Ac

TD

a

ARRAA

KABTOGE

1

zhtm1trpwWii

atcigc

h0

Sensors and Actuators B 204 (2014) 421–428

Contents lists available at ScienceDirect

Sensors and Actuators B: Chemical

jo ur nal home page: www.elsev ier .com/ locate /snb

n integrated optoelectronic chip for sensing aromatic hydrocarbonontaminants in groundwater

im A. Erickson ∗, Kevin L. Learepartment of Electrical and Computer Engineering, Colorado State University, Fort Collins, CO, USA

r t i c l e i n f o

rticle history:eceived 25 March 2014eceived in revised form 28 May 2014ccepted 7 June 2014vailable online 28 June 2014

eywords:

a b s t r a c t

The local evanescent array coupled (LEAC) chip-scale sensor is demonstrated for rapid detection of ben-zene, toluene and xylenes (BTX) in water at sub-ppm concentrations with minimal sample preparation.BTX analytes are found to concentrate into the chip’s Teflon AF upper cladding sensing region, causingan increase in refractive index, which modulates evanescent coupling into the chip’s integrated pho-todetector array. Integration of the photodetector array simplifies system instrumentation and permitsincorporation of an on-chip photocurrent reference region in the immediate vicinity of the sensing region,

romatic hydrocarbon sensingTEXeflon AFptical waveguide sensorroundwater qualitynvironmental monitoring

reducing drift due to temperature fluctuations. A nearly linear response is demonstrated for all testedconcentrations between 1 ppm and 30 ppm, and a limit of detection of 359 ppb, 249 ppb and 103 ppb isdemonstrated for benzene, toluene and xylene in water, respectively. The effect of matrix interferencecontaminants is investigated and a method for identifying individual contaminants based on extractionof the contaminant’s diffusion coefficient is presented.

© 2014 Elsevier B.V. All rights reserved.

. Introduction

Aromatic hydrocarbons such as benzene, toluene, ethylben-ene and xylenes (BTEX) are carcinogenic and hazardous to humanealth even at relatively low concentrations [1–3]. To ensurehe safety of municipal water supplies, the EPA has mandated

aximum allowable concentrations of 5 ppb, 1 ppm, 700 ppb and0 ppm for BTEX, respectively. In recent years, industrial activi-ies related to hydrocarbon production and transportation haveesulted in BTEX contamination of water supplies. For instance,ipeline leaks have resulted in benzene contamination of ground-ater in Parachute, Colorado [4] and well water in Jackson,isconsin [5]. Additionally, hydraulic fracturing activity has been

ndicated as the cause of BTEX and gasoline-range organic contam-nation of the Pavilion, Wyoming aquifer [6].

Due to the increased health risks and remediation costs associ-ted with larger contamination events, there is a strong motivationo develop portable, low-cost technologies, which can automati-ally sense BTEX contaminants and localize contamination events

n real-time. Current state-of-the art sensing technologies such asas chromatography–mass spectrometry (GC–MS) [7,8] and gashromatography–flame ionization detectors (GC–FID) [9–11] can

∗ Corresponding author. Tel.: +1 970 491 0718.E-mail address: kllear@engr.colostate.edu (K.L. Lear).

ttp://dx.doi.org/10.1016/j.snb.2014.06.020925-4005/© 2014 Elsevier B.V. All rights reserved.

selectively detect BTEX contaminants with exquisite sensitivityat ppt concentrations. However, due to cost, size, and energyrequirements, these systems are not typically deployed in the fieldas real-time monitoring systems. Current monitoring methodstypically involve sample collection in the field and transport backto a lab where the assay can be performed, resulting in delayedresults and increased analysis costs [12]. Thus, there is a need todevelop robust, field-deployable sensing technologies.

In recent years, a number of groups have developed increasinglyportable devices capable of detecting BTEX solutes in water. Thesedevices include absorption spectroscopy based sensors in the UV[13], near-IR [14,15,16], mid-IR [17,18] and IR [19], quartz crystalmicrobalance sensors [20], enzymatic fluorescence based sensors[21], and surface wave acoustic sensors [22]. While these devicesgenerally have simpler instrumentation requirements than eitherGC–MS or GC–FID, they still require significant external instrumen-tation, including optical spectrum analyzers, network analyzers,external broadband sources and external optics.

In this paper, we report a CMOS-compatible, optoelectronicchip which is capable of rapid, refractive index-based detectionof BTX in water. A novel aspect of the device is the inclusionof on-chip, low-noise (SNR > 3000) photodetectors, which greatly

simplifies instrumentation requirements relative to other sensingplatforms. We demonstrate the chip to be capable of broad-basedexclusionary detection of hydrophobic contaminants and a limitof detection of 359 ppb, 249 ppb and 103 ppb is established for

422 T.A. Erickson, K.L. Lear / Sensors and Actuators B 204 (2014) 421–428

btadd

2

2

gdupahmpimrplts

awpcIttddbtacFtnm

Fig. 2. (a) Effect of lower cladding thickness on device sensitivity and coupling for

Fig. 1. Diagram of device functionality.

enzene, toluene, and xylenes, respectively. Effects from sampleemperature fluctuations and potential interfering contaminantsre evaluated. Analyte selectivity is demonstrated by employing aiffusion theory model, which permits identification of the solute’siffusion coefficient.

. Materials and methods

.1. Device operation and sensing mechanism

The local evanescent array coupled (LEAC) chip is an inte-rated optical waveguide sensor, employing a single mode siliconioxide/silicon nitride (SiO2/SiNx) optical waveguide with anpper cladding sensing region. For aromatic hydrocarbon sensing,lasma-etched superhydrophobic Teflon AF 1600 is employed asn upper cladding sensing film, which preferentially concentratesydrophobic BTX contaminants at factors exceeding two orders ofagnitude [23]. As BTX solutes (n ≈ 1.5) present in a water sam-

le partition from the aqueous phase and diffuse into the lowndex (n = 1.31) film, the film’s refractive index increases causing

ore guided mode power to shift upward into the upper claddingegion of the waveguide, leading to a decrease in evanescent cou-ling into the integrated silicon (n = 3.83 + 0.014i) photodetector

ocated beneath the SiO2 lower cladding. The decrease in photode-ector coupling causes the measured photocurrent to decrease. Thisensing mechanism is referred to as the local evanescent field shift.

The sensing concept is visually summarized in Fig. 1. Light from red (660 nm) laser diode is end-fire coupled into the chip’s opticalaveguide. An impermeable metal blocking layer (gray rectangle)revents BTX contaminants from diffusing into the Teflon AF upperladding in the chip’s fixed index photocurrent reference region.n the downstream sensing region, BTX contaminants diffuse intohe upper cladding. As a result, there is a decrease in photodetec-or coupling compared to the reference region, as indicated by theecreased optical intensity leaking into the underlying silicon. Theecrease in coupling is measured as a decrease in photocurrenty the chip’s integrated metal–silicon–metal (MSM) photodetec-or array. Each detector in the array has a length of 300 �m and

width of 25 �m. There are a total of eight photodetectors on thehip, one reference detector followed by seven sensing detectors. Inig. 1, only the reference detector and the 1st and 2nd sensing detec-

ors are depicted. The first 3 mm of the chip’s waveguide, which isot depicted, is used to filter out unguided and substrate coupledodes.

a 65 nm thick waveguide with no scattering loss. (b) Device sensitivity as a functionof waveguide core and lower cladding dimensions.

2.2. Device modeling using a full-vector finite-differencemodesolver

Device performance is highly dependent on the dimensions ofthe chip’s waveguide. A two-dimensional, full-vector finite differ-ence modesolver [24] was used to optimize the dimensions ofthe core height (n = 1.8) and lower cladding (n = 1.46) subject toscattering losses. The photodetector coupling coefficient (cm−1)for both a pure Teflon AF upper cladding (n = 1.31) and Teflon AFwith diffused BTEX solutes (n = 1.311) was computed for a widerange of core and lower cladding thicknesses for a 4 �m wide, TE,single-mode waveguide. The 4 �m core width is preferred oversmaller single mode widths to reduce sidewall scattering losses[25].

Interestingly, simulation results show that device sensitivity (%change in photodetector coupling/refractive index unit) increaseslinearly with lower cladding thickness, but that photodetectorcoupling falls of exponentially with lower cladding thickness, asindicated in Fig. 2a. Thus, with an ideal, scatter-free waveguide, thechip’s sensitivity and corresponding limit of detection would onlybe restricted by photodetector noise constraints related to shot-noise, amplifier noise and dark current fluctuations, as sensitivitycan be arbitrarily increased by increasing lower cladding thick-

ness. This sensing mechanism is unique to single-mode waveguidesoperated near cutoff employing the local evanescent field shifteffect.

and A

2

aomodpPtsta

M

baeato(tsri

M

uistitaptbesn˛itf

2

idiopsf

gS

to the waveguide facet after careful alignment under a microscope.For fluid handling, a custom-machined flow cell with a volume of

T.A. Erickson, K.L. Lear / Sensors

.3. Effect of scattering on device performance and sensitivity

Practically speaking, device sensitivity is limited by scattering,nd the effect of scattering loss s must be taken into account whenptimizing device performance, as scattering confounds measure-ent of the evanescently coupled signal. Let P be the amount of

ptical power evanescently coupled into a given photodetector. Weefine the signal modulation M as the change in photodetector cou-ling from its initial value Po before solute diffusion to its final valuef after solute diffusion, such that M = (Po − Pf/Po). Let ˛o and ˛f behe coefficient of evanescent power coupling in before and afterolution diffusion, respectively. Assuming a 300 �m (.03 cm) pho-odetector length l, the modulation at the first sensing detector for

scatter-free waveguide is given by

= 1 − 1 − e−˛f l

1 − e−˛ol

Scattering losses of 8 ± 4 dB/cm were measured using cali-rated camera measurements of scattered light attenuation on

series of test waveguides with a lower cladding (3 �m) thicknough, so as to render evanescent coupling negligible. The corend cladding thickness were optimized subject to measured scat-ering losses. Assuming isotropic scattering, approximately 45.7%f scattered light is incident on the underlying photodetectortan−1(12.5/1.7)). Of this 45.7%, only about 60% is absorbed due tohe angular averaged Fresnel reflectance. Thus, only about 27.4% or

= 2.2 ± 0.3 dB/cm = 0.5 ± 0.07 cm−1 is actually absorbed by sensingegion photodetectors. The modulation when scattering is takennto account Ms is then given by,

S = 1 − (1 − e−˛ol) + (1 − e−sl)

(1 − e−˛f l) + (1 − e−sl)

Note that as s increases, modulation decreases.The sensor sensitivity, dMs/dn is given units of %/refractive index

nit (RIU) and parameterizes the fractional change in coupled lightntensity for a given change in refractive index. For the index changeimulated (�n = 0.001), the sensitivity is plotted in Fig. 2b for a scat-ering power absorption rate of s = 2.4 dB/cm. Note that sensitivityncreases monotonically with lower cladding thickness up to a cer-ain point for all core heights and then begins to fall of rapidlys the lower cladding thickness increases further. This inflectionoint is caused by the fixed scattered light intensity overwhelminghe evanescently coupled signal for larger lower cladding valuesecause the magnitude of evanescently coupled power falls ofxponentially as the lower cladding thickness increases. An optimalensitivity of approximately 1700%/RIU is found for a core thick-ess of 65 nm and a lower cladding thickness of 2000 nm, whereo = 16 dB/cm for n = 1.31. This sensitivity value is comparable to

ntensity-based surface plasmon resonance systems [26]. Simula-ion results were used to inform selection of waveguide dimensionsor device fabrication.

.4. Device fabrication

Chips are fabricated using standard photolithography, wet etch-ng, dry etching, metal deposition, metal liftoff and thin filmeposition techniques. The fabrication process flow is illustrated

n Fig. 3. For brevity, photolithography and liftoff steps have beenmitted from the diagram, but we note that Futurrex NR9-1500Photoresist is used for wet etching; Futurrex NR71-3000P photore-ist is used for metal liftoff, and Shipley S1808 photoresist is used

or dry etching. The fabrication process occurs as follows.

First, an RCA cleaned [27], n-type 〈1 0 0〉, 1–5 �-cm primerade wafer is thermally oxidized at 1150 ◦C to form a 1700 nmiO2 layer using a dry-wet-dry oxidation scheme [28]. After

ctuators B 204 (2014) 421–428 423

photolithography, the oxide is etched 1600 nm with 6:1 bufferedoxide etch (BOE), in order to from the waveguide’s lowercladding.

In order to produce a high Schottky barrier, low dark currentcontacts for the integrated MSM photodetectors, Au is selected asthe silicon contact metal [29]. However, Au adheres very poorlyto oxide, so a thin layer of Cr/Au is deposited on top of the oxideto serve as a metal adhesion layer for the Au/n-Si contacts. Aftermetal adhesion layer patterning, the remaining 100 nm of oxide isBOE etched to reveal the bare silicon. After photolithography, theSi native oxide layer is removed using a 3 s BOE dip followed by adeionized water rinse and N2 drying. Then 35 nm of Au, 20 nm ofCr, and 75 nm of Al are deposited at 0.1 nm/s to form the integratedMSM photodetector array. Au, Cr, and Al serve as contact, bar-rier and probing metals, respectively [30]. After metal deposition,70 nm SiNx (n = 1.80) is deposited using plasma-enhanced chemicalvapor deposition (PECVD) [31]. The 4 �m waveguide core is formedafter photolithography by dry etching the SiNx 35 nm using a reac-tive ion etcher (RIE) flowing CF4/O2 (8%) at 50 sccm with a plasmapower of 50 W. After core formation, Teflon AF 1600 is spin coatedonto the chip. As Teflon does not readily adhere to silicon nitride, afluorosilane adhesion promoter is applied [23]. Then Teflon is spincoated onto the chip 4 times at 800 RPM. Between each spin coat,the film is cured in an oven at 300 ◦C for 1 h with a 25 ◦C/min ramptime. After curing, the film is etched in an RIE for 5 s in oxygenplasma at a flow rate of 50 sccm and a power of 50 W. The plasmaetch is necessary to ensure that the layers of Teflon adhere to eachother. This process yields a ∼6 �m Teflon film as measured by whitelight spectrometry. After Teflon patterning, the film is once againplasma etched prior to the next photolithography step, in order toensure photoresist adhesion to the Teflon. After photolithography,200 nm of aluminum is deposited to form a BTEX impermeableblocking layer for the reference detector (Fig. 4a). After this step,only the sensing region detectors are directly exposed to theanalyte solution. This is done to prevent other regions of the chipfrom absorbing solutes from the sample volume. After the metalreference region patterning step, the remaining Teflon in thesensing region is oxygen plasma etched to a depth of 1200 nm. Thisthickness is sufficient to ensure that 99.9998% of evanescent fieldpower is present in the Teflon upper cladding, in order to ensuremode matching and field continuity. Teflon serves the dual purposeof sensing film and insulating layer. Plasma etching has the addedadvantage of oxidizing the aluminum, forming a hydrophilic metaloxide capping layer [32]. In the next step (not shown in Fig. 3) theremaining Teflon in the probe pad region of the chip is removedthrough oxygen plasma etching to permit electrical contact withthe chip. In the final process step, the waveguide facet is polished,so that light can be end-fire coupled into the waveguide. Fig. 4bis an SEM image of the chip before Teflon coating and aluminumdeposition.

3. Experiment and results

3.1. Experimental setup

For all experiments, ∼1 mW of optical power from a 660 nm laserdiode was fiber-coupled into the waveguide. To minimize fluctu-ations in coupled light intensity, the 4/125 �m fiber was epoxied

200 �L was clamped onto the chip and sealed with a 1/16′′ diame-ter O-ring. Sample solutions were manually injected into the fluidicchamber using a syringe. A probe card was used to contact the chip’sreadout pads for photocurrent measurement.

424 T.A. Erickson, K.L. Lear / Sensors and Actuators B 204 (2014) 421–428

Fig. 3. Fabrication

Fig. 4. (a) Optical microscope image of the chip’s reference and sensing region withcorresponding 300 �m reference and sensing photodetectors. The waveguide coreand lower cladding are seen in the middle of the image. As BTX solutes can diffuselaterally, a 100 �m sweep detector is used to separate the reference detector fromthe 1st sensing detector. (b) SEM image taken before Teflon AF and reference regionpatterning. The reference detector, sweep detector and first sensing detector arelabeled in the image along with the shared bias contact. For n-type Si, the individ-ual photodetectors are reverse biased at 2 V, by applying a 5 V bias to the sharedcontact and a 3 V bias to the individual detectors. The apparent overlap of separatemetal contacts is an artifact resulting from the 45◦ tilt at which the SEM image wasacquired.

process flow.

3.2. Data acquisition system

Photocurrents on the chip’s photodetectors were measuredusing a custom data acquisition system, which permitted simul-taneous photocurrent measurement on all photodetectors at asampling rate of 2 kHz. The system consisted of a battery-powered,single-supply, 8-channel transimpedance amplifier (TIA) con-nected to an 8-channel ADC, which then interfaced with a computerrunning Labview for automated data acquisition. The TIA consistedof a gain stage, which ranged from 0.4 V/�A to 2 V/�A dependingon detector number (waveguide position). The gain stage was fol-lowed by a two-pole Butterworth 6 Hz low pass filter. To reducedrift and noise, “zero drift” op amps (LTC 1050, Linear Technology)were used for both amplifier stages. To collect generated carriers,a 2 V bias is supplied across the contacts.

3.3. Signal processing and conversion to normalized photocurrent

Prior to each experiment, the dark current on each detector isrecorded and averaged over a period of 8 s to give Idark,i for eachdetector. The dark current on each detector is then subtracted fromthe measured current Imeas,i when light is coupled into the waveg-uide, to give the photocurrent on the ith detector Idet,i = Imeas,i −Idark,i. The dark current on each detector was approximately 400 pA,corresponding to a Schottky barrier height of 0.7 Ev [33,34].

To increase SNR, photocurrent measurements on all channelsare averaged over 8 s (16,000 samples total). After averaging, theraw photocurrent on the ith sensing photodetector Idet,i is nor-malized by dividing by the reference region photocurrent Iref suchthat Inorm,i = Idet,i/Iref. The on-chip photocurrent reference region isused to correct for small changes in coupled light intensity andtemperature-related drift, in order to resolve changes in photo-current solely caused by solute diffusion into the chip’s uppercladding sensing film. After reference photocurrent normalization,all photocurrents are divided by the maximum value on the 1stsensing detector (Inorm,i) so that all normalized photocurrent valuesare between 0 and 1.

3.4. BTEX sensing experiments

A series of experiments was conducted to characterize thesensing performance of six nominally identical chips (B1, B2, B3,

T.A. Erickson, K.L. Lear / Sensors and Actuators B 204 (2014) 421–428 425

Xdsc

3

stpCcctst3

cl1satAfitrbbbichr

3

gatpof

Fig. 5. Xylene flow experiment.

1, T1 and E1) fabricated on the same wafer. Experiments wereesigned, in order to determine a limit of detection for BTX solutes,ensor reusability, sensor repeatability, and interference from otherontaminants and temperature fluctuations.

.4.1. Temporal resolution of solute diffusionFor benzene, toluene and xylene sensing experiments, three

eparate chips (X1, T1, and B1) were used, respectively. To mimicrue environmental sampling conditions, test solutions were pre-ared using water drawn from Horsetooth Reservoir (Fort Collins,O). (An initial experiment on chip X1 showed no measurablehange in photocurrent when DI Water was injected into the flowell followed by reservoir water, implying that any potential con-amination in the reservoir was below the limit of detection of theystem). Solutions of high-purity benzene, toluene, and a mixture ofhe three xylene isomers were prepared at concentrations of 1 ppm,

ppm, 10 ppm, and 30 ppm in reservoir water.For each experiment, reservoir water is injected into the flow

ell and data is recorded for 2 min in order to establish a base-ine. Then the 1 ppm solution is injected and data is acquired for5 min followed reservoir water injection and another 15 min mea-urement period. This procedure is then repeated for the 3, 10nd 30 ppm concentrations. A time trace of the normalized pho-ocurrent on sensing detector #1 is displayed for xylene in Fig. 5.s expected, the photocurrent decreases as xylene diffuses into thelm at increasing concentrations. As water flows into the channel,he xylene diffuses out of the film and the modulation begins toeturn to baseline. While the film cannot be quickly regenerated toaseline due to the high affinity of BTX solutes for the hydropho-ic film, separate experiments have demonstrated that the film cane fully regenerated by a 24-h soak in deionized water or by dry-

ng the chip and heating it to 150 ◦C for 5 min. In this context, thehips are re-usable, but require separate processing after the filmas encountered a diffusing hydrophobic contaminant to be fullyegenerated.

.4.2. Effect of solute diffusion on mode propagationIn order to investigate how solute diffusion affects light propa-

ation and evanescent coupling along the length of the waveguide, total of 7 sensing detectors were included on the chip. In Fig. 6,

he measured raw photocurrent on each sensing detector islotted for water and a water sample contaminated with 30 ppmf xylene, which was allowed to equilibrate for 15 min. The datarom the chip’s seven detectors are shown and plotted vs. position

Fig. 6. Effect of 30 ppm xylene on mode propagation.

along the waveguide from the first sensing detector. The rawphotocurrent data is presented to demonstrate the effect of solutediffusion on evanescent coupling. (In order to resolve changes inevanescent photodetector coupling with higher precision, the ref-erence photodetector must be used to compute the dimensionlessnormalized photocurrent.)

As a result of xylene diffusion into the sensing region,the calculated evanescent coupling strength decreases from˛water = 6.89 cm−1 to ˛xylene = 6.49 cm−1. The greatest change inmodulation is found on the 1st sensing photodetector (M = 8.4%),whereas the modulation on the 7th photodetector in only 1.4%.The decrease in modulation results from two competing effects.Due to the increased upper cladding index in the upstream sensingregion, more optical power reaches the 7th photodetector. As aresult, the decrease in localized photodetector coupling is coun-teracted by the increased optical power entering the 7th sensingregion. Thus in practice, only the reference detector and the 1stsensing photodetector are required to detect contaminants.

3.4.3. Response linearityThe responses of chips X1, T1, and B1 to 1, 3, 10 and 30 ppm

concentrations of xylene, toluene, and benzene, respectively, areplotted in Fig. 7. Results are for a 15 min equilibration time periodat each concentration. A nearly linear response is found for all BTXsolutes, as indicated by R-square values exceeding 0.99 for all linesof best fit.

3.4.4. Device repeatabilityTo evaluate measurement repeatability across multiple chips,

the benzene flow experiment used for chip B1 was repeated onchips B2 and B3 at concentrations of 1, 3, 10 and 30 ppm. The aver-age standard deviation in sensitivity across the four concentrationstested is 18.3% for a 15 min exposure period. The deviation in sensi-tivity between chips B1, B2 and B3 is likely caused by differences inscattering loss among the chips. Based on the scattering model pre-sented in Section 2.3, a waveguide with a scattering loss of 4 dB/cmwould provide a sensitivity of 1520%/RIU vs. 1398%/RIU for a loss of12 dB/cm, an 8% difference. While care was taken to prepare cappedsolutions identically followed by sonication before sample extrac-

tion with a syringe, it is possible that the deviation in response couldhave also resulted from concentration differences in the solutionbeing sampled.

426 T.A. Erickson, K.L. Lear / Sensors and Actuators B 204 (2014) 421–428

Fr

3

tefispluuewoamc

3

pitgb[T(rcu

aaifittama

of detection is the given by LOD = (3�/�M) × (1 ppm) and is found tobe 103 ppb, 249 ppb, and 359 ppb for benzene, toluene and xylene,respectively.

ig. 7. Modulation response of chips B1, X1 and T1 to benzene, xylene, and toluene,espectively at concentrations of 1, 3, 10 and 30 ppm.

.5. Effect of temperature fluctuations

For laboratory experiments, all sample solutions were allowedo equilibrate to room temperature in a controlled temperaturenvironment. For field measurements, it would likely be more dif-cult to control the temperature of both the pure water referenceolution used for establishing a baseline reading and the sam-le undergoing analysis. Additionally, if the chip were used for

ong-term monitoring, the ambient sample medium would likelyndergo relatively large temperature fluctuations. In order to eval-ate the effect of temperature on device performance, the followingxperiment was conducted on chip E1. First a room temperatureater sample (22 ◦C) was injected in the flow cell. Then a sample

f chilled ice water was injected into the sample flow cell, yielding near-instantaneous temperature deviation of ∼22 ◦C. The maxi-um deviation in the normalized photocurrent was found to 0.44%,

orresponding to temperature stability of 0.02%/◦C.

.6. Effect of interference from other matrix contaminants

A given water sample may contain a variety of ionic and non-olar contaminants, in addition to BTX contaminants. Thus, it is

mportant to investigate how other potential sample matrix con-aminants can affect the signal when BTX are not present, therebyiving a potentially false positive reading. Previously, solute uptakey Teflon films has been shown to exhibit a strong dependence23] on the solute’s octanol–water partition coefficient, Pow [35].hus, 30 ppm of hexane (log(Pow) = 4.0) and 30 ppm of acetonelog(Pow) = −0.24) were selected as test solutions to span a largeange of octanol–water coefficients. To evaluate the effect of ionicontaminants, a 4% NaCl solution, used to mimic ocean water wassed.

Chip E1, which had previously been use for temperature fluctu-tion measurements was used for the following experiment. First,

water baseline reading was taken. Then the 4% saline sample wasnjected into the flow cell and allowed to equilibrate for 10 minollowed by a 15 min water rinse. Then the acetone solution wasnjected, followed by a 15 min water rinse. Lastly, the hexane solu-ion was injected into the channel. The rinse step was included

o avoid cross-contamination of the interfering solutes beingnalyzed and bring the film back to baseline. The data showed noeasurable change for the 4% NaCl solution. The data for hexane

nd acetone solution are plotted in Fig. 8 where t = 0 corresponds

Fig. 8. Modulation shift due to 30 ppm concentrations of acetone and hexane.

to the time when each solute is injected into the flow cell. The nor-malized photocurrent shift is 9.96% for hexane, but only 0.16% foracetone, consistent with the partitioning behavior of these solutesinto Teflon AF [23].

4. Analysis and discussion

4.1. Experimental device sensitivity and limit of detection

The device sensitivity is 0.22%/ppm, 0.11%/ppm and 0.07%/ppmfor xylene, toluene and benzene, respectively based on the lines ofbest fit in Fig. 7. In order to calculate the chip’s limit of detection(LOD) for a 5-min measurement time window, which is sufficientto allow nearly full equilibration for all BTX solutes, the data forthe 1 ppm concentration tested (Fig. 9) were analyzed in the fol-lowing manner. First, the change in normalized photocurrent �Mis calculated from t = 0 to the average value for last five data pointsthe sample window. Then the standard deviation � in the nor-malized photocurrent is computed for last five data points in thesample window (t = 4.46–5 min) Per IUPAC definition [36], the limit

Fig. 9. Modulation response to 1 ppm samples of benzene, toluene and xylene.

T.A. Erickson, K.L. Lear / Sensors and A

4

ifautcmstt

b(zt[

C

dfii

1apttaifcatoi

Fig. 10. 30 ppm diffusion curves for xylene, toluene and benzene.

.2. Identifying contaminants from their diffusion coefficients

By employing a diffusion theory model, it is possible to identifyndividual BTX solutes based on extraction of their unique dif-usion coefficients [23] provided that the solution contains only

single dominant contaminant. It must be stated upfront thatsing the diffusion coefficient as a single parameter has its limita-ions for molecular identification, as other potential contaminantsould have similar diffusion coefficients. Whereas absorption basedethods employ multispectral fitting and GC–MS uses both diffu-

ion (time of flight) and molecular mass for analyte determination,he LEAC chip can only employ diffusion data, which is inherent tohe chip’s sensing mechanism and simplified instrumentation.

The diffusion of BTX solutes in Teflon AF films has previouslyeen shown to closely obey the solution to the diffusion equationEq. (1)), where K is the solute’s partition coefficient into the film;film is the film thickness, Co is the solute’s concentration in the con-aminated water sample and D is the solute’s diffusion coefficient37].

(z, t) = KCo − 4KCo

�

∞∑n=0

(−1)n

2n + 1exp

{−D(2n + 1)2�2t

4zfilm2

}

× cos

{(2n + 1)�z

2zfilm

}(1)

Curve fitting [23] can be used to rigorously extract a solute’siffusion coefficient from the data. From the time trace data shownor 30 ppm concentrations of benzene, toluene and xylene in Fig. 10,t is evident that the rate at which each BTX contaminant diffusesnto the film is unique.

The extracted diffusion coefficients are 2.22 × 10−10 cm2/s,.54 × 10−10 cm2/s, and 0.81 × 10−10 cm2/s for benzene, toluenend xylene, respectively. These values are in good agreement withreviously published values [23]. For field measurements, thisechnique would permit a simple approach for discriminating con-aminants based on the measured diffusion coefficient. However,

library of diffusion coefficients for various hydrophobic solutesn Teflon AF would first have to be generated, which is reservedor future work. Further enhancements to the chip’s selectivityould potentially be achieved by employing a molecular imprinting

pproach, whereby the sensing polymer film is tuned to preferen-ially uptake a target solute by adding a small concentration (∼1%)f the solute into the liquid polymer prior to curing. Using molecularmprinting, Dickert et al. were able to show that polystyrene films

ctuators B 204 (2014) 421–428 427

could be modified to selectively uptake o-xylene with a selectivityof 6:1 and 75:1 over p-xylene and m-xylene, respectively [38].

4.3. Comparison to other sensing platforms and ways to improveLOD

The chip’s demonstrated limit of detection is below the EPA’smaximum allowable concentration limit for toluene (1 ppm) andxylenes (10 ppm), but above the maximum allowable level forbenzene (5 ppb). Systems which can resolve benzene at concen-trations of 5 ppb or lower typically require external equipment forpre-concentrating the sample solution [7–11,13]. While externalpre-concentration equipment could be used to further improvethe LEAC chip’s limit of detection, a key advantage of the chip isthat it requires no sample pre-treatment (only a 2 min DI waterbaseline establishment) and has relatively simple external instru-mentation requirements. By employing a grating coupler with aflip-chip laser [39], the chip would require no external optical hard-ware. Only electronic readout circuitry, which could potentially beintegrated onto the chip, would be required, paving the way for atrue lab-on-a-chip groundwater contaminant sensor.

The chip’s demonstrated limit of detection is very comparableto surface acoustic wave [22] (LOD = 500 ppb for benzene), enzy-matic [21] (LOD = 276 ppb for toluene) and absorption spectroscopysystems [14,16,19] (LOD = 100–1200 ppb for xylene) presented inrecent literature, which do require significant external opticalhardware such photomultiplier tubes and external spectrome-ters. While the LEAC chip clearly does not provide the samelevel of multi-analyte specificity as either GC–MS, GC–FID orabsorption-based spectroscopy, measurement of the solute’s dif-fusion coefficient can be used to aid identification of contaminantsin water samples containing a single dominant contaminant. Thisenables the chip to be readily used for exclusionary testing withinthe measured limits of detection. However, it would not necessarilyprevent potential false positives caused the presence of contami-nants with similar diffusion coefficients. Flagged samples wouldneed to be analyzed using other techniques, which offer improvedspecificity.

Ultimately, the chip’s limit of detection depends on its sensi-tivity (%/RIU) and the SNR of the chip’s integrated photodetectors.Assuming [23] a partition coefficient of 800 for xylene (n = 1.46),the chip’s measured sensitivity at 1 ppm is 1804%/RIU which iscomparable to the model simulated value of 1500%/RIU. Simula-tion results indicate that a symmetric waveguide structure (lowercladding with index n = 1.31) with scattering loss of 0.01 dB/cm [40]could provide a sensitivity exceeding 16000%/RIU. By modulatingthe laser and employing a lock-in amplifier [41], photodetectorSNR could be further improved. These two potential improvementscould yield a device with a significantly improved limit of detec-tion, capable of rapid exclusionary sensing of BTX contaminants,potentially at levels below 5 ppb for benzene.

5. Conclusion

We have demonstrated a robust lab-on-a-chip sensing platform,which is capable of rapid (<5 min) exclusionary detection of BTXcontaminants at sub-ppm concentrations, as indicated in Fig. 9.The chip can also be used for broad-based exclusionary testing ofhydrophobic contaminants, such as hexane. A limit of detection of359 ppb, 249 ppb, and 103 ppb has been demonstrated for benzene,toluene and xylene, respectively. Measurement of a contaminant’s

diffusion coefficient can be used to aid in identification. Effects fromother potential interfering contaminants have been characterizedand the on-chip reference region has been shown to render thechip relatively immune to temperature fluctuations. The sensing

4 and A

pocm

A

AaBf

R

[[[

[

[

[

[

[

[

[

[

[

[

[[[

[[[[[[

[[[

[[[

[

[

[

[

28 T.A. Erickson, K.L. Lear / Sensors

latform has potential as a field-deployable measurement systemr for consumers who may benefit from utilizing a relatively low-ost system to monitor water quality from household wells thatay be contaminated with hydrophobic solutes.

cknowledgements

The authors acknowledge the Colorado Cleantech Industriesssociation Foundation, the Advanced Energy Economy Institutend Colorado OEDIT for their financial support. We also thank Ryanrow and Jan Van Zoeghbroeck of the Colorado Nanofabrication Lab

or assistance in operating fabrication equipment.

eferences

[1] M. Smith, Ann. Rev. Public Health 31 (2010) 133–148.[2] V. Benignus, K. Muller, J. Graham, C. Barton, Environ. Res. 33 (1984) 39–46.[3] F. Gamberale, G. Annwall, M. Hultengren, Scand. J. Work Environ. Health 4

(1978) 204–211.[4] Benzene from Gas Plant Leak Polluting Water near Parachute Creek, Denver

Post, 2013, March.[5] Town of Jackson residents ponder aftermath of gasoline spill, Milwaukee J.

Sentinel (September) (2012).[6] D. DiGiuli, Investigation of Ground Water Contamination near Pavillion,

Wyoming, Environmental Protection Agency, Washington, DC, 2011.[7] H. Kim, S. Kim, S. Lee, Talanta 97 (2012) 432–437.[8] H. Faraji, M. Tajbakhsh, M. Helalizadeh, Anal. Methods 4 (2012) 3372–3380.[9] A. Sarafraz-Yazdi, Z. Es’haghi, S. Sepehr, Anal. Methods 2 (2010) 746–752.10] A. Sarafraz-Yazdi, A.H. Amiri, Z. Es’haghi, Talanta 78 (2009) 936–941.11] Y. Assadi, F. Ahmadi, M. Hossieni, Chromatographia 71 (2010) 1137–1141.12] A Review of Emerging Sensor Technologies for Facilitating Long-Term Ground

Water Monitoring of Volatile Organic Compounds, United States Environmen-tal Protection Agency, 2003, August (EPA 542-R-03-007).

13] S. Camou, A. Shimizu, T. Horiuchi, T. Haga, IEEE Sens. 132 (2009)2021–2024.

14] K. Lima, I. Raimundo, M. Pimentel, Sens. Actuators B: Chem. 125 (2007)229–233.

15] J. Buerck, S. Roth, K. Kraemer, M. Scholz, N. Klaas, J. Hazard. Mater. 83 (2001)11–28.

16] W. Lai, S. Chakravarty, X. Wang, C. Lin, R. Chen, Appl. Phys. Lett. 98 (2011)

023304.

17] R. McCue, J. Walsh, F. Walsh, F. Regan, Sens. Actuators B: Chem. 114 (2006)438–444.

18] B. Pejcic, L. Boyd, M. Myers, A. Ross, Y. Raichlin, A. Katzir, R. Lu, B. Mizaikoff,Org. Geochem. 55 (2013) 63–71.

ctuators B 204 (2014) 421–428

19] A. Silva, M. Santana, M. Pimentel, I. Raimundo, Y. Almeida, Sens. Actuators B:Chem. 139 (2009) 222–230.

20] B. Pejcic, M. Myers, N. Ranwala, L. Boyd, M. Baker, A. Ross, Talanta 85 (2011)1648–1657.

21] Z. Zhong, M. Fritzsche, S. Pieper, T. Wood, K. Lear, D. Dandy, K. Reardon, Biosens.Bioelectron. 26 (2011) 2407–2412.

22] F. Bender, F. Josse, R. Mohler, A. Ricco, 14th International Meeting on ChemicalSensors, 2012, pp. 473–476.

23] T. Erickson, R. Nijjar, M. Kipper, K. Lear, Talanta 119 (2014) 151–155.24] A. Fallahkhair, K. Li, T. Murphy, J. Lightw. Technol. 26 (2008) 1423–1431.25] D. Bulla, B. Borges, M. Romero, N. Morimoto, L. Neto, A. Cortes, Proc. SBMO/IEEE

MTT-S, APS, LEOS, Int. Microw. Optoelectron. Conf. 2 (1999) 454–457.26] J. Homola, S. Yee, G. Gauglitz, Sens. Actuators B: Chem. 54 (1999) 3–15.27] W. Kern, J. Electrochem. Soc. 137 (1990) 1887–1892.28] R. Razouk, B. Deal, J. Electrochem. Soc. 126 (1979) 1573–1581.29] M. Turner, E. Rhoderick, Solid State Electron. 11 (1968) 291–300.30] B. Yan, T. Lin, D. Mao, C. Yang, Thin Solid Films 173 (1989) 39–51.31] A. Gorin, A. Jaouad, E. Grondin, V. Aimez, P. Charette, Opt. Express 16 (2008)

13509–13516.32] A. Bulusu, H. Kim, D. Samet, S. Graham, J. Phys. D: Appl. Phys. 46 (2013) 084014.33] M. Saglam, F. Cimilli, A. Türüt, Phys. B: Condens. Matter 384 (2004) 397–403.34] C. Detavernier, R. Van Meirhaeghe, R. Donaton, K. Maex, F. Cardon, J. Appl. Phys.

84 (1998) 3226–3231.35] J. Sangster, J. Phys. Chem. Ref. Data 18 (1989) 1111–1227.36] G. Long, J. Winefordner, Anal. Chem. 55 (1983) 712–724.37] J. Crank, The Mathematics of Diffusion, Eq. 2.67, Oxford University Press, Oxford,

1975.38] F. Dickert, P. Lieberzeit, S. Miarecka, K. Mann, O. Hayden, C. Palfinger, Biosens.

Bioelectron. 20 (2004) 1040–1044.39] C. Liu, Y. Lin, M. Houng, Y. Wang, I.E.E.E. Trans, Compon. Packag. Technol. 26

(2003) 635–641.40] J. Bauters, M. Heck, D. John, D. Dai, M. Tien, J. Barton, A. Leinse, R. Heideman, D.

Blumenthal, J. Bowers, Opt. Express 19 (2011) 3163–3174.41] R. Yan, N. Lynn, L. Kingry, Z. Yi, R. Slayden, D. Dandy, K. Lear, Appl. Phys. Lett.

98 (2011) 013702.

Biographies

Tim A. Erickson received the M.S. degree in biomedical engineering from the Uni-versity of Texas at Austin, Austin, TX, USA, in 2009. He is currently a Ph.D. candidatein electrical engineering under the guidance of Dr. Kevin Lear at Colorado StateUniversity, Fort Collins, CO, USA. His current research interests include design ofportable sensing systems for environmental monitoring and medical diagnostics.

Kevin L. Lear is a Professor of Electrical and Computer Engineering and AssociateDirector of the School of Biomedical Engineering with Colorado State University, FortCollins, CO, USA. His research interests include optoelectronics, portable biosensors,and vertical-cavity surface emitting lasers.