhys 35 6110 (1996)

2 B O Regan and M Gratzel A low cost Nature 353 737 (1991)3 A Fujishima and K Honda Nature 238 37 (1972)4 K Sato A Tsuzuki H Taoda Y Torii T Kato and Y Butsugan

J Mater Sci 29 5911 (1994)5 K Sato A Tsuzuki Y Torii H Taoda T Kato and Y Butsugan

J Mater Sci 30 837 (1995)6 N Ozer and C M Lampert Sol Energy Mater Sol Cells 54 147

(1998)7 A Manz and H Becker editors Microsystem Technology in

Chemistry and Life Sciences Springer-Verlag Berlin 19988 E Kim Y Xia and G M Whitesides Nature 376 581 (1995)9 M Templin A Franck A Du Chesne A Leist A Zhang

R Ulrich V Schadler and U Weisner Science 278 1795 (1997)10 D Zhao J Feng Q Huo N Melosh G H Frederickson

B Chmelka and G D Stucky Science 279 548 (1998)11 O D Velev T A Jede R F Lobo and A M Lenhoff Nature

389 447 (1997)12 B T Holland C F Blanford and A Stein Science 281 538

(1998)13 S Matsushita T Miwa and A Fujishima Chem Lett 9 925

(1997)14 A Imhof and D J Pine Adv Mater 10 697 (1998)15 T Tatsuma A Ikezawa Y Ohko T Miwa T Matsue and

A Fujishima Adv Mater 12 643 (2000)16 Z Zhong Y Yin B Gates and Y Xia Adv Mater 12 206 (2000)17 O D Velev and A M Lenhoff Curr Opin Colloids Interface

Sci 5 56 (2000)18 O D Velev and E W Kaler Adv Mater 12 531 (2000)19 P Jiang J Cizeron J F Bertone and V L Colvin J Am Chem

Soc 121 7957 (1999)20 M E Turner T J Trentler and V L Colvin Adv Mater 13 180

(2001)21 K M Kulinowski P Jiang H Vaswani and V L Colvin Adv

Mater 12 833 (2000)22 P Jiang J F Bertone and V L Colvin Science 291 453 (2001)23 R Seshadri and F C Meldrum Adv Mater 12 1149 (2000)24 F C Meldrum and R Seshadri Chem Commun 29 (2000)25 Q B Meng Z Z Gu O Sato and A Fujishima Appl Phys

Lett 77 4313 (2000)26 G Widawski B Rawiso and B Francois Nature 369 3897

(1994)27 B Francois O Pitois and J Francois Adv Mater 7 1041 (1995)28 O Pitois and B Francois Eur Phys J B 8 225 (1999)29 S A Jenekhe and X L Chen Science 283 372 (1999)30 M Shimomuara and T Sawadaishi Curr Opin Colloid Interface

Sci 6 11 (2001)31 O Karthaus N Maruyama X Cieren M Shimomura

H Hasegawa and T Hashimoto Langmuir 16 6071 (2000)32 T Nishikawa J Nishida R Ookura S Nishimura S Wada

T Karino and M Shimomura Mater Sci Eng C 10 141 (1999)33 K Kajihara K Nakanishi K Tanaka K Hirao and N Soga

J Am Ceram Soc 81 2670 (1998)34 C A Grimes R S Singh E C Dickey and O K Varghese

J Mater Res 16 1686 (2001)35 M Srinivasrao D Collings A Phillips and S Patel Science 292

79 (2001)36 P G de Gennes Rev Mod Phys 57 827 (1985)

37 D Beysens A Steyer P Guenoun D Fritter and C M KnoblerPhase Transitions 31 219 (1991)

38 B J Briscoe and K P Galvin J Phys D 23 422 (1990)39 B J Briscoe and K P Galvin J Phys D 23 1265 (1990)40 D Beysens and C M Knobler Phys Rev Lett 57 1433 (1986)41 F Family and P Meakin Phys Rev Lett 61 428 (1988)42 A V Limaye R D Narhe A M Dhote and S B Ogale Phys

Rev Lett 76 3762 (1996)43 P Dellrsquo Aversana and G P Neitzel Phys Today 51 38 (1998)44 P Dellrsquo Aversana J R Banavar and J Koplik Phys Fluids 8 15

(1986)45 P Dellrsquo Aversana V Tontodonato and L Carotenuto Phys Fluids

9 2475 (1997)46 M Q Li S Q Xu and E Kumacheva Langmuir 16 7275 (2000)47 P Freymuth Rev Sci Instrum 64 1 (1993)48 P Ball The Self-Made Tapestry Pattern Formation in Nature

Oxford University Press New York 199949 M F Schatz S J VanHook W D McCormick J B Swift and

H L Swinney Phys Fluids 11 2577 (1999)50 S R Coriell M R Cordes and W J Boettinger J Cryst Growth

49 13 (1980)51 T S Sullivan Y M Liu and R E Ecke Phys Rev E 54 486

(1996)52 J L Rogers M F Schatz J L Bougie and J B Swift 84 87

(2000)53 K Nitschke 52 R5772 (1995)54 E Bodenschatz J R De Bruyn G Ahlers and D S Cannell

Phys Rev Lett 67 3078 (1991)55 C W Meyer D S Cannell G Ahlers J B Swift and P C

Hohenberg Phys Rev Lett 61 947 (1988)56 R F Ismagilov D Rosmarin D H Gracias A D Stroock and

G M Whitesides Appl Phys Lett 79 439 (2001)57 E A Barringer and H K Bowen Langmuir 1 420 (1985)58 M Tschapek C Wasowski and R M T Sanchez J Electroanal

Chem 74 167 (1976)59 H Masuda and K Fukuda Science 268 1466 (1995)60 H Masuda F Hasegwa and S Ono J Electrochem Soc 144

L 127 (1997)61 O Jessensky F Muller and U Gosele Appl Phys Lett 72 1173

(1998)62 D Gong C A Grimes O K Varghese W Hu R S Singh

Z Chen and E C Dickey J Mater Res 16 3331 (2001)63 B M Kulwicki J Am Ceram Soc 74 697 (1991)64 H Arai and T Seiyama in Sensors A Comprehensive Survey

edited by W Gopel J Hesse and J N Zemel VCH Weinheim(1992) Vol 3 pp 981ndash1012

65 E Traversa Sens Actuators B 23 135 (1995)66 N Yamazoe and Y Shimizu Sens Actuators 10 379 (1986)67 A Bearzotti I Fratoddi L Palummo S Petrocco A Furlani

C Lo Sterzo and M V Russo Sens Actuators B 76 316 (2001)68 Y Sakai Y Sadaoka and M Matsuguchi Sens Actuators B 35

85 (1996)69 C A Grimes and D Kouzoudis Sens Actuators A 84 205

(2000)70 K G Ong C A Grimes C L Robbins and R S Singh Sens

Actuators A 93 33 (2001)71 T Seiyama N Yamazoe and H Arai Sens Actuators 4 85

(1983)72 L Ketron Ceram Bull 68 860 (1989)73 E Traversa G Gusmano and A Montenero Eur J Solid State

Inorg Chem 32 719 (1995)74 M K Jain M C Bhatnagar and G L Sharma Sens Actuators

B 55 180 (1999)75 O K Varghese and L K Malhotra J Appl Phys 87 7457 (2000)76 G Sberveglieri R Murri and N Pinto Sens Actuators B 23

177 (1995)

292

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

in a semilog plot For sensor DO15 R1 decreases to afew hundred ohms at high RH levels indicating that theobserved conductivity is not due merely to the protonsprovided by the adsorbed water layer Since deionizedwater and argon gas are used to create the required humid-ity environment it is reasonable to assume that the sur-faces of the alumina lms contain anions presumablyfrom the electrolyte used for anodization which controlsconduction through the adsorbed layer108

With reference to Figure 18 Bn1 of the larger pore sam-ples SO50 and SO30 increases signi cantly only at thehigher humidity levels This result is in contrast to thatof the smaller pore size sensors which shows that Bn1

increases at lower humidity levels The relative humid-ity values where a signi cant increase in Bn1 takes placegenerally match the point seen in Figure 11 where theimpedance drops signi cantly which in turn is a func-tion of pore size These same relative humidity valuescorrelate with the point at which the spur appears inthe impedance diagrams given in Figure 15 The resultslead to the following conclusions Since hydrogen-bondedwater molecules possess a high dielectric constant theincrease in Bn1 indicates the presence of these bondson the surface which means that a liquid-like networkmore than one monolayer of physisorbed molecules hasformed at that humidity level This liquid-like network isresponsible for the sharp reduction in impedance seen inFigure 11 and the ion migration that creates the spur inthe low-frequency portion of the Cole-Cole plots as seenin Figure 15 According to Eq (1) capillary condensationdoes not take place in the alumina test sensors

In light of the above observations we propose thefollowing model to explain the behavior of nanoporousalumina in response to humidity The anodization pro-cess results in a certain amount of electrolyte anionstrapped within the pores1041 1071 108 Even in a low-humidityregime the presence of these anions on the sensor sur-face provides a high charge density for easy physisorp-tion of water molecules The result is the formationof a liquid-like network within the pores The impu-rity anions also act as proton donors by a dissociativemechanism104 joining the protons donated by the watermolecules The smaller the pore size the lower the RHlevel at which the liquid-like network is formed Hencefor the small pore sensors there is variation in Bn1 aswell as impedance magnitude in low-humidity regimesthe exponential dependence of R1 on RH also suggeststhe same phenomenon

In the case of DO15 the rst semicircle is highly dis-torted by the low-frequency spur originating with ionmigration A relatively large number of mobile ions mightbe present in these sensors leading to an enhancedaccumulation of protons at the electrode-sample contactAs mentioned earlier the dipoles in the singly bondedadsorbed water layer are free to rotate in an electric eld

Hence in the region close to the electrode-alumina inter-face the adsorbed water layer exhibits a net dipolar ori-entation and therefore it does not display its normaldielectric constant1011 103 Depending on ion size up toa monolayer of charge can exist at the electrode101 thediffuse (Gouy) layer of counter-ions spreads away fromthis inner (Stern) layer The number of ions available toform the Stern and Gouy layers at a particular frequencyincreases with increasing number of physisorbed layersHence as the number of physisorbed layers increases theStern and Gouy layers start forming at higher frequencies

With reference to Figure 12 the relative humiditylevel at which the sharp decrease in impedance occursshifts higher with increasing frequency This result can beexplained with the equivalent circuit shown in Figure 16athat is the circuit in the absence of ion accumulation Athigh frequencies the reactive impedance Xc of Cn14mdash5equal to Xc

D 1=jmdashCn14mdash5 is low and hence impedanceis largely determined by R1 As the humidity increasesR1 decreases however this translates into a reduction inimpedance only at relatively low frequencies where thevalue of R1 is comparable to or less than the capacitivereactance of Cn1(mdash) As the measurement frequency isreduced the reactance of Cn1(mdash) decreases hence the effectof R1 becomes more signi cant at lower humidity levels

32 Hydrogen Sensing with Titania Nanotubes

We recently reported62 the fabrication of self-organizedtitania nanotube arrays with an anodization techniqueAlthough the as-prepared nanotubes are amorphous theycrystallize on annealing at elevated temperatures and arestructurally stable to at least 600 C This stability ofstructure which is an essential criterion of a gas-sensingmaterial prompted us to study the gas-sensing behavior ofthese nanotubes for technologically important gases suchas oxygen carbon monoxide ammonia carbon dioxideand hydrogen

Titania has earned much attention for its oxygen-sensing capabilities110ndash115 Furthermore with propermanipulation of the microstructure or crystalline phaseandor the addition of the proper impurities or surfacefunctionalization titania can also be used as a reducinggas sensor116ndash127 The interaction of a gas with a metaloxide semiconductor is primarily a surface phenomenonand nanoporous metal oxides1221 1231 1281 129 offer the advan-tage of providing large sensing surface areas

Room-temperature metal oxide hydrogen sensors aregenerally based on Schottky barrier modulation of deviceslike PdTiO2 or PtTiO2 by hydrogen130ndash132 Elevated-temperature hydrogen sensors examine electrical resis-tance with hydrogen concentration for example Birke-feld et al133 observed that the resistance of the anatasephase of titania varies in the presence of carbon monoxideand hydrogen at temperatures above 500 C but on dop-ing with 10 alumina it becomes selective for hydrogen

287

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Barrier layer

Titanium Nano tube array

Platinum electrodes

Fig 19 Schematic representation of the hydrogen sensor geometry

Munuera et al134 used rhenium doping in the anatasephase to perform room-temperature hydrogen monitor-ing Recently Shimizu et al122 reported that anodizednanoporous titania lms with a Pd Schottky barrier aresensitive to hydrogen at 250 C

Titania nanotubes were grown from titanium foil(995 pure from Alfa Aesar Ward Hill MA) with athickness of 025 mm The anodization was performed inan electrolyte medium of 05 hydro uoric acid in waterwith the use of a platinum foil cathode Well-de nednanotube arrays were grown with anodizing potentialsranging from 12 V to 20 V Nanotube length increaseswith anodization time reaching 400 nm in approximately20 min and then remains constant For the present studythe samples were anodized for 25 min The samples werethen annealed at 500 C in a pure oxygen ambient for 6 hwith a heating and cooling rate of 1 Cmin

The electrode geometry of the titania nanotube sen-sors is shown in Figure 19 The sensor consists of a basetitanium metal foil with a nanotube array grown on topAn insulating barrier layer separates the nanotubes fromthe conducting titanium foil Electrical connections weremade by two parallel 10 mm pound 2 mm platinum pads of100-Œm thickness The test chamber consists of a 13-literquartz tube with stainless steel end caps placed insidea furnace (Thermolyne USA model 21100 tubular fur-nace) Gas ow through the test chamber was controlledvia a computer-controlled mass ow controller The elec-trical resistance of the titania sensors was measured witha computer-controlled Keithly digital multimeter Prior todata collection the test chamber was initially evacuatedwith a mechanical pump whereupon nitrogen (99999pure) was passed while the sensor under test was heatedto the desired temperature The test gases were mixed inappropriate ratios with nitrogen to create the necessarytest gas ambient

The surface morphology of nanotube arrays preparedwith an anodization potential of 20 V and annealed at500 C for 6 h in a pure oxygen ambient is shown inFigure 20a and b It can be seen from these images thatthe nanotubes are uniform over the surface The nano-tubes are approximately 400 nm in length and have abarrier layer62 thickness of ordm50 nm For the nanotubes

100 nm

(a)

(b)

1 mm (c)

100 nm Fig 20 The surface morphology of the titania nanotubes after anneal-ing at 500 C (a) High- and (b) low-magni cation images of a 20-Vsample (c) A high-magni cation image of a 12-V sample

fabricated with 20-V anodization the average pore diam-eter as determined from FESEM images is 76 nm (thestandard deviation 15 nm) with a wall thickness of 27 nm(standard deviation 6 nm) The sample anodized at 12 Vshown in Figure 2c was found to have an average porediameter of 46 nm (standard deviation 8 nm) with a wallthickness of 17 nm (standard deviation 2 nm) The porosi-ties of the 20-V and 12-V samples were calculated as 45and 61 respectively

Glancing-angle X-ray diffraction patterns of a 20-Vsample annealed at 500 C for 6 h in oxygen ambientis shown in Figure 21 It can be seen that both anataseand rutile phases of titania are present in the sampleA detailed study135 of these structures by high-resolutiontransmission electron microscope (HRTEM) showed that

288

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

20 30 40 50 60 70 80

2 q (deg)

(101

)

(110

)

A

R

TR

T

T

R R A TRR T R R T TT

Fig 21 Glancing-angle X-ray diffraction pattern of a 20-V sample(glancing angle D 2) annealed at 500 C in oxygen ambient A R andT represent respectively the re ections from anatase crystallites rutilecrystallites and the titanium substrate

the anatase crystallites were concentrated on the walls ofthe nanotubes and rutile on the barrier layer

Nanotubes annealed in a pure oxygen ambient werefound to be stable (intact) to temperatures of approxi-mately 580 C Above this temperature protrusions wereseen to be coming out through the nanotubes an effectthat spread with increasing temperature These protru-sions which are due to oxidation of the titanium substratecollapse the nanotubes

Figure 22 shows the response of the 20-V (76 nm) sam-ple as a function of ambient temperature as it is switchedfrom a nitrogen environment to one containing 1000 ppmhydrogen and then back to nitrogen The plot was madeusing 4Rg=R05

ƒ1 versus time where R0 is the base resis-tance of the sensor that is the sensor resistance before thetest gas was introduced and Rg is the measured resistancein the presence of test gas The sensor shows increasinghydrogen sensitivity with temperature with a change by3 orders of magnitude in resistance at temperatures aboveordm300 C At all of the temperatures the original resistancerecovered without hysteresis

The sensitivity S is de ned by the formula

S DR0

ƒ Rgs

Rgs

10 0

10 1

10 2

10 3

0 10 0 5 10 2 1 10 3 15 103 2 10 3 25 103 3 10 3 35 103 4 10 3

Time (Sec)

(Rg

R0)

-1

180degC

240degC

265degC

290degC

345degC

375degC

H2+N

2

N2

Fig 22 Variation of resistance Rg normalized with respect to baselineresistance R0 of a 20-V sample with time on exposure to 1000 ppmhydrogen at different temperatures It may be noted that the inverse ofRg=R0 was used in the plot to represent data in the positive y direction

10-1

100

101

102

103

150 200 250 300 350 400 450

Temperature (degC)

Sens

itivi

ty (

R0-

Rgs

)R

gs

Fig 23 The sensitivity temperature dependence of a 20-V sample to1000 ppm hydrogen

where R0 is the resistance of the sensor before the gas ispassed and Rgs is the resistance after gas is passed andreaches the saturation value The temperature-dependentsensitivity of a 20-V sample to 1000 ppm hydrogen isshown in Figure 23 Sensitivity is seen to increase withtemperature to approximately 380 C where the increasein sensitivity with temperature begins to saturate

The response time de ned as the time needed for thesensor to reach 90 of the nal signal for a given concen-tration of gas is plotted against temperature in Figure 24(the time includes that required for the gas to equili-brate inside the measurement chamber estimated to beordm30 s) The response time decreases exponentially withtemperature

To check the behavior of the sensor on repeated hydro-gen exposure the hydrogen concentration was varied indiscrete steps of 100 ppm from 0 to 500 ppm while thetemperature was kept constant at 290 C the chamber was ushed with nitrogen after each exposure to hydrogenThe response of the 20-V prepared nanotube sensorskept at 290 C is shown in Figure 25 The behaviorof the sensor is consistent recovering its original resis-tance after repeated exposure to varying hydrogen gasconcentrations The sensitivities of the 76-nm- 53-nm-and 22-nm-diameter nanotube sensors to hydrogen con-centrations ranging from 100 ppm to 1 are shown inFigure 26 The sensitivity of the 76-nm (20-V) sensor tolow concentrations is shown in Figure 27 there is a linearincrease in sensitivity at low concentrations The lower

10 1

10 2

10 3

10 4

150 200 250 300 350 400 450

Temperature (degC)

Res

pons

e T

ime

(Sec

)

Fig 24 Variation in response time for a 20-V sensor as a function oftemperature The dots represent measured data

289

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

10 2

0 4 4 4 4 4

10 3

0 10 1 10 2 10 3 10 4 10 5 10

Time (Sec)

Res

ista

nce

(W)

H2 (ppm)

200

300

400

500

400

300

200

100

100

Fig 25 Resistance of a 20-V sample when exposed to different con-centrations of hydrogen at 290 C The nitrogen-hydrogen mixture waspassed for 1500 s the chamber was then ushed with nitrogen for 3000 sbefore the nitrogen-hydrogen mixture was passed again

test limit of 20 ppm is determined by the experimentalapparatus and not the sensor

Figure 28 shows the change in resistance of the 76-nm-46-nm- and 22-nm-diameter nanotube sensors with expo-sure to 1000 ppm hydrogen at 290 C Whereas smallerdiameter nanotubes had greater sensitivity to hydrogenthe samples made at lower anodizing voltages tended tobe more brittle and harder to handle mechanically withoutbreaking

The 20-V 76-nm-diameter sample was exposed to oxy-gen carbon monoxide ammonia and carbon dioxide at290 C to study the cross-sensitivity The sensor wasfound to have no detectable variation in resistance onexposure to carbon dioxide The sensitivities of the tita-nia nanotubes to the other gases are shown in Figure 29The sensitivities of the nanotubes to carbon monoxide andammonia are negligible compared with that of hydrogenThe resistance of the nanotubes increased in the presenceof oxygen and did not regain their original electrical con-ductivity even after several hours in a 290 C nitrogenenvironment

1

10

100

1000

104

105

0 02 04 06 08 1

Hydrogen Concentration ()

Pore diameter = 76 nm

Pore diameter = 53 nm

Pore diameter = 22nm

Sen

sitiv

ity (

R0-

Rg)

Rg

Fig 26 The sensitivity variation of a samples with pore diameters of22 nm 53 nm and 76 nm versus time and hydrogen concentrations

0 100 200 300 400 5001

10

100

Hydrogen concentration (ppm)

Sen

sitiv

ity

(R0-

Rgs

)R

gs

Fig 27 Sensitivity of a 20-V (76 nm) sample at 290 C for low hydro-gen concentrations

Since the sensor measurements were made in atmo-spheres without oxygen the increase in conductivitycannot be due to hydrogen removing oxygen from thelattice132ndash134 or the removal of chemisorbed oxygen135ndash137

Hydrogen molecules can dissociate at defects on thetitania surface which can subsequently diffuse into thetitania lattice acting as electron donors1281 1381 139 How-ever a dissociation-driven process would result in aslow response and recovery times with complete recov-ery unlikely since the sensor completely regains its orig-inal resistance with hydrogen cycling it appears that thisis not the dominant mechanism behind high hydrogensensitivity We believe chemisorption of the dissociatedhydrogen on the titania surface is the underlying sensingmechanism137 During chemisorption hydrogen acts as asurface state and a partial charge transfer from hydrogento the titania conduction band takes place This createsan electron accumulation layer on the nanotube surfaceenhancing its electrical conductance Upon removal of thehydrogen ambient electron transfer back to the hydro-gen molecule takes place which subsequently desorbsrestoring the original electrical resistance of the mate-rial Another factor that may play a role in the hydrogensensitivity (and selectivity) is the platinum electrodes Atelevated temperatures hydrogen dissociation can occur onplatinum surfaces These dissociated hydrogen atoms mayspill1361 140 onto the nanotube surface where they diffuse

10-1

101

103

105

0 05 1 15

Time (Hour)

Pore size - 76 nm

Pore size - 53 nm

Pore size - 22 nm

(Rg

R0)

-1

Fig 28 A comparison of the variation in resistance of samples havingpore diameters of 22 nm 53 nm and 76 nm versus time upon exposureto 1000 ppm of hydrogen at 290 C

290

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

10-2

10-1

100

0 100 2 102 4 102 6 102 8 102 1 103

OxygenAmmoniaCarbon monoxide

Time (Sec)

(Rg

R0)

-1

Fig 29 Variation of resistance of a 20-V sample when exposed to1000 ppm carbon monoxide 5 ammonia and 1 oxygen at 290 C

into the material affecting its electrical properties Fromthe present study it was not clear how signi cant a rolethe platinum electrodes play

Anatase the polymorph of titania has been reported tohave high sensitivity for reducing gases like hydrogen andcarbon monoxide1211 1241 133 Our nanotube samples containanatase phase mainly on the walls and rutile in the barrierlayer As the diffusing hydrogen atoms go to the intersti-tial sites1331 142 and as the ca ratio of anatase is almostfour times that of rutile it appears that the anatase latticeaccommodates hydrogen easily and hence makes a highercontribution to hydrogen sensitivity

The effect of chemisorption can be neglected in theoxygen-sensing experiments As the recovery requiresseveral hours it appears that the nanotubes contain oxy-gen vacancies or titanium interstitial defects in the pres-ence of nitrogen On exposure of the sensor to oxygenambient the lattice reoxidizes and hence the conductiv-ity of the sensor decreases On the removal of oxygenthe reduction of the lattice will not immediately occurhence the sensor requires several hours to regain its orig-inal conductivity

4 CONCLUSIONS

This paper summarizes different routes of fabrication ofmetal oxide nanoarchitectures useful for sensing applica-tions Structures made via anodization of a starting metaland breath gure formation from a sol-gel are discussedand presented as are TiO2 nanowires made by lling ananoporous alumina membrane with subsequent removalof the supporting template Two test cases illustratingthe utility of these nanoporous metal oxide structures forsensing are presented highly ordered nanoporous aluminais used for humidity sensing and TiO2 nanotubes are usedfor hydrogen sensing These study test cases reveal sev-eral points worth brie y noting

In the rst sensing test case uniform nanoporous alu-mina sensors exhibit pore size and frequency-dependent

humidity-sensing characteristics The mechanism respon-sible for the sensing action of the nanoporous alumina sen-sors was studied by representing the impedance spectra atdifferent humidity levels in Cole-Cole plots This makes itpossible to identify the competing impedance-determiningprocesses occurring in the sensors namely charge transferand mass transport through the physisorbed water layeras dependent upon the humidity level and input signalfrequency Charge transfer was found to enhance the sen-sitivity by causing a large variation in impedance withrespect to humidity Mass transport leads to accumulationof ions at the electrodes which tends to decrease the mea-surement sensitivity by increasing the impedance Hencea proper choice of operating frequency high enough toavoid ion accumulation is essential for optimum sensorperformance

It has been inferred from our results that anions becomeincorporated into the alumina lm during the anodiza-tion process supporting the hypothesis of Khanna andNahar104ndash106 where they act as proton donors and facili-tate the adsorption of water molecules Hence liquid-likecharge transfer networks are formed at relatively lowerhumidity levels for the smaller pore-size sensors Thisprocess rather than capillary condensation is primarilyresponsible for the observed humidity-sensing behavior ofthe sensors Our studies show that a distribution of poresizes as suggested by Shimizu et al94 is not necessary toobtain sensitive performance over a wide humidity rangerather such optimal performance can be obtained with uni-form nanodimensional pores containing adsorbed anionicimpurities

In the second sensing test case titania nanotubes pre-pared by anodization and annealed in an oxygen atmo-sphere at a temperature of 500 C were found to behighly sensitive to hydrogen The nanotube sensors con-tained both anatase and rutile phases of titania and showedappreciable sensitivity toward hydrogen at temperaturesas low as 180 C The sensitivity increased drasticallywith temperature showing a variation of 3 orders in mag-nitude of resistance to 1000 ppm of hydrogen at 400 CThe response time decreased with increasing temperatureat 290 C full switching of the sensor took approximately3 min Results were highly reproducible with no indica-tion of hysteresis Our results showed that these sensorsare at least capable of monitoring hydrogen levels from20 ppm to 4 At 290 C nanotubes with smaller porediameters showed higher sensitivity to hydrogen com-pared with their larger pore counterparts The sensorsshowed high selectivity to hydrogen compared with car-bon monoxide ammonia and carbon dioxide Althoughthe sensor was sensitive to high concentrations of oxy-gen the response time was high and the sensor did notcompletely regain the original condition We believe thehydrogen sensitivity of the nanotubes is due to hydrogendiffusion into the titania lattice where they act as electrondonors

291

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Acknowledgments Support of this work by the NationalScience Foundation through grant ECS-0210033 is grate-fully acknowledged

References and Notes

1 K Kajihara K Tanaka K Horao and N Soga Jpn J ApplPhys 35 6110 (1996)

2 B O Regan and M Gratzel A low cost Nature 353 737 (1991)3 A Fujishima and K Honda Nature 238 37 (1972)4 K Sato A Tsuzuki H Taoda Y Torii T Kato and Y Butsugan

J Mater Sci 29 5911 (1994)5 K Sato A Tsuzuki Y Torii H Taoda T Kato and Y Butsugan

J Mater Sci 30 837 (1995)6 N Ozer and C M Lampert Sol Energy Mater Sol Cells 54 147

(1998)7 A Manz and H Becker editors Microsystem Technology in

Chemistry and Life Sciences Springer-Verlag Berlin 19988 E Kim Y Xia and G M Whitesides Nature 376 581 (1995)9 M Templin A Franck A Du Chesne A Leist A Zhang

R Ulrich V Schadler and U Weisner Science 278 1795 (1997)10 D Zhao J Feng Q Huo N Melosh G H Frederickson

B Chmelka and G D Stucky Science 279 548 (1998)11 O D Velev T A Jede R F Lobo and A M Lenhoff Nature

389 447 (1997)12 B T Holland C F Blanford and A Stein Science 281 538

(1998)13 S Matsushita T Miwa and A Fujishima Chem Lett 9 925

(1997)14 A Imhof and D J Pine Adv Mater 10 697 (1998)15 T Tatsuma A Ikezawa Y Ohko T Miwa T Matsue and

A Fujishima Adv Mater 12 643 (2000)16 Z Zhong Y Yin B Gates and Y Xia Adv Mater 12 206 (2000)17 O D Velev and A M Lenhoff Curr Opin Colloids Interface

Sci 5 56 (2000)18 O D Velev and E W Kaler Adv Mater 12 531 (2000)19 P Jiang J Cizeron J F Bertone and V L Colvin J Am Chem

Soc 121 7957 (1999)20 M E Turner T J Trentler and V L Colvin Adv Mater 13 180

(2001)21 K M Kulinowski P Jiang H Vaswani and V L Colvin Adv

Mater 12 833 (2000)22 P Jiang J F Bertone and V L Colvin Science 291 453 (2001)23 R Seshadri and F C Meldrum Adv Mater 12 1149 (2000)24 F C Meldrum and R Seshadri Chem Commun 29 (2000)25 Q B Meng Z Z Gu O Sato and A Fujishima Appl Phys

Lett 77 4313 (2000)26 G Widawski B Rawiso and B Francois Nature 369 3897

(1994)27 B Francois O Pitois and J Francois Adv Mater 7 1041 (1995)28 O Pitois and B Francois Eur Phys J B 8 225 (1999)29 S A Jenekhe and X L Chen Science 283 372 (1999)30 M Shimomuara and T Sawadaishi Curr Opin Colloid Interface

Sci 6 11 (2001)31 O Karthaus N Maruyama X Cieren M Shimomura

H Hasegawa and T Hashimoto Langmuir 16 6071 (2000)32 T Nishikawa J Nishida R Ookura S Nishimura S Wada

T Karino and M Shimomura Mater Sci Eng C 10 141 (1999)33 K Kajihara K Nakanishi K Tanaka K Hirao and N Soga

J Am Ceram Soc 81 2670 (1998)34 C A Grimes R S Singh E C Dickey and O K Varghese

J Mater Res 16 1686 (2001)35 M Srinivasrao D Collings A Phillips and S Patel Science 292

79 (2001)36 P G de Gennes Rev Mod Phys 57 827 (1985)

37 D Beysens A Steyer P Guenoun D Fritter and C M KnoblerPhase Transitions 31 219 (1991)

38 B J Briscoe and K P Galvin J Phys D 23 422 (1990)39 B J Briscoe and K P Galvin J Phys D 23 1265 (1990)40 D Beysens and C M Knobler Phys Rev Lett 57 1433 (1986)41 F Family and P Meakin Phys Rev Lett 61 428 (1988)42 A V Limaye R D Narhe A M Dhote and S B Ogale Phys

Rev Lett 76 3762 (1996)43 P Dellrsquo Aversana and G P Neitzel Phys Today 51 38 (1998)44 P Dellrsquo Aversana J R Banavar and J Koplik Phys Fluids 8 15

(1986)45 P Dellrsquo Aversana V Tontodonato and L Carotenuto Phys Fluids

9 2475 (1997)46 M Q Li S Q Xu and E Kumacheva Langmuir 16 7275 (2000)47 P Freymuth Rev Sci Instrum 64 1 (1993)48 P Ball The Self-Made Tapestry Pattern Formation in Nature

Oxford University Press New York 199949 M F Schatz S J VanHook W D McCormick J B Swift and

H L Swinney Phys Fluids 11 2577 (1999)50 S R Coriell M R Cordes and W J Boettinger J Cryst Growth

49 13 (1980)51 T S Sullivan Y M Liu and R E Ecke Phys Rev E 54 486

(1996)52 J L Rogers M F Schatz J L Bougie and J B Swift 84 87

(2000)53 K Nitschke 52 R5772 (1995)54 E Bodenschatz J R De Bruyn G Ahlers and D S Cannell

Phys Rev Lett 67 3078 (1991)55 C W Meyer D S Cannell G Ahlers J B Swift and P C

Hohenberg Phys Rev Lett 61 947 (1988)56 R F Ismagilov D Rosmarin D H Gracias A D Stroock and

G M Whitesides Appl Phys Lett 79 439 (2001)57 E A Barringer and H K Bowen Langmuir 1 420 (1985)58 M Tschapek C Wasowski and R M T Sanchez J Electroanal

Chem 74 167 (1976)59 H Masuda and K Fukuda Science 268 1466 (1995)60 H Masuda F Hasegwa and S Ono J Electrochem Soc 144

L 127 (1997)61 O Jessensky F Muller and U Gosele Appl Phys Lett 72 1173

(1998)62 D Gong C A Grimes O K Varghese W Hu R S Singh

Z Chen and E C Dickey J Mater Res 16 3331 (2001)63 B M Kulwicki J Am Ceram Soc 74 697 (1991)64 H Arai and T Seiyama in Sensors A Comprehensive Survey

edited by W Gopel J Hesse and J N Zemel VCH Weinheim(1992) Vol 3 pp 981ndash1012

65 E Traversa Sens Actuators B 23 135 (1995)66 N Yamazoe and Y Shimizu Sens Actuators 10 379 (1986)67 A Bearzotti I Fratoddi L Palummo S Petrocco A Furlani

C Lo Sterzo and M V Russo Sens Actuators B 76 316 (2001)68 Y Sakai Y Sadaoka and M Matsuguchi Sens Actuators B 35

85 (1996)69 C A Grimes and D Kouzoudis Sens Actuators A 84 205

(2000)70 K G Ong C A Grimes C L Robbins and R S Singh Sens

Actuators A 93 33 (2001)71 T Seiyama N Yamazoe and H Arai Sens Actuators 4 85

(1983)72 L Ketron Ceram Bull 68 860 (1989)73 E Traversa G Gusmano and A Montenero Eur J Solid State

Inorg Chem 32 719 (1995)74 M K Jain M C Bhatnagar and G L Sharma Sens Actuators

B 55 180 (1999)75 O K Varghese and L K Malhotra J Appl Phys 87 7457 (2000)76 G Sberveglieri R Murri and N Pinto Sens Actuators B 23

177 (1995)

292

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Barrier layer

Titanium Nano tube array

Platinum electrodes

Fig 19 Schematic representation of the hydrogen sensor geometry

Munuera et al134 used rhenium doping in the anatasephase to perform room-temperature hydrogen monitor-ing Recently Shimizu et al122 reported that anodizednanoporous titania lms with a Pd Schottky barrier aresensitive to hydrogen at 250 C

Titania nanotubes were grown from titanium foil(995 pure from Alfa Aesar Ward Hill MA) with athickness of 025 mm The anodization was performed inan electrolyte medium of 05 hydro uoric acid in waterwith the use of a platinum foil cathode Well-de nednanotube arrays were grown with anodizing potentialsranging from 12 V to 20 V Nanotube length increaseswith anodization time reaching 400 nm in approximately20 min and then remains constant For the present studythe samples were anodized for 25 min The samples werethen annealed at 500 C in a pure oxygen ambient for 6 hwith a heating and cooling rate of 1 Cmin

The electrode geometry of the titania nanotube sen-sors is shown in Figure 19 The sensor consists of a basetitanium metal foil with a nanotube array grown on topAn insulating barrier layer separates the nanotubes fromthe conducting titanium foil Electrical connections weremade by two parallel 10 mm pound 2 mm platinum pads of100-Œm thickness The test chamber consists of a 13-literquartz tube with stainless steel end caps placed insidea furnace (Thermolyne USA model 21100 tubular fur-nace) Gas ow through the test chamber was controlledvia a computer-controlled mass ow controller The elec-trical resistance of the titania sensors was measured witha computer-controlled Keithly digital multimeter Prior todata collection the test chamber was initially evacuatedwith a mechanical pump whereupon nitrogen (99999pure) was passed while the sensor under test was heatedto the desired temperature The test gases were mixed inappropriate ratios with nitrogen to create the necessarytest gas ambient

The surface morphology of nanotube arrays preparedwith an anodization potential of 20 V and annealed at500 C for 6 h in a pure oxygen ambient is shown inFigure 20a and b It can be seen from these images thatthe nanotubes are uniform over the surface The nano-tubes are approximately 400 nm in length and have abarrier layer62 thickness of ordm50 nm For the nanotubes

100 nm

(a)

(b)

1 mm (c)

100 nm Fig 20 The surface morphology of the titania nanotubes after anneal-ing at 500 C (a) High- and (b) low-magni cation images of a 20-Vsample (c) A high-magni cation image of a 12-V sample

fabricated with 20-V anodization the average pore diam-eter as determined from FESEM images is 76 nm (thestandard deviation 15 nm) with a wall thickness of 27 nm(standard deviation 6 nm) The sample anodized at 12 Vshown in Figure 2c was found to have an average porediameter of 46 nm (standard deviation 8 nm) with a wallthickness of 17 nm (standard deviation 2 nm) The porosi-ties of the 20-V and 12-V samples were calculated as 45and 61 respectively

Glancing-angle X-ray diffraction patterns of a 20-Vsample annealed at 500 C for 6 h in oxygen ambientis shown in Figure 21 It can be seen that both anataseand rutile phases of titania are present in the sampleA detailed study135 of these structures by high-resolutiontransmission electron microscope (HRTEM) showed that

288

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

20 30 40 50 60 70 80

2 q (deg)

(101

)

(110

)

A

R

TR

T

T

R R A TRR T R R T TT

Fig 21 Glancing-angle X-ray diffraction pattern of a 20-V sample(glancing angle D 2) annealed at 500 C in oxygen ambient A R andT represent respectively the re ections from anatase crystallites rutilecrystallites and the titanium substrate

the anatase crystallites were concentrated on the walls ofthe nanotubes and rutile on the barrier layer

Nanotubes annealed in a pure oxygen ambient werefound to be stable (intact) to temperatures of approxi-mately 580 C Above this temperature protrusions wereseen to be coming out through the nanotubes an effectthat spread with increasing temperature These protru-sions which are due to oxidation of the titanium substratecollapse the nanotubes

Figure 22 shows the response of the 20-V (76 nm) sam-ple as a function of ambient temperature as it is switchedfrom a nitrogen environment to one containing 1000 ppmhydrogen and then back to nitrogen The plot was madeusing 4Rg=R05

ƒ1 versus time where R0 is the base resis-tance of the sensor that is the sensor resistance before thetest gas was introduced and Rg is the measured resistancein the presence of test gas The sensor shows increasinghydrogen sensitivity with temperature with a change by3 orders of magnitude in resistance at temperatures aboveordm300 C At all of the temperatures the original resistancerecovered without hysteresis

The sensitivity S is de ned by the formula

S DR0

ƒ Rgs

Rgs

10 0

10 1

10 2

10 3

0 10 0 5 10 2 1 10 3 15 103 2 10 3 25 103 3 10 3 35 103 4 10 3

Time (Sec)

(Rg

R0)

-1

180degC

240degC

265degC

290degC

345degC

375degC

H2+N

2

N2

Fig 22 Variation of resistance Rg normalized with respect to baselineresistance R0 of a 20-V sample with time on exposure to 1000 ppmhydrogen at different temperatures It may be noted that the inverse ofRg=R0 was used in the plot to represent data in the positive y direction

10-1

100

101

102

103

150 200 250 300 350 400 450

Temperature (degC)

Sens

itivi

ty (

R0-

Rgs

)R

gs

Fig 23 The sensitivity temperature dependence of a 20-V sample to1000 ppm hydrogen

where R0 is the resistance of the sensor before the gas ispassed and Rgs is the resistance after gas is passed andreaches the saturation value The temperature-dependentsensitivity of a 20-V sample to 1000 ppm hydrogen isshown in Figure 23 Sensitivity is seen to increase withtemperature to approximately 380 C where the increasein sensitivity with temperature begins to saturate

The response time de ned as the time needed for thesensor to reach 90 of the nal signal for a given concen-tration of gas is plotted against temperature in Figure 24(the time includes that required for the gas to equili-brate inside the measurement chamber estimated to beordm30 s) The response time decreases exponentially withtemperature

To check the behavior of the sensor on repeated hydro-gen exposure the hydrogen concentration was varied indiscrete steps of 100 ppm from 0 to 500 ppm while thetemperature was kept constant at 290 C the chamber was ushed with nitrogen after each exposure to hydrogenThe response of the 20-V prepared nanotube sensorskept at 290 C is shown in Figure 25 The behaviorof the sensor is consistent recovering its original resis-tance after repeated exposure to varying hydrogen gasconcentrations The sensitivities of the 76-nm- 53-nm-and 22-nm-diameter nanotube sensors to hydrogen con-centrations ranging from 100 ppm to 1 are shown inFigure 26 The sensitivity of the 76-nm (20-V) sensor tolow concentrations is shown in Figure 27 there is a linearincrease in sensitivity at low concentrations The lower

10 1

10 2

10 3

10 4

150 200 250 300 350 400 450

Temperature (degC)

Res

pons

e T

ime

(Sec

)

Fig 24 Variation in response time for a 20-V sensor as a function oftemperature The dots represent measured data

289

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

10 2

0 4 4 4 4 4

10 3

0 10 1 10 2 10 3 10 4 10 5 10

Time (Sec)

Res

ista

nce

(W)

H2 (ppm)

200

300

400

500

400

300

200

100

100

Fig 25 Resistance of a 20-V sample when exposed to different con-centrations of hydrogen at 290 C The nitrogen-hydrogen mixture waspassed for 1500 s the chamber was then ushed with nitrogen for 3000 sbefore the nitrogen-hydrogen mixture was passed again

test limit of 20 ppm is determined by the experimentalapparatus and not the sensor

Figure 28 shows the change in resistance of the 76-nm-46-nm- and 22-nm-diameter nanotube sensors with expo-sure to 1000 ppm hydrogen at 290 C Whereas smallerdiameter nanotubes had greater sensitivity to hydrogenthe samples made at lower anodizing voltages tended tobe more brittle and harder to handle mechanically withoutbreaking

The 20-V 76-nm-diameter sample was exposed to oxy-gen carbon monoxide ammonia and carbon dioxide at290 C to study the cross-sensitivity The sensor wasfound to have no detectable variation in resistance onexposure to carbon dioxide The sensitivities of the tita-nia nanotubes to the other gases are shown in Figure 29The sensitivities of the nanotubes to carbon monoxide andammonia are negligible compared with that of hydrogenThe resistance of the nanotubes increased in the presenceof oxygen and did not regain their original electrical con-ductivity even after several hours in a 290 C nitrogenenvironment

1

10

100

1000

104

105

0 02 04 06 08 1

Hydrogen Concentration ()

Pore diameter = 76 nm

Pore diameter = 53 nm

Pore diameter = 22nm

Sen

sitiv

ity (

R0-

Rg)

Rg

Fig 26 The sensitivity variation of a samples with pore diameters of22 nm 53 nm and 76 nm versus time and hydrogen concentrations

0 100 200 300 400 5001

10

100

Hydrogen concentration (ppm)

Sen

sitiv

ity

(R0-

Rgs

)R

gs

Fig 27 Sensitivity of a 20-V (76 nm) sample at 290 C for low hydro-gen concentrations

Since the sensor measurements were made in atmo-spheres without oxygen the increase in conductivitycannot be due to hydrogen removing oxygen from thelattice132ndash134 or the removal of chemisorbed oxygen135ndash137

Hydrogen molecules can dissociate at defects on thetitania surface which can subsequently diffuse into thetitania lattice acting as electron donors1281 1381 139 How-ever a dissociation-driven process would result in aslow response and recovery times with complete recov-ery unlikely since the sensor completely regains its orig-inal resistance with hydrogen cycling it appears that thisis not the dominant mechanism behind high hydrogensensitivity We believe chemisorption of the dissociatedhydrogen on the titania surface is the underlying sensingmechanism137 During chemisorption hydrogen acts as asurface state and a partial charge transfer from hydrogento the titania conduction band takes place This createsan electron accumulation layer on the nanotube surfaceenhancing its electrical conductance Upon removal of thehydrogen ambient electron transfer back to the hydro-gen molecule takes place which subsequently desorbsrestoring the original electrical resistance of the mate-rial Another factor that may play a role in the hydrogensensitivity (and selectivity) is the platinum electrodes Atelevated temperatures hydrogen dissociation can occur onplatinum surfaces These dissociated hydrogen atoms mayspill1361 140 onto the nanotube surface where they diffuse

10-1

101

103

105

0 05 1 15

Time (Hour)

Pore size - 76 nm

Pore size - 53 nm

Pore size - 22 nm

(Rg

R0)

-1

Fig 28 A comparison of the variation in resistance of samples havingpore diameters of 22 nm 53 nm and 76 nm versus time upon exposureto 1000 ppm of hydrogen at 290 C

290

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

10-2

10-1

100

0 100 2 102 4 102 6 102 8 102 1 103

OxygenAmmoniaCarbon monoxide

Time (Sec)

(Rg

R0)

-1

Fig 29 Variation of resistance of a 20-V sample when exposed to1000 ppm carbon monoxide 5 ammonia and 1 oxygen at 290 C

into the material affecting its electrical properties Fromthe present study it was not clear how signi cant a rolethe platinum electrodes play

Anatase the polymorph of titania has been reported tohave high sensitivity for reducing gases like hydrogen andcarbon monoxide1211 1241 133 Our nanotube samples containanatase phase mainly on the walls and rutile in the barrierlayer As the diffusing hydrogen atoms go to the intersti-tial sites1331 142 and as the ca ratio of anatase is almostfour times that of rutile it appears that the anatase latticeaccommodates hydrogen easily and hence makes a highercontribution to hydrogen sensitivity

The effect of chemisorption can be neglected in theoxygen-sensing experiments As the recovery requiresseveral hours it appears that the nanotubes contain oxy-gen vacancies or titanium interstitial defects in the pres-ence of nitrogen On exposure of the sensor to oxygenambient the lattice reoxidizes and hence the conductiv-ity of the sensor decreases On the removal of oxygenthe reduction of the lattice will not immediately occurhence the sensor requires several hours to regain its orig-inal conductivity

4 CONCLUSIONS

This paper summarizes different routes of fabrication ofmetal oxide nanoarchitectures useful for sensing applica-tions Structures made via anodization of a starting metaland breath gure formation from a sol-gel are discussedand presented as are TiO2 nanowires made by lling ananoporous alumina membrane with subsequent removalof the supporting template Two test cases illustratingthe utility of these nanoporous metal oxide structures forsensing are presented highly ordered nanoporous aluminais used for humidity sensing and TiO2 nanotubes are usedfor hydrogen sensing These study test cases reveal sev-eral points worth brie y noting

In the rst sensing test case uniform nanoporous alu-mina sensors exhibit pore size and frequency-dependent

humidity-sensing characteristics The mechanism respon-sible for the sensing action of the nanoporous alumina sen-sors was studied by representing the impedance spectra atdifferent humidity levels in Cole-Cole plots This makes itpossible to identify the competing impedance-determiningprocesses occurring in the sensors namely charge transferand mass transport through the physisorbed water layeras dependent upon the humidity level and input signalfrequency Charge transfer was found to enhance the sen-sitivity by causing a large variation in impedance withrespect to humidity Mass transport leads to accumulationof ions at the electrodes which tends to decrease the mea-surement sensitivity by increasing the impedance Hencea proper choice of operating frequency high enough toavoid ion accumulation is essential for optimum sensorperformance

It has been inferred from our results that anions becomeincorporated into the alumina lm during the anodiza-tion process supporting the hypothesis of Khanna andNahar104ndash106 where they act as proton donors and facili-tate the adsorption of water molecules Hence liquid-likecharge transfer networks are formed at relatively lowerhumidity levels for the smaller pore-size sensors Thisprocess rather than capillary condensation is primarilyresponsible for the observed humidity-sensing behavior ofthe sensors Our studies show that a distribution of poresizes as suggested by Shimizu et al94 is not necessary toobtain sensitive performance over a wide humidity rangerather such optimal performance can be obtained with uni-form nanodimensional pores containing adsorbed anionicimpurities

In the second sensing test case titania nanotubes pre-pared by anodization and annealed in an oxygen atmo-sphere at a temperature of 500 C were found to behighly sensitive to hydrogen The nanotube sensors con-tained both anatase and rutile phases of titania and showedappreciable sensitivity toward hydrogen at temperaturesas low as 180 C The sensitivity increased drasticallywith temperature showing a variation of 3 orders in mag-nitude of resistance to 1000 ppm of hydrogen at 400 CThe response time decreased with increasing temperatureat 290 C full switching of the sensor took approximately3 min Results were highly reproducible with no indica-tion of hysteresis Our results showed that these sensorsare at least capable of monitoring hydrogen levels from20 ppm to 4 At 290 C nanotubes with smaller porediameters showed higher sensitivity to hydrogen com-pared with their larger pore counterparts The sensorsshowed high selectivity to hydrogen compared with car-bon monoxide ammonia and carbon dioxide Althoughthe sensor was sensitive to high concentrations of oxy-gen the response time was high and the sensor did notcompletely regain the original condition We believe thehydrogen sensitivity of the nanotubes is due to hydrogendiffusion into the titania lattice where they act as electrondonors

291

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Acknowledgments Support of this work by the NationalScience Foundation through grant ECS-0210033 is grate-fully acknowledged

References and Notes

1 K Kajihara K Tanaka K Horao and N Soga Jpn J ApplPhys 35 6110 (1996)

2 B O Regan and M Gratzel A low cost Nature 353 737 (1991)3 A Fujishima and K Honda Nature 238 37 (1972)4 K Sato A Tsuzuki H Taoda Y Torii T Kato and Y Butsugan

J Mater Sci 29 5911 (1994)5 K Sato A Tsuzuki Y Torii H Taoda T Kato and Y Butsugan

J Mater Sci 30 837 (1995)6 N Ozer and C M Lampert Sol Energy Mater Sol Cells 54 147

(1998)7 A Manz and H Becker editors Microsystem Technology in

Chemistry and Life Sciences Springer-Verlag Berlin 19988 E Kim Y Xia and G M Whitesides Nature 376 581 (1995)9 M Templin A Franck A Du Chesne A Leist A Zhang

R Ulrich V Schadler and U Weisner Science 278 1795 (1997)10 D Zhao J Feng Q Huo N Melosh G H Frederickson

B Chmelka and G D Stucky Science 279 548 (1998)11 O D Velev T A Jede R F Lobo and A M Lenhoff Nature

389 447 (1997)12 B T Holland C F Blanford and A Stein Science 281 538

(1998)13 S Matsushita T Miwa and A Fujishima Chem Lett 9 925

(1997)14 A Imhof and D J Pine Adv Mater 10 697 (1998)15 T Tatsuma A Ikezawa Y Ohko T Miwa T Matsue and

A Fujishima Adv Mater 12 643 (2000)16 Z Zhong Y Yin B Gates and Y Xia Adv Mater 12 206 (2000)17 O D Velev and A M Lenhoff Curr Opin Colloids Interface

Sci 5 56 (2000)18 O D Velev and E W Kaler Adv Mater 12 531 (2000)19 P Jiang J Cizeron J F Bertone and V L Colvin J Am Chem

Soc 121 7957 (1999)20 M E Turner T J Trentler and V L Colvin Adv Mater 13 180

(2001)21 K M Kulinowski P Jiang H Vaswani and V L Colvin Adv

Mater 12 833 (2000)22 P Jiang J F Bertone and V L Colvin Science 291 453 (2001)23 R Seshadri and F C Meldrum Adv Mater 12 1149 (2000)24 F C Meldrum and R Seshadri Chem Commun 29 (2000)25 Q B Meng Z Z Gu O Sato and A Fujishima Appl Phys

Lett 77 4313 (2000)26 G Widawski B Rawiso and B Francois Nature 369 3897

(1994)27 B Francois O Pitois and J Francois Adv Mater 7 1041 (1995)28 O Pitois and B Francois Eur Phys J B 8 225 (1999)29 S A Jenekhe and X L Chen Science 283 372 (1999)30 M Shimomuara and T Sawadaishi Curr Opin Colloid Interface

Sci 6 11 (2001)31 O Karthaus N Maruyama X Cieren M Shimomura

H Hasegawa and T Hashimoto Langmuir 16 6071 (2000)32 T Nishikawa J Nishida R Ookura S Nishimura S Wada

T Karino and M Shimomura Mater Sci Eng C 10 141 (1999)33 K Kajihara K Nakanishi K Tanaka K Hirao and N Soga

J Am Ceram Soc 81 2670 (1998)34 C A Grimes R S Singh E C Dickey and O K Varghese

J Mater Res 16 1686 (2001)35 M Srinivasrao D Collings A Phillips and S Patel Science 292

79 (2001)36 P G de Gennes Rev Mod Phys 57 827 (1985)

37 D Beysens A Steyer P Guenoun D Fritter and C M KnoblerPhase Transitions 31 219 (1991)

38 B J Briscoe and K P Galvin J Phys D 23 422 (1990)39 B J Briscoe and K P Galvin J Phys D 23 1265 (1990)40 D Beysens and C M Knobler Phys Rev Lett 57 1433 (1986)41 F Family and P Meakin Phys Rev Lett 61 428 (1988)42 A V Limaye R D Narhe A M Dhote and S B Ogale Phys

Rev Lett 76 3762 (1996)43 P Dellrsquo Aversana and G P Neitzel Phys Today 51 38 (1998)44 P Dellrsquo Aversana J R Banavar and J Koplik Phys Fluids 8 15

(1986)45 P Dellrsquo Aversana V Tontodonato and L Carotenuto Phys Fluids

9 2475 (1997)46 M Q Li S Q Xu and E Kumacheva Langmuir 16 7275 (2000)47 P Freymuth Rev Sci Instrum 64 1 (1993)48 P Ball The Self-Made Tapestry Pattern Formation in Nature

Oxford University Press New York 199949 M F Schatz S J VanHook W D McCormick J B Swift and

H L Swinney Phys Fluids 11 2577 (1999)50 S R Coriell M R Cordes and W J Boettinger J Cryst Growth

49 13 (1980)51 T S Sullivan Y M Liu and R E Ecke Phys Rev E 54 486

(1996)52 J L Rogers M F Schatz J L Bougie and J B Swift 84 87

(2000)53 K Nitschke 52 R5772 (1995)54 E Bodenschatz J R De Bruyn G Ahlers and D S Cannell

Phys Rev Lett 67 3078 (1991)55 C W Meyer D S Cannell G Ahlers J B Swift and P C

Hohenberg Phys Rev Lett 61 947 (1988)56 R F Ismagilov D Rosmarin D H Gracias A D Stroock and

G M Whitesides Appl Phys Lett 79 439 (2001)57 E A Barringer and H K Bowen Langmuir 1 420 (1985)58 M Tschapek C Wasowski and R M T Sanchez J Electroanal

Chem 74 167 (1976)59 H Masuda and K Fukuda Science 268 1466 (1995)60 H Masuda F Hasegwa and S Ono J Electrochem Soc 144

L 127 (1997)61 O Jessensky F Muller and U Gosele Appl Phys Lett 72 1173

(1998)62 D Gong C A Grimes O K Varghese W Hu R S Singh

Z Chen and E C Dickey J Mater Res 16 3331 (2001)63 B M Kulwicki J Am Ceram Soc 74 697 (1991)64 H Arai and T Seiyama in Sensors A Comprehensive Survey

edited by W Gopel J Hesse and J N Zemel VCH Weinheim(1992) Vol 3 pp 981ndash1012

65 E Traversa Sens Actuators B 23 135 (1995)66 N Yamazoe and Y Shimizu Sens Actuators 10 379 (1986)67 A Bearzotti I Fratoddi L Palummo S Petrocco A Furlani

C Lo Sterzo and M V Russo Sens Actuators B 76 316 (2001)68 Y Sakai Y Sadaoka and M Matsuguchi Sens Actuators B 35

85 (1996)69 C A Grimes and D Kouzoudis Sens Actuators A 84 205

(2000)70 K G Ong C A Grimes C L Robbins and R S Singh Sens

Actuators A 93 33 (2001)71 T Seiyama N Yamazoe and H Arai Sens Actuators 4 85

(1983)72 L Ketron Ceram Bull 68 860 (1989)73 E Traversa G Gusmano and A Montenero Eur J Solid State

Inorg Chem 32 719 (1995)74 M K Jain M C Bhatnagar and G L Sharma Sens Actuators

B 55 180 (1999)75 O K Varghese and L K Malhotra J Appl Phys 87 7457 (2000)76 G Sberveglieri R Murri and N Pinto Sens Actuators B 23

177 (1995)

292

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

20 30 40 50 60 70 80

2 q (deg)

(101

)

(110

)

A

R

TR

T

T

R R A TRR T R R T TT

Fig 21 Glancing-angle X-ray diffraction pattern of a 20-V sample(glancing angle D 2) annealed at 500 C in oxygen ambient A R andT represent respectively the re ections from anatase crystallites rutilecrystallites and the titanium substrate

the anatase crystallites were concentrated on the walls ofthe nanotubes and rutile on the barrier layer

Nanotubes annealed in a pure oxygen ambient werefound to be stable (intact) to temperatures of approxi-mately 580 C Above this temperature protrusions wereseen to be coming out through the nanotubes an effectthat spread with increasing temperature These protru-sions which are due to oxidation of the titanium substratecollapse the nanotubes

Figure 22 shows the response of the 20-V (76 nm) sam-ple as a function of ambient temperature as it is switchedfrom a nitrogen environment to one containing 1000 ppmhydrogen and then back to nitrogen The plot was madeusing 4Rg=R05

ƒ1 versus time where R0 is the base resis-tance of the sensor that is the sensor resistance before thetest gas was introduced and Rg is the measured resistancein the presence of test gas The sensor shows increasinghydrogen sensitivity with temperature with a change by3 orders of magnitude in resistance at temperatures aboveordm300 C At all of the temperatures the original resistancerecovered without hysteresis

The sensitivity S is de ned by the formula

S DR0

ƒ Rgs

Rgs

10 0

10 1

10 2

10 3

0 10 0 5 10 2 1 10 3 15 103 2 10 3 25 103 3 10 3 35 103 4 10 3

Time (Sec)

(Rg

R0)

-1

180degC

240degC

265degC

290degC

345degC

375degC

H2+N

2

N2

Fig 22 Variation of resistance Rg normalized with respect to baselineresistance R0 of a 20-V sample with time on exposure to 1000 ppmhydrogen at different temperatures It may be noted that the inverse ofRg=R0 was used in the plot to represent data in the positive y direction

10-1

100

101

102

103

150 200 250 300 350 400 450

Temperature (degC)

Sens

itivi

ty (

R0-

Rgs

)R

gs

Fig 23 The sensitivity temperature dependence of a 20-V sample to1000 ppm hydrogen

where R0 is the resistance of the sensor before the gas ispassed and Rgs is the resistance after gas is passed andreaches the saturation value The temperature-dependentsensitivity of a 20-V sample to 1000 ppm hydrogen isshown in Figure 23 Sensitivity is seen to increase withtemperature to approximately 380 C where the increasein sensitivity with temperature begins to saturate

The response time de ned as the time needed for thesensor to reach 90 of the nal signal for a given concen-tration of gas is plotted against temperature in Figure 24(the time includes that required for the gas to equili-brate inside the measurement chamber estimated to beordm30 s) The response time decreases exponentially withtemperature

To check the behavior of the sensor on repeated hydro-gen exposure the hydrogen concentration was varied indiscrete steps of 100 ppm from 0 to 500 ppm while thetemperature was kept constant at 290 C the chamber was ushed with nitrogen after each exposure to hydrogenThe response of the 20-V prepared nanotube sensorskept at 290 C is shown in Figure 25 The behaviorof the sensor is consistent recovering its original resis-tance after repeated exposure to varying hydrogen gasconcentrations The sensitivities of the 76-nm- 53-nm-and 22-nm-diameter nanotube sensors to hydrogen con-centrations ranging from 100 ppm to 1 are shown inFigure 26 The sensitivity of the 76-nm (20-V) sensor tolow concentrations is shown in Figure 27 there is a linearincrease in sensitivity at low concentrations The lower

10 1

10 2

10 3

10 4

150 200 250 300 350 400 450

Temperature (degC)

Res

pons

e T

ime

(Sec

)

Fig 24 Variation in response time for a 20-V sensor as a function oftemperature The dots represent measured data

289

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

10 2

0 4 4 4 4 4

10 3

0 10 1 10 2 10 3 10 4 10 5 10

Time (Sec)

Res

ista

nce

(W)

H2 (ppm)

200

300

400

500

400

300

200

100

100

Fig 25 Resistance of a 20-V sample when exposed to different con-centrations of hydrogen at 290 C The nitrogen-hydrogen mixture waspassed for 1500 s the chamber was then ushed with nitrogen for 3000 sbefore the nitrogen-hydrogen mixture was passed again

test limit of 20 ppm is determined by the experimentalapparatus and not the sensor

Figure 28 shows the change in resistance of the 76-nm-46-nm- and 22-nm-diameter nanotube sensors with expo-sure to 1000 ppm hydrogen at 290 C Whereas smallerdiameter nanotubes had greater sensitivity to hydrogenthe samples made at lower anodizing voltages tended tobe more brittle and harder to handle mechanically withoutbreaking

The 20-V 76-nm-diameter sample was exposed to oxy-gen carbon monoxide ammonia and carbon dioxide at290 C to study the cross-sensitivity The sensor wasfound to have no detectable variation in resistance onexposure to carbon dioxide The sensitivities of the tita-nia nanotubes to the other gases are shown in Figure 29The sensitivities of the nanotubes to carbon monoxide andammonia are negligible compared with that of hydrogenThe resistance of the nanotubes increased in the presenceof oxygen and did not regain their original electrical con-ductivity even after several hours in a 290 C nitrogenenvironment

1

10

100

1000

104

105

0 02 04 06 08 1

Hydrogen Concentration ()

Pore diameter = 76 nm

Pore diameter = 53 nm

Pore diameter = 22nm

Sen

sitiv

ity (

R0-

Rg)

Rg

Fig 26 The sensitivity variation of a samples with pore diameters of22 nm 53 nm and 76 nm versus time and hydrogen concentrations

0 100 200 300 400 5001

10

100

Hydrogen concentration (ppm)

Sen

sitiv

ity

(R0-

Rgs

)R

gs

Fig 27 Sensitivity of a 20-V (76 nm) sample at 290 C for low hydro-gen concentrations

Since the sensor measurements were made in atmo-spheres without oxygen the increase in conductivitycannot be due to hydrogen removing oxygen from thelattice132ndash134 or the removal of chemisorbed oxygen135ndash137

Hydrogen molecules can dissociate at defects on thetitania surface which can subsequently diffuse into thetitania lattice acting as electron donors1281 1381 139 How-ever a dissociation-driven process would result in aslow response and recovery times with complete recov-ery unlikely since the sensor completely regains its orig-inal resistance with hydrogen cycling it appears that thisis not the dominant mechanism behind high hydrogensensitivity We believe chemisorption of the dissociatedhydrogen on the titania surface is the underlying sensingmechanism137 During chemisorption hydrogen acts as asurface state and a partial charge transfer from hydrogento the titania conduction band takes place This createsan electron accumulation layer on the nanotube surfaceenhancing its electrical conductance Upon removal of thehydrogen ambient electron transfer back to the hydro-gen molecule takes place which subsequently desorbsrestoring the original electrical resistance of the mate-rial Another factor that may play a role in the hydrogensensitivity (and selectivity) is the platinum electrodes Atelevated temperatures hydrogen dissociation can occur onplatinum surfaces These dissociated hydrogen atoms mayspill1361 140 onto the nanotube surface where they diffuse

10-1

101

103

105

0 05 1 15

Time (Hour)

Pore size - 76 nm

Pore size - 53 nm

Pore size - 22 nm

(Rg

R0)

-1

Fig 28 A comparison of the variation in resistance of samples havingpore diameters of 22 nm 53 nm and 76 nm versus time upon exposureto 1000 ppm of hydrogen at 290 C

290

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

10-2

10-1

100

0 100 2 102 4 102 6 102 8 102 1 103

OxygenAmmoniaCarbon monoxide

Time (Sec)

(Rg

R0)

-1

Fig 29 Variation of resistance of a 20-V sample when exposed to1000 ppm carbon monoxide 5 ammonia and 1 oxygen at 290 C

into the material affecting its electrical properties Fromthe present study it was not clear how signi cant a rolethe platinum electrodes play

Anatase the polymorph of titania has been reported tohave high sensitivity for reducing gases like hydrogen andcarbon monoxide1211 1241 133 Our nanotube samples containanatase phase mainly on the walls and rutile in the barrierlayer As the diffusing hydrogen atoms go to the intersti-tial sites1331 142 and as the ca ratio of anatase is almostfour times that of rutile it appears that the anatase latticeaccommodates hydrogen easily and hence makes a highercontribution to hydrogen sensitivity

The effect of chemisorption can be neglected in theoxygen-sensing experiments As the recovery requiresseveral hours it appears that the nanotubes contain oxy-gen vacancies or titanium interstitial defects in the pres-ence of nitrogen On exposure of the sensor to oxygenambient the lattice reoxidizes and hence the conductiv-ity of the sensor decreases On the removal of oxygenthe reduction of the lattice will not immediately occurhence the sensor requires several hours to regain its orig-inal conductivity

4 CONCLUSIONS

This paper summarizes different routes of fabrication ofmetal oxide nanoarchitectures useful for sensing applica-tions Structures made via anodization of a starting metaland breath gure formation from a sol-gel are discussedand presented as are TiO2 nanowires made by lling ananoporous alumina membrane with subsequent removalof the supporting template Two test cases illustratingthe utility of these nanoporous metal oxide structures forsensing are presented highly ordered nanoporous aluminais used for humidity sensing and TiO2 nanotubes are usedfor hydrogen sensing These study test cases reveal sev-eral points worth brie y noting

In the rst sensing test case uniform nanoporous alu-mina sensors exhibit pore size and frequency-dependent

humidity-sensing characteristics The mechanism respon-sible for the sensing action of the nanoporous alumina sen-sors was studied by representing the impedance spectra atdifferent humidity levels in Cole-Cole plots This makes itpossible to identify the competing impedance-determiningprocesses occurring in the sensors namely charge transferand mass transport through the physisorbed water layeras dependent upon the humidity level and input signalfrequency Charge transfer was found to enhance the sen-sitivity by causing a large variation in impedance withrespect to humidity Mass transport leads to accumulationof ions at the electrodes which tends to decrease the mea-surement sensitivity by increasing the impedance Hencea proper choice of operating frequency high enough toavoid ion accumulation is essential for optimum sensorperformance

It has been inferred from our results that anions becomeincorporated into the alumina lm during the anodiza-tion process supporting the hypothesis of Khanna andNahar104ndash106 where they act as proton donors and facili-tate the adsorption of water molecules Hence liquid-likecharge transfer networks are formed at relatively lowerhumidity levels for the smaller pore-size sensors Thisprocess rather than capillary condensation is primarilyresponsible for the observed humidity-sensing behavior ofthe sensors Our studies show that a distribution of poresizes as suggested by Shimizu et al94 is not necessary toobtain sensitive performance over a wide humidity rangerather such optimal performance can be obtained with uni-form nanodimensional pores containing adsorbed anionicimpurities

In the second sensing test case titania nanotubes pre-pared by anodization and annealed in an oxygen atmo-sphere at a temperature of 500 C were found to behighly sensitive to hydrogen The nanotube sensors con-tained both anatase and rutile phases of titania and showedappreciable sensitivity toward hydrogen at temperaturesas low as 180 C The sensitivity increased drasticallywith temperature showing a variation of 3 orders in mag-nitude of resistance to 1000 ppm of hydrogen at 400 CThe response time decreased with increasing temperatureat 290 C full switching of the sensor took approximately3 min Results were highly reproducible with no indica-tion of hysteresis Our results showed that these sensorsare at least capable of monitoring hydrogen levels from20 ppm to 4 At 290 C nanotubes with smaller porediameters showed higher sensitivity to hydrogen com-pared with their larger pore counterparts The sensorsshowed high selectivity to hydrogen compared with car-bon monoxide ammonia and carbon dioxide Althoughthe sensor was sensitive to high concentrations of oxy-gen the response time was high and the sensor did notcompletely regain the original condition We believe thehydrogen sensitivity of the nanotubes is due to hydrogendiffusion into the titania lattice where they act as electrondonors

291

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Acknowledgments Support of this work by the NationalScience Foundation through grant ECS-0210033 is grate-fully acknowledged

References and Notes

1 K Kajihara K Tanaka K Horao and N Soga Jpn J ApplPhys 35 6110 (1996)

2 B O Regan and M Gratzel A low cost Nature 353 737 (1991)3 A Fujishima and K Honda Nature 238 37 (1972)4 K Sato A Tsuzuki H Taoda Y Torii T Kato and Y Butsugan

J Mater Sci 29 5911 (1994)5 K Sato A Tsuzuki Y Torii H Taoda T Kato and Y Butsugan

J Mater Sci 30 837 (1995)6 N Ozer and C M Lampert Sol Energy Mater Sol Cells 54 147

(1998)7 A Manz and H Becker editors Microsystem Technology in

Chemistry and Life Sciences Springer-Verlag Berlin 19988 E Kim Y Xia and G M Whitesides Nature 376 581 (1995)9 M Templin A Franck A Du Chesne A Leist A Zhang

R Ulrich V Schadler and U Weisner Science 278 1795 (1997)10 D Zhao J Feng Q Huo N Melosh G H Frederickson

B Chmelka and G D Stucky Science 279 548 (1998)11 O D Velev T A Jede R F Lobo and A M Lenhoff Nature

389 447 (1997)12 B T Holland C F Blanford and A Stein Science 281 538

(1998)13 S Matsushita T Miwa and A Fujishima Chem Lett 9 925

(1997)14 A Imhof and D J Pine Adv Mater 10 697 (1998)15 T Tatsuma A Ikezawa Y Ohko T Miwa T Matsue and

A Fujishima Adv Mater 12 643 (2000)16 Z Zhong Y Yin B Gates and Y Xia Adv Mater 12 206 (2000)17 O D Velev and A M Lenhoff Curr Opin Colloids Interface

Sci 5 56 (2000)18 O D Velev and E W Kaler Adv Mater 12 531 (2000)19 P Jiang J Cizeron J F Bertone and V L Colvin J Am Chem

Soc 121 7957 (1999)20 M E Turner T J Trentler and V L Colvin Adv Mater 13 180

(2001)21 K M Kulinowski P Jiang H Vaswani and V L Colvin Adv

Mater 12 833 (2000)22 P Jiang J F Bertone and V L Colvin Science 291 453 (2001)23 R Seshadri and F C Meldrum Adv Mater 12 1149 (2000)24 F C Meldrum and R Seshadri Chem Commun 29 (2000)25 Q B Meng Z Z Gu O Sato and A Fujishima Appl Phys

Lett 77 4313 (2000)26 G Widawski B Rawiso and B Francois Nature 369 3897

(1994)27 B Francois O Pitois and J Francois Adv Mater 7 1041 (1995)28 O Pitois and B Francois Eur Phys J B 8 225 (1999)29 S A Jenekhe and X L Chen Science 283 372 (1999)30 M Shimomuara and T Sawadaishi Curr Opin Colloid Interface

Sci 6 11 (2001)31 O Karthaus N Maruyama X Cieren M Shimomura

H Hasegawa and T Hashimoto Langmuir 16 6071 (2000)32 T Nishikawa J Nishida R Ookura S Nishimura S Wada

T Karino and M Shimomura Mater Sci Eng C 10 141 (1999)33 K Kajihara K Nakanishi K Tanaka K Hirao and N Soga

J Am Ceram Soc 81 2670 (1998)34 C A Grimes R S Singh E C Dickey and O K Varghese

J Mater Res 16 1686 (2001)35 M Srinivasrao D Collings A Phillips and S Patel Science 292

79 (2001)36 P G de Gennes Rev Mod Phys 57 827 (1985)

37 D Beysens A Steyer P Guenoun D Fritter and C M KnoblerPhase Transitions 31 219 (1991)

38 B J Briscoe and K P Galvin J Phys D 23 422 (1990)39 B J Briscoe and K P Galvin J Phys D 23 1265 (1990)40 D Beysens and C M Knobler Phys Rev Lett 57 1433 (1986)41 F Family and P Meakin Phys Rev Lett 61 428 (1988)42 A V Limaye R D Narhe A M Dhote and S B Ogale Phys

Rev Lett 76 3762 (1996)43 P Dellrsquo Aversana and G P Neitzel Phys Today 51 38 (1998)44 P Dellrsquo Aversana J R Banavar and J Koplik Phys Fluids 8 15

(1986)45 P Dellrsquo Aversana V Tontodonato and L Carotenuto Phys Fluids

9 2475 (1997)46 M Q Li S Q Xu and E Kumacheva Langmuir 16 7275 (2000)47 P Freymuth Rev Sci Instrum 64 1 (1993)48 P Ball The Self-Made Tapestry Pattern Formation in Nature

Oxford University Press New York 199949 M F Schatz S J VanHook W D McCormick J B Swift and

H L Swinney Phys Fluids 11 2577 (1999)50 S R Coriell M R Cordes and W J Boettinger J Cryst Growth

49 13 (1980)51 T S Sullivan Y M Liu and R E Ecke Phys Rev E 54 486

(1996)52 J L Rogers M F Schatz J L Bougie and J B Swift 84 87

(2000)53 K Nitschke 52 R5772 (1995)54 E Bodenschatz J R De Bruyn G Ahlers and D S Cannell

Phys Rev Lett 67 3078 (1991)55 C W Meyer D S Cannell G Ahlers J B Swift and P C

Hohenberg Phys Rev Lett 61 947 (1988)56 R F Ismagilov D Rosmarin D H Gracias A D Stroock and

G M Whitesides Appl Phys Lett 79 439 (2001)57 E A Barringer and H K Bowen Langmuir 1 420 (1985)58 M Tschapek C Wasowski and R M T Sanchez J Electroanal

Chem 74 167 (1976)59 H Masuda and K Fukuda Science 268 1466 (1995)60 H Masuda F Hasegwa and S Ono J Electrochem Soc 144

L 127 (1997)61 O Jessensky F Muller and U Gosele Appl Phys Lett 72 1173

(1998)62 D Gong C A Grimes O K Varghese W Hu R S Singh

Z Chen and E C Dickey J Mater Res 16 3331 (2001)63 B M Kulwicki J Am Ceram Soc 74 697 (1991)64 H Arai and T Seiyama in Sensors A Comprehensive Survey

edited by W Gopel J Hesse and J N Zemel VCH Weinheim(1992) Vol 3 pp 981ndash1012

65 E Traversa Sens Actuators B 23 135 (1995)66 N Yamazoe and Y Shimizu Sens Actuators 10 379 (1986)67 A Bearzotti I Fratoddi L Palummo S Petrocco A Furlani

C Lo Sterzo and M V Russo Sens Actuators B 76 316 (2001)68 Y Sakai Y Sadaoka and M Matsuguchi Sens Actuators B 35

85 (1996)69 C A Grimes and D Kouzoudis Sens Actuators A 84 205

(2000)70 K G Ong C A Grimes C L Robbins and R S Singh Sens

Actuators A 93 33 (2001)71 T Seiyama N Yamazoe and H Arai Sens Actuators 4 85

(1983)72 L Ketron Ceram Bull 68 860 (1989)73 E Traversa G Gusmano and A Montenero Eur J Solid State

Inorg Chem 32 719 (1995)74 M K Jain M C Bhatnagar and G L Sharma Sens Actuators

B 55 180 (1999)75 O K Varghese and L K Malhotra J Appl Phys 87 7457 (2000)76 G Sberveglieri R Murri and N Pinto Sens Actuators B 23

177 (1995)

292

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

10 2

0 4 4 4 4 4

10 3

0 10 1 10 2 10 3 10 4 10 5 10

Time (Sec)

Res

ista

nce

(W)

H2 (ppm)

200

300

400

500

400

300

200

100

100

Fig 25 Resistance of a 20-V sample when exposed to different con-centrations of hydrogen at 290 C The nitrogen-hydrogen mixture waspassed for 1500 s the chamber was then ushed with nitrogen for 3000 sbefore the nitrogen-hydrogen mixture was passed again

test limit of 20 ppm is determined by the experimentalapparatus and not the sensor

Figure 28 shows the change in resistance of the 76-nm-46-nm- and 22-nm-diameter nanotube sensors with expo-sure to 1000 ppm hydrogen at 290 C Whereas smallerdiameter nanotubes had greater sensitivity to hydrogenthe samples made at lower anodizing voltages tended tobe more brittle and harder to handle mechanically withoutbreaking

The 20-V 76-nm-diameter sample was exposed to oxy-gen carbon monoxide ammonia and carbon dioxide at290 C to study the cross-sensitivity The sensor wasfound to have no detectable variation in resistance onexposure to carbon dioxide The sensitivities of the tita-nia nanotubes to the other gases are shown in Figure 29The sensitivities of the nanotubes to carbon monoxide andammonia are negligible compared with that of hydrogenThe resistance of the nanotubes increased in the presenceof oxygen and did not regain their original electrical con-ductivity even after several hours in a 290 C nitrogenenvironment

1

10

100

1000

104

105

0 02 04 06 08 1

Hydrogen Concentration ()

Pore diameter = 76 nm

Pore diameter = 53 nm

Pore diameter = 22nm

Sen

sitiv

ity (

R0-

Rg)

Rg

Fig 26 The sensitivity variation of a samples with pore diameters of22 nm 53 nm and 76 nm versus time and hydrogen concentrations

0 100 200 300 400 5001

10

100

Hydrogen concentration (ppm)

Sen

sitiv

ity

(R0-

Rgs

)R

gs

Fig 27 Sensitivity of a 20-V (76 nm) sample at 290 C for low hydro-gen concentrations

Since the sensor measurements were made in atmo-spheres without oxygen the increase in conductivitycannot be due to hydrogen removing oxygen from thelattice132ndash134 or the removal of chemisorbed oxygen135ndash137

Hydrogen molecules can dissociate at defects on thetitania surface which can subsequently diffuse into thetitania lattice acting as electron donors1281 1381 139 How-ever a dissociation-driven process would result in aslow response and recovery times with complete recov-ery unlikely since the sensor completely regains its orig-inal resistance with hydrogen cycling it appears that thisis not the dominant mechanism behind high hydrogensensitivity We believe chemisorption of the dissociatedhydrogen on the titania surface is the underlying sensingmechanism137 During chemisorption hydrogen acts as asurface state and a partial charge transfer from hydrogento the titania conduction band takes place This createsan electron accumulation layer on the nanotube surfaceenhancing its electrical conductance Upon removal of thehydrogen ambient electron transfer back to the hydro-gen molecule takes place which subsequently desorbsrestoring the original electrical resistance of the mate-rial Another factor that may play a role in the hydrogensensitivity (and selectivity) is the platinum electrodes Atelevated temperatures hydrogen dissociation can occur onplatinum surfaces These dissociated hydrogen atoms mayspill1361 140 onto the nanotube surface where they diffuse

10-1

101

103

105

0 05 1 15

Time (Hour)

Pore size - 76 nm

Pore size - 53 nm

Pore size - 22 nm

(Rg

R0)

-1

Fig 28 A comparison of the variation in resistance of samples havingpore diameters of 22 nm 53 nm and 76 nm versus time upon exposureto 1000 ppm of hydrogen at 290 C

290

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

10-2

10-1

100

0 100 2 102 4 102 6 102 8 102 1 103

OxygenAmmoniaCarbon monoxide

Time (Sec)

(Rg

R0)

-1

Fig 29 Variation of resistance of a 20-V sample when exposed to1000 ppm carbon monoxide 5 ammonia and 1 oxygen at 290 C

into the material affecting its electrical properties Fromthe present study it was not clear how signi cant a rolethe platinum electrodes play

Anatase the polymorph of titania has been reported tohave high sensitivity for reducing gases like hydrogen andcarbon monoxide1211 1241 133 Our nanotube samples containanatase phase mainly on the walls and rutile in the barrierlayer As the diffusing hydrogen atoms go to the intersti-tial sites1331 142 and as the ca ratio of anatase is almostfour times that of rutile it appears that the anatase latticeaccommodates hydrogen easily and hence makes a highercontribution to hydrogen sensitivity

The effect of chemisorption can be neglected in theoxygen-sensing experiments As the recovery requiresseveral hours it appears that the nanotubes contain oxy-gen vacancies or titanium interstitial defects in the pres-ence of nitrogen On exposure of the sensor to oxygenambient the lattice reoxidizes and hence the conductiv-ity of the sensor decreases On the removal of oxygenthe reduction of the lattice will not immediately occurhence the sensor requires several hours to regain its orig-inal conductivity

4 CONCLUSIONS

This paper summarizes different routes of fabrication ofmetal oxide nanoarchitectures useful for sensing applica-tions Structures made via anodization of a starting metaland breath gure formation from a sol-gel are discussedand presented as are TiO2 nanowires made by lling ananoporous alumina membrane with subsequent removalof the supporting template Two test cases illustratingthe utility of these nanoporous metal oxide structures forsensing are presented highly ordered nanoporous aluminais used for humidity sensing and TiO2 nanotubes are usedfor hydrogen sensing These study test cases reveal sev-eral points worth brie y noting

In the rst sensing test case uniform nanoporous alu-mina sensors exhibit pore size and frequency-dependent

humidity-sensing characteristics The mechanism respon-sible for the sensing action of the nanoporous alumina sen-sors was studied by representing the impedance spectra atdifferent humidity levels in Cole-Cole plots This makes itpossible to identify the competing impedance-determiningprocesses occurring in the sensors namely charge transferand mass transport through the physisorbed water layeras dependent upon the humidity level and input signalfrequency Charge transfer was found to enhance the sen-sitivity by causing a large variation in impedance withrespect to humidity Mass transport leads to accumulationof ions at the electrodes which tends to decrease the mea-surement sensitivity by increasing the impedance Hencea proper choice of operating frequency high enough toavoid ion accumulation is essential for optimum sensorperformance

It has been inferred from our results that anions becomeincorporated into the alumina lm during the anodiza-tion process supporting the hypothesis of Khanna andNahar104ndash106 where they act as proton donors and facili-tate the adsorption of water molecules Hence liquid-likecharge transfer networks are formed at relatively lowerhumidity levels for the smaller pore-size sensors Thisprocess rather than capillary condensation is primarilyresponsible for the observed humidity-sensing behavior ofthe sensors Our studies show that a distribution of poresizes as suggested by Shimizu et al94 is not necessary toobtain sensitive performance over a wide humidity rangerather such optimal performance can be obtained with uni-form nanodimensional pores containing adsorbed anionicimpurities

In the second sensing test case titania nanotubes pre-pared by anodization and annealed in an oxygen atmo-sphere at a temperature of 500 C were found to behighly sensitive to hydrogen The nanotube sensors con-tained both anatase and rutile phases of titania and showedappreciable sensitivity toward hydrogen at temperaturesas low as 180 C The sensitivity increased drasticallywith temperature showing a variation of 3 orders in mag-nitude of resistance to 1000 ppm of hydrogen at 400 CThe response time decreased with increasing temperatureat 290 C full switching of the sensor took approximately3 min Results were highly reproducible with no indica-tion of hysteresis Our results showed that these sensorsare at least capable of monitoring hydrogen levels from20 ppm to 4 At 290 C nanotubes with smaller porediameters showed higher sensitivity to hydrogen com-pared with their larger pore counterparts The sensorsshowed high selectivity to hydrogen compared with car-bon monoxide ammonia and carbon dioxide Althoughthe sensor was sensitive to high concentrations of oxy-gen the response time was high and the sensor did notcompletely regain the original condition We believe thehydrogen sensitivity of the nanotubes is due to hydrogendiffusion into the titania lattice where they act as electrondonors

291

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Acknowledgments Support of this work by the NationalScience Foundation through grant ECS-0210033 is grate-fully acknowledged

References and Notes

1 K Kajihara K Tanaka K Horao and N Soga Jpn J ApplPhys 35 6110 (1996)

2 B O Regan and M Gratzel A low cost Nature 353 737 (1991)3 A Fujishima and K Honda Nature 238 37 (1972)4 K Sato A Tsuzuki H Taoda Y Torii T Kato and Y Butsugan

J Mater Sci 29 5911 (1994)5 K Sato A Tsuzuki Y Torii H Taoda T Kato and Y Butsugan

J Mater Sci 30 837 (1995)6 N Ozer and C M Lampert Sol Energy Mater Sol Cells 54 147

(1998)7 A Manz and H Becker editors Microsystem Technology in

Chemistry and Life Sciences Springer-Verlag Berlin 19988 E Kim Y Xia and G M Whitesides Nature 376 581 (1995)9 M Templin A Franck A Du Chesne A Leist A Zhang

R Ulrich V Schadler and U Weisner Science 278 1795 (1997)10 D Zhao J Feng Q Huo N Melosh G H Frederickson

B Chmelka and G D Stucky Science 279 548 (1998)11 O D Velev T A Jede R F Lobo and A M Lenhoff Nature

389 447 (1997)12 B T Holland C F Blanford and A Stein Science 281 538

(1998)13 S Matsushita T Miwa and A Fujishima Chem Lett 9 925

(1997)14 A Imhof and D J Pine Adv Mater 10 697 (1998)15 T Tatsuma A Ikezawa Y Ohko T Miwa T Matsue and

A Fujishima Adv Mater 12 643 (2000)16 Z Zhong Y Yin B Gates and Y Xia Adv Mater 12 206 (2000)17 O D Velev and A M Lenhoff Curr Opin Colloids Interface

Sci 5 56 (2000)18 O D Velev and E W Kaler Adv Mater 12 531 (2000)19 P Jiang J Cizeron J F Bertone and V L Colvin J Am Chem

Soc 121 7957 (1999)20 M E Turner T J Trentler and V L Colvin Adv Mater 13 180

(2001)21 K M Kulinowski P Jiang H Vaswani and V L Colvin Adv

Mater 12 833 (2000)22 P Jiang J F Bertone and V L Colvin Science 291 453 (2001)23 R Seshadri and F C Meldrum Adv Mater 12 1149 (2000)24 F C Meldrum and R Seshadri Chem Commun 29 (2000)25 Q B Meng Z Z Gu O Sato and A Fujishima Appl Phys

Lett 77 4313 (2000)26 G Widawski B Rawiso and B Francois Nature 369 3897

(1994)27 B Francois O Pitois and J Francois Adv Mater 7 1041 (1995)28 O Pitois and B Francois Eur Phys J B 8 225 (1999)29 S A Jenekhe and X L Chen Science 283 372 (1999)30 M Shimomuara and T Sawadaishi Curr Opin Colloid Interface

Sci 6 11 (2001)31 O Karthaus N Maruyama X Cieren M Shimomura

H Hasegawa and T Hashimoto Langmuir 16 6071 (2000)32 T Nishikawa J Nishida R Ookura S Nishimura S Wada

T Karino and M Shimomura Mater Sci Eng C 10 141 (1999)33 K Kajihara K Nakanishi K Tanaka K Hirao and N Soga

J Am Ceram Soc 81 2670 (1998)34 C A Grimes R S Singh E C Dickey and O K Varghese

J Mater Res 16 1686 (2001)35 M Srinivasrao D Collings A Phillips and S Patel Science 292

79 (2001)36 P G de Gennes Rev Mod Phys 57 827 (1985)

37 D Beysens A Steyer P Guenoun D Fritter and C M KnoblerPhase Transitions 31 219 (1991)

38 B J Briscoe and K P Galvin J Phys D 23 422 (1990)39 B J Briscoe and K P Galvin J Phys D 23 1265 (1990)40 D Beysens and C M Knobler Phys Rev Lett 57 1433 (1986)41 F Family and P Meakin Phys Rev Lett 61 428 (1988)42 A V Limaye R D Narhe A M Dhote and S B Ogale Phys

Rev Lett 76 3762 (1996)43 P Dellrsquo Aversana and G P Neitzel Phys Today 51 38 (1998)44 P Dellrsquo Aversana J R Banavar and J Koplik Phys Fluids 8 15

(1986)45 P Dellrsquo Aversana V Tontodonato and L Carotenuto Phys Fluids

9 2475 (1997)46 M Q Li S Q Xu and E Kumacheva Langmuir 16 7275 (2000)47 P Freymuth Rev Sci Instrum 64 1 (1993)48 P Ball The Self-Made Tapestry Pattern Formation in Nature

Oxford University Press New York 199949 M F Schatz S J VanHook W D McCormick J B Swift and

H L Swinney Phys Fluids 11 2577 (1999)50 S R Coriell M R Cordes and W J Boettinger J Cryst Growth

49 13 (1980)51 T S Sullivan Y M Liu and R E Ecke Phys Rev E 54 486

(1996)52 J L Rogers M F Schatz J L Bougie and J B Swift 84 87

(2000)53 K Nitschke 52 R5772 (1995)54 E Bodenschatz J R De Bruyn G Ahlers and D S Cannell

Phys Rev Lett 67 3078 (1991)55 C W Meyer D S Cannell G Ahlers J B Swift and P C

Hohenberg Phys Rev Lett 61 947 (1988)56 R F Ismagilov D Rosmarin D H Gracias A D Stroock and

G M Whitesides Appl Phys Lett 79 439 (2001)57 E A Barringer and H K Bowen Langmuir 1 420 (1985)58 M Tschapek C Wasowski and R M T Sanchez J Electroanal

Chem 74 167 (1976)59 H Masuda and K Fukuda Science 268 1466 (1995)60 H Masuda F Hasegwa and S Ono J Electrochem Soc 144

L 127 (1997)61 O Jessensky F Muller and U Gosele Appl Phys Lett 72 1173

(1998)62 D Gong C A Grimes O K Varghese W Hu R S Singh

Z Chen and E C Dickey J Mater Res 16 3331 (2001)63 B M Kulwicki J Am Ceram Soc 74 697 (1991)64 H Arai and T Seiyama in Sensors A Comprehensive Survey

edited by W Gopel J Hesse and J N Zemel VCH Weinheim(1992) Vol 3 pp 981ndash1012

65 E Traversa Sens Actuators B 23 135 (1995)66 N Yamazoe and Y Shimizu Sens Actuators 10 379 (1986)67 A Bearzotti I Fratoddi L Palummo S Petrocco A Furlani

C Lo Sterzo and M V Russo Sens Actuators B 76 316 (2001)68 Y Sakai Y Sadaoka and M Matsuguchi Sens Actuators B 35

85 (1996)69 C A Grimes and D Kouzoudis Sens Actuators A 84 205

(2000)70 K G Ong C A Grimes C L Robbins and R S Singh Sens

Actuators A 93 33 (2001)71 T Seiyama N Yamazoe and H Arai Sens Actuators 4 85

(1983)72 L Ketron Ceram Bull 68 860 (1989)73 E Traversa G Gusmano and A Montenero Eur J Solid State

Inorg Chem 32 719 (1995)74 M K Jain M C Bhatnagar and G L Sharma Sens Actuators

B 55 180 (1999)75 O K Varghese and L K Malhotra J Appl Phys 87 7457 (2000)76 G Sberveglieri R Murri and N Pinto Sens Actuators B 23

177 (1995)

292

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

10-2

10-1

100

0 100 2 102 4 102 6 102 8 102 1 103

OxygenAmmoniaCarbon monoxide

Time (Sec)

(Rg

R0)

-1

Fig 29 Variation of resistance of a 20-V sample when exposed to1000 ppm carbon monoxide 5 ammonia and 1 oxygen at 290 C

into the material affecting its electrical properties Fromthe present study it was not clear how signi cant a rolethe platinum electrodes play

Anatase the polymorph of titania has been reported tohave high sensitivity for reducing gases like hydrogen andcarbon monoxide1211 1241 133 Our nanotube samples containanatase phase mainly on the walls and rutile in the barrierlayer As the diffusing hydrogen atoms go to the intersti-tial sites1331 142 and as the ca ratio of anatase is almostfour times that of rutile it appears that the anatase latticeaccommodates hydrogen easily and hence makes a highercontribution to hydrogen sensitivity

The effect of chemisorption can be neglected in theoxygen-sensing experiments As the recovery requiresseveral hours it appears that the nanotubes contain oxy-gen vacancies or titanium interstitial defects in the pres-ence of nitrogen On exposure of the sensor to oxygenambient the lattice reoxidizes and hence the conductiv-ity of the sensor decreases On the removal of oxygenthe reduction of the lattice will not immediately occurhence the sensor requires several hours to regain its orig-inal conductivity

4 CONCLUSIONS

This paper summarizes different routes of fabrication ofmetal oxide nanoarchitectures useful for sensing applica-tions Structures made via anodization of a starting metaland breath gure formation from a sol-gel are discussedand presented as are TiO2 nanowires made by lling ananoporous alumina membrane with subsequent removalof the supporting template Two test cases illustratingthe utility of these nanoporous metal oxide structures forsensing are presented highly ordered nanoporous aluminais used for humidity sensing and TiO2 nanotubes are usedfor hydrogen sensing These study test cases reveal sev-eral points worth brie y noting

In the rst sensing test case uniform nanoporous alu-mina sensors exhibit pore size and frequency-dependent

humidity-sensing characteristics The mechanism respon-sible for the sensing action of the nanoporous alumina sen-sors was studied by representing the impedance spectra atdifferent humidity levels in Cole-Cole plots This makes itpossible to identify the competing impedance-determiningprocesses occurring in the sensors namely charge transferand mass transport through the physisorbed water layeras dependent upon the humidity level and input signalfrequency Charge transfer was found to enhance the sen-sitivity by causing a large variation in impedance withrespect to humidity Mass transport leads to accumulationof ions at the electrodes which tends to decrease the mea-surement sensitivity by increasing the impedance Hencea proper choice of operating frequency high enough toavoid ion accumulation is essential for optimum sensorperformance

It has been inferred from our results that anions becomeincorporated into the alumina lm during the anodiza-tion process supporting the hypothesis of Khanna andNahar104ndash106 where they act as proton donors and facili-tate the adsorption of water molecules Hence liquid-likecharge transfer networks are formed at relatively lowerhumidity levels for the smaller pore-size sensors Thisprocess rather than capillary condensation is primarilyresponsible for the observed humidity-sensing behavior ofthe sensors Our studies show that a distribution of poresizes as suggested by Shimizu et al94 is not necessary toobtain sensitive performance over a wide humidity rangerather such optimal performance can be obtained with uni-form nanodimensional pores containing adsorbed anionicimpurities

In the second sensing test case titania nanotubes pre-pared by anodization and annealed in an oxygen atmo-sphere at a temperature of 500 C were found to behighly sensitive to hydrogen The nanotube sensors con-tained both anatase and rutile phases of titania and showedappreciable sensitivity toward hydrogen at temperaturesas low as 180 C The sensitivity increased drasticallywith temperature showing a variation of 3 orders in mag-nitude of resistance to 1000 ppm of hydrogen at 400 CThe response time decreased with increasing temperatureat 290 C full switching of the sensor took approximately3 min Results were highly reproducible with no indica-tion of hysteresis Our results showed that these sensorsare at least capable of monitoring hydrogen levels from20 ppm to 4 At 290 C nanotubes with smaller porediameters showed higher sensitivity to hydrogen com-pared with their larger pore counterparts The sensorsshowed high selectivity to hydrogen compared with car-bon monoxide ammonia and carbon dioxide Althoughthe sensor was sensitive to high concentrations of oxy-gen the response time was high and the sensor did notcompletely regain the original condition We believe thehydrogen sensitivity of the nanotubes is due to hydrogendiffusion into the titania lattice where they act as electrondonors

291

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Acknowledgments Support of this work by the NationalScience Foundation through grant ECS-0210033 is grate-fully acknowledged

References and Notes

1 K Kajihara K Tanaka K Horao and N Soga Jpn J ApplPhys 35 6110 (1996)

2 B O Regan and M Gratzel A low cost Nature 353 737 (1991)3 A Fujishima and K Honda Nature 238 37 (1972)4 K Sato A Tsuzuki H Taoda Y Torii T Kato and Y Butsugan

J Mater Sci 29 5911 (1994)5 K Sato A Tsuzuki Y Torii H Taoda T Kato and Y Butsugan

J Mater Sci 30 837 (1995)6 N Ozer and C M Lampert Sol Energy Mater Sol Cells 54 147

(1998)7 A Manz and H Becker editors Microsystem Technology in

Chemistry and Life Sciences Springer-Verlag Berlin 19988 E Kim Y Xia and G M Whitesides Nature 376 581 (1995)9 M Templin A Franck A Du Chesne A Leist A Zhang

R Ulrich V Schadler and U Weisner Science 278 1795 (1997)10 D Zhao J Feng Q Huo N Melosh G H Frederickson

B Chmelka and G D Stucky Science 279 548 (1998)11 O D Velev T A Jede R F Lobo and A M Lenhoff Nature

389 447 (1997)12 B T Holland C F Blanford and A Stein Science 281 538

(1998)13 S Matsushita T Miwa and A Fujishima Chem Lett 9 925

(1997)14 A Imhof and D J Pine Adv Mater 10 697 (1998)15 T Tatsuma A Ikezawa Y Ohko T Miwa T Matsue and

A Fujishima Adv Mater 12 643 (2000)16 Z Zhong Y Yin B Gates and Y Xia Adv Mater 12 206 (2000)17 O D Velev and A M Lenhoff Curr Opin Colloids Interface

Sci 5 56 (2000)18 O D Velev and E W Kaler Adv Mater 12 531 (2000)19 P Jiang J Cizeron J F Bertone and V L Colvin J Am Chem

Soc 121 7957 (1999)20 M E Turner T J Trentler and V L Colvin Adv Mater 13 180

(2001)21 K M Kulinowski P Jiang H Vaswani and V L Colvin Adv

Mater 12 833 (2000)22 P Jiang J F Bertone and V L Colvin Science 291 453 (2001)23 R Seshadri and F C Meldrum Adv Mater 12 1149 (2000)24 F C Meldrum and R Seshadri Chem Commun 29 (2000)25 Q B Meng Z Z Gu O Sato and A Fujishima Appl Phys

Lett 77 4313 (2000)26 G Widawski B Rawiso and B Francois Nature 369 3897

(1994)27 B Francois O Pitois and J Francois Adv Mater 7 1041 (1995)28 O Pitois and B Francois Eur Phys J B 8 225 (1999)29 S A Jenekhe and X L Chen Science 283 372 (1999)30 M Shimomuara and T Sawadaishi Curr Opin Colloid Interface

Sci 6 11 (2001)31 O Karthaus N Maruyama X Cieren M Shimomura

H Hasegawa and T Hashimoto Langmuir 16 6071 (2000)32 T Nishikawa J Nishida R Ookura S Nishimura S Wada

T Karino and M Shimomura Mater Sci Eng C 10 141 (1999)33 K Kajihara K Nakanishi K Tanaka K Hirao and N Soga

J Am Ceram Soc 81 2670 (1998)34 C A Grimes R S Singh E C Dickey and O K Varghese

J Mater Res 16 1686 (2001)35 M Srinivasrao D Collings A Phillips and S Patel Science 292

79 (2001)36 P G de Gennes Rev Mod Phys 57 827 (1985)

37 D Beysens A Steyer P Guenoun D Fritter and C M KnoblerPhase Transitions 31 219 (1991)

38 B J Briscoe and K P Galvin J Phys D 23 422 (1990)39 B J Briscoe and K P Galvin J Phys D 23 1265 (1990)40 D Beysens and C M Knobler Phys Rev Lett 57 1433 (1986)41 F Family and P Meakin Phys Rev Lett 61 428 (1988)42 A V Limaye R D Narhe A M Dhote and S B Ogale Phys

Rev Lett 76 3762 (1996)43 P Dellrsquo Aversana and G P Neitzel Phys Today 51 38 (1998)44 P Dellrsquo Aversana J R Banavar and J Koplik Phys Fluids 8 15

(1986)45 P Dellrsquo Aversana V Tontodonato and L Carotenuto Phys Fluids

9 2475 (1997)46 M Q Li S Q Xu and E Kumacheva Langmuir 16 7275 (2000)47 P Freymuth Rev Sci Instrum 64 1 (1993)48 P Ball The Self-Made Tapestry Pattern Formation in Nature

Oxford University Press New York 199949 M F Schatz S J VanHook W D McCormick J B Swift and

H L Swinney Phys Fluids 11 2577 (1999)50 S R Coriell M R Cordes and W J Boettinger J Cryst Growth

49 13 (1980)51 T S Sullivan Y M Liu and R E Ecke Phys Rev E 54 486

(1996)52 J L Rogers M F Schatz J L Bougie and J B Swift 84 87

(2000)53 K Nitschke 52 R5772 (1995)54 E Bodenschatz J R De Bruyn G Ahlers and D S Cannell

Phys Rev Lett 67 3078 (1991)55 C W Meyer D S Cannell G Ahlers J B Swift and P C

Hohenberg Phys Rev Lett 61 947 (1988)56 R F Ismagilov D Rosmarin D H Gracias A D Stroock and

G M Whitesides Appl Phys Lett 79 439 (2001)57 E A Barringer and H K Bowen Langmuir 1 420 (1985)58 M Tschapek C Wasowski and R M T Sanchez J Electroanal

Chem 74 167 (1976)59 H Masuda and K Fukuda Science 268 1466 (1995)60 H Masuda F Hasegwa and S Ono J Electrochem Soc 144

L 127 (1997)61 O Jessensky F Muller and U Gosele Appl Phys Lett 72 1173

(1998)62 D Gong C A Grimes O K Varghese W Hu R S Singh

Z Chen and E C Dickey J Mater Res 16 3331 (2001)63 B M Kulwicki J Am Ceram Soc 74 697 (1991)64 H Arai and T Seiyama in Sensors A Comprehensive Survey

edited by W Gopel J Hesse and J N Zemel VCH Weinheim(1992) Vol 3 pp 981ndash1012

65 E Traversa Sens Actuators B 23 135 (1995)66 N Yamazoe and Y Shimizu Sens Actuators 10 379 (1986)67 A Bearzotti I Fratoddi L Palummo S Petrocco A Furlani

C Lo Sterzo and M V Russo Sens Actuators B 76 316 (2001)68 Y Sakai Y Sadaoka and M Matsuguchi Sens Actuators B 35

85 (1996)69 C A Grimes and D Kouzoudis Sens Actuators A 84 205

(2000)70 K G Ong C A Grimes C L Robbins and R S Singh Sens

Actuators A 93 33 (2001)71 T Seiyama N Yamazoe and H Arai Sens Actuators 4 85

(1983)72 L Ketron Ceram Bull 68 860 (1989)73 E Traversa G Gusmano and A Montenero Eur J Solid State

Inorg Chem 32 719 (1995)74 M K Jain M C Bhatnagar and G L Sharma Sens Actuators

B 55 180 (1999)75 O K Varghese and L K Malhotra J Appl Phys 87 7457 (2000)76 G Sberveglieri R Murri and N Pinto Sens Actuators B 23

177 (1995)

292

RE

VIE

WVarghese and GrimesMetal Oxide for Sensing J Nanosci Nanotech 2003 3 277ndash293

Acknowledgments Support of this work by the NationalScience Foundation through grant ECS-0210033 is grate-fully acknowledged

References and Notes

1 K Kajihara K Tanaka K Horao and N Soga Jpn J ApplPhys 35 6110 (1996)

2 B O Regan and M Gratzel A low cost Nature 353 737 (1991)3 A Fujishima and K Honda Nature 238 37 (1972)4 K Sato A Tsuzuki H Taoda Y Torii T Kato and Y Butsugan

J Mater Sci 29 5911 (1994)5 K Sato A Tsuzuki Y Torii H Taoda T Kato and Y Butsugan

J Mater Sci 30 837 (1995)6 N Ozer and C M Lampert Sol Energy Mater Sol Cells 54 147

(1998)7 A Manz and H Becker editors Microsystem Technology in

Chemistry and Life Sciences Springer-Verlag Berlin 19988 E Kim Y Xia and G M Whitesides Nature 376 581 (1995)9 M Templin A Franck A Du Chesne A Leist A Zhang

R Ulrich V Schadler and U Weisner Science 278 1795 (1997)10 D Zhao J Feng Q Huo N Melosh G H Frederickson

B Chmelka and G D Stucky Science 279 548 (1998)11 O D Velev T A Jede R F Lobo and A M Lenhoff Nature

389 447 (1997)12 B T Holland C F Blanford and A Stein Science 281 538

(1998)13 S Matsushita T Miwa and A Fujishima Chem Lett 9 925

(1997)14 A Imhof and D J Pine Adv Mater 10 697 (1998)15 T Tatsuma A Ikezawa Y Ohko T Miwa T Matsue and

A Fujishima Adv Mater 12 643 (2000)16 Z Zhong Y Yin B Gates and Y Xia Adv Mater 12 206 (2000)17 O D Velev and A M Lenhoff Curr Opin Colloids Interface

Sci 5 56 (2000)18 O D Velev and E W Kaler Adv Mater 12 531 (2000)19 P Jiang J Cizeron J F Bertone and V L Colvin J Am Chem

Soc 121 7957 (1999)20 M E Turner T J Trentler and V L Colvin Adv Mater 13 180

(2001)21 K M Kulinowski P Jiang H Vaswani and V L Colvin Adv

Mater 12 833 (2000)22 P Jiang J F Bertone and V L Colvin Science 291 453 (2001)23 R Seshadri and F C Meldrum Adv Mater 12 1149 (2000)24 F C Meldrum and R Seshadri Chem Commun 29 (2000)25 Q B Meng Z Z Gu O Sato and A Fujishima Appl Phys

Lett 77 4313 (2000)26 G Widawski B Rawiso and B Francois Nature 369 3897

(1994)27 B Francois O Pitois and J Francois Adv Mater 7 1041 (1995)28 O Pitois and B Francois Eur Phys J B 8 225 (1999)29 S A Jenekhe and X L Chen Science 283 372 (1999)30 M Shimomuara and T Sawadaishi Curr Opin Colloid Interface

Sci 6 11 (2001)31 O Karthaus N Maruyama X Cieren M Shimomura

H Hasegawa and T Hashimoto Langmuir 16 6071 (2000)32 T Nishikawa J Nishida R Ookura S Nishimura S Wada

T Karino and M Shimomura Mater Sci Eng C 10 141 (1999)33 K Kajihara K Nakanishi K Tanaka K Hirao and N Soga

J Am Ceram Soc 81 2670 (1998)34 C A Grimes R S Singh E C Dickey and O K Varghese

J Mater Res 16 1686 (2001)35 M Srinivasrao D Collings A Phillips and S Patel Science 292

79 (2001)36 P G de Gennes Rev Mod Phys 57 827 (1985)

37 D Beysens A Steyer P Guenoun D Fritter and C M KnoblerPhase Transitions 31 219 (1991)

38 B J Briscoe and K P Galvin J Phys D 23 422 (1990)39 B J Briscoe and K P Galvin J Phys D 23 1265 (1990)40 D Beysens and C M Knobler Phys Rev Lett 57 1433 (1986)41 F Family and P Meakin Phys Rev Lett 61 428 (1988)42 A V Limaye R D Narhe A M Dhote and S B Ogale Phys

Rev Lett 76 3762 (1996)43 P Dellrsquo Aversana and G P Neitzel Phys Today 51 38 (1998)44 P Dellrsquo Aversana J R Banavar and J Koplik Phys Fluids 8 15

(1986)45 P Dellrsquo Aversana V Tontodonato and L Carotenuto Phys Fluids

9 2475 (1997)46 M Q Li S Q Xu and E Kumacheva Langmuir 16 7275 (2000)47 P Freymuth Rev Sci Instrum 64 1 (1993)48 P Ball The Self-Made Tapestry Pattern Formation in Nature

Oxford University Press New York 199949 M F Schatz S J VanHook W D McCormick J B Swift and

H L Swinney Phys Fluids 11 2577 (1999)50 S R Coriell M R Cordes and W J Boettinger J Cryst Growth

49 13 (1980)51 T S Sullivan Y M Liu and R E Ecke Phys Rev E 54 486

(1996)52 J L Rogers M F Schatz J L Bougie and J B Swift 84 87

(2000)53 K Nitschke 52 R5772 (1995)54 E Bodenschatz J R De Bruyn G Ahlers and D S Cannell

Phys Rev Lett 67 3078 (1991)55 C W Meyer D S Cannell G Ahlers J B Swift and P C

Hohenberg Phys Rev Lett 61 947 (1988)56 R F Ismagilov D Rosmarin D H Gracias A D Stroock and

G M Whitesides Appl Phys Lett 79 439 (2001)57 E A Barringer and H K Bowen Langmuir 1 420 (1985)58 M Tschapek C Wasowski and R M T Sanchez J Electroanal

Chem 74 167 (1976)59 H Masuda and K Fukuda Science 268 1466 (1995)60 H Masuda F Hasegwa and S Ono J Electrochem Soc 144

L 127 (1997)61 O Jessensky F Muller and U Gosele Appl Phys Lett 72 1173

(1998)62 D Gong C A Grimes O K Varghese W Hu R S Singh

Z Chen and E C Dickey J Mater Res 16 3331 (2001)63 B M Kulwicki J Am Ceram Soc 74 697 (1991)64 H Arai and T Seiyama in Sensors A Comprehensive Survey

edited by W Gopel J Hesse and J N Zemel VCH Weinheim(1992) Vol 3 pp 981ndash1012

65 E Traversa Sens Actuators B 23 135 (1995)66 N Yamazoe and Y Shimizu Sens Actuators 10 379 (1986)67 A Bearzotti I Fratoddi L Palummo S Petrocco A Furlani

C Lo Sterzo and M V Russo Sens Actuators B 76 316 (2001)68 Y Sakai Y Sadaoka and M Matsuguchi Sens Actuators B 35

85 (1996)69 C A Grimes and D Kouzoudis Sens Actuators A 84 205

(2000)70 K G Ong C A Grimes C L Robbins and R S Singh Sens

Actuators A 93 33 (2001)71 T Seiyama N Yamazoe and H Arai Sens Actuators 4 85

(1983)72 L Ketron Ceram Bull 68 860 (1989)73 E Traversa G Gusmano and A Montenero Eur J Solid State

Inorg Chem 32 719 (1995)74 M K Jain M C Bhatnagar and G L Sharma Sens Actuators

B 55 180 (1999)75 O K Varghese and L K Malhotra J Appl Phys 87 7457 (2000)76 G Sberveglieri R Murri and N Pinto Sens Actuators B 23

177 (1995)

292

RE

VIE

WJ Nanosci Nanotech 2003 3 277ndash293 Varghese and GrimesMetal Oxide for Sensing

77 E Traversa M Baroncini E D Bartolomeo G GusmanoP Innocenzi A Martucci and A Bearzotti J Eur Ceram Soc19 753 (1999)

78 K-S Chou T-K Lee and F-J Liu Sens Actuators B 56 106(1999)

79 E Traversa G Gnappi A Montenero and G Gusmano SensActuators B 31 59 (1996)

80 Y-C Yeh T-Y Tseng and D-A Chang J Am Ceram Soc 721472 (1989)

81 E Traversa J Am Ceram Soc 78 2625 (1995)82 T Yamamoto and K Murukami in Chemical Sensor Technology

edited by T Seiyama (1989) Vol 2 pp 133ndash14983 F Ansbacher and A C Jason Nature 171 177 (1953)84 V K Khanna and R K Nahar Sens Actuators 5 187 (1984)85 S Basu S Chatterjee M Saha S Bhandyopadhay K K Mistry

and K Sengupta Sens Actuators B 79 182 (2001)86 G Sberveglieri R Anchisini R Murri C Ercoli and N Pinto

Sens Actuators B 32 1 (1996)87 L H Mai P T M Hoa N T Binh N T T Ha and D K An

Sens Actuators B 66 63 (2000)88 S Chatterjee S Basu S Bandyopadhay K K Mistry and

K Sengupta Rev Sci Instrum 72 2792 (2001)89 R K Nahar Sens Actuators B 63 49 (2000)90 Z Chen M-C Jin and C Zhen Sens Actuators 2 167 (1990)91 S Basu M Saha S Chatterjee K K Mistry and K Sengupta

Mater Lett 49 29 (2001)92 Y Sadaoka Y Sakai and S Matsumoto J Mater Sci 21 1269

(1986)93 T Seiyama K Fueki J Shiokawa and S Suzuki editors Chem-

ical Sensors Elsevier New York (1983)94 Y Shimisu H Arai and T Seiyama Sens Actuators 7 11 (1985)95 T Morimoto M Nagao and F Tokuda J Phys Chem 73 243

(1969)96 W J Fleming Soc Automot Eng Trans Section 2 90 1656

(1981)97 H Kamiya M Mitsui H Takano and S Miyazawa J Am

Ceram Soc 83 287 (2000)98 S H-Tao W M-Tang L Ping and Y Xi Sens Actuators 19 61

(1989)99 D Gong V Yadavalli M Paulose M Pishko and C A Grimes

J Biomed Microdevices (2002)100 httpwwwtesatapecom101 J R Macdonald Impedance Spectroscopy Wiley New York (1987)102 A E Falk B M Lacquet and P L Swart Electron Lett 28 166

(1992)103 A K Jonscher J Mater Sci 13 553 (1978)104 V K Khanna and R K Nahar Appl Surf Sci 28 247 (1987)105 V K Khanna and R K Nahar J Phys D Appl Phys 19 L141

(1986)106 R K Nahar V K Khanna and W S Khokle J Phys D 17

2087 (1984)107 R K Nahar and V K Khanna Sens Actuators B 46 35 (1998)108 G C Wood P Skeldon G E Thompson and K Shimizu J Elec-

trochem Soc 143 74 (1996)109 G Patermarakis and K Moussoutzanis J Electrochem Soc 142

737 (1995)110 T Takeuchi Sens Actuators 14 109 (1988)111 U Kirner K D Schierbaum and W Gopel Sens Actuators B 1

103 (1990)112 Y Xu K Yao X Zhou and Q Cao Sens Actuators B 13ndash14

492 (1993)

113 V Demarne S Balkanova A Grisel D Rosenfeld and F LevySens Actuators B 13ndash14 497 (1993)

114 S Hasegawa Y Sasaki and S Matsuhara Sens Actuators B13ndash14 509 (1993)

115 A Rothschild F Edelman Y Komem and F Cosandey SensActuators B 67 282 (2000)

116 G Sakai N S Baik N Miura and N Yamazoe Sens ActuatorsB 77 116 (2001)

117 V N Misra and R P Agarwal Sens Actuators B 21 209(1991)

118 O K Varghese L K Malhotra and G L Sharma Sens Actua-tors B 55 161 (1999)

119 V A Chaudhary I S Mulla and K Vijayamohanan Sens Actu-ators B 50 45 (1998)

120 H Tang K Prasad R Sanjines and F Levy Sens Actuators B26ndash27 71 (1995)

121 S A Akbar and L B Younkman J Electrochem Soc 144 1750(1997)

122 Y Shimizu N Kuwano T Hyodo and M Egashira Sens Actu-ators B 83 195 (2002)

123 M C Carotta M Ferroni D Gnani V Guidi M Merli G Mar-tinelli M C Casale and M Notaro Sens Actuators B 58 310(1999)

124 N O Savage S A Akbar and P K Dutta Sens Actuators B72 239 (2001)

125 E Comini G Faglia G Sberveglieri Y X Li W Wlodarski andM K Ghantasala Sens Actuators B 64 169 (2000)

126 I Hayakawa Y Iwamoto K Kikuta and S Hirano Sens Actua-tors B 62 55 (2000)

127 N Savage B Chwieroth A Ginwalla B R Patton S A Akbarand P K Dutta Sens Actuators B 79 17 (2002)

128 C C Koch editor Nanostructured Materials Processing Prop-erties and Applications Noyes Publications New York (2002)

129 H-M Lin C-H Keng and C-Y Tung Nanostruct Mater 9 747(1997)

130 H Kobayashi K Kishimoto and Y Nakato Surf Sci 306 393(1994)

131 L A Harris J Electrochem Soc Solid State Sci Technol 1272657 (1980)

132 K D Schierbaum U K Kirner J F Geiger and W Gopel SensActuators B 4 87 (1991)

133 L D Birkefeld A M Azad and S A Akbar J Am Ceram Soc75 2964 (1992)

134 G Munuera A R G Elipe A Munoz A Fernandez J SoriaJ Conesa and J Sanz Sens Actuators 18 337 (1989)

135 O K Varghese D Gong M Paulose C A Grimes and E CDickey J Mater Res 18 156 (2003)

136 R D Shannon J Appl Phys 35 3414 (1964)137 K H Kim E J Ju and J S Choi J Phys Chem Solids 45

1265 (1984)138 C A Grimes J Waves Random Media 1 265 (1991)139 R M Walton D J Dwyer J W Schwank and J L Gland Appl

Surf Sci 125 199 (1998)140 M J Madou and S R Morrison Chemical Sensing with Solid

State Devices Academic Press New York (1989)141 G B Raupp and J A Dumesic J Phys Chem 89 5240 (1985)142 J B Bates J C Wang and R A Perkins Phys Rev B 19 4130

(1979)143 G J Hill Br J Appl Phys 1 1151 (1968)144 U Roland T Braunschweig and F Roessner J Mol Catal A

Chem 127 61 (1997)

Received 19 September 2002 RevisedAccepted 26 November 2002

293