REVIEW ARTICLE Electrochemical deposition of metals onto ......In semiconductor electrochemistry,...

J. Phys. D: Appl. Phys. 31 (1998) 1927–1949. Printed in the UK PII: S0022-3727(98)78690-5

REVIEW ARTICLE

Electrochemical deposition of metalsonto silicon

G Oskam , J G Long, A Natarajan an d P C Searson

Department of Materials Science and Engineering, The Johns Hopkins University,Baltimore, MD 21218, USA

Received 26 November 1997

Abstract. The general concepts governing the electrochemical deposition of metalfilms onto semiconductors are discussed. Deposition onto semiconductor surfacesis complicated due to the band structure of the semiconductor, which affects boththe thermodynamics and the kinetics of metal deposition processes. The influenceof the potential distribution at the semiconductor/solution interface on the chargetransfer mechanisms involved in deposition of metals is discussed. Models forelectrochemical nucleation and growth are described and the influence of theunique physical properties of semiconductors is analysed. Finally, we presentrecent results for electrochemical deposition of gold, copper and platinum onton-type silicon.

1. Introduction

Electrochemical deposition of metals and alloys ontometallic substrates plays an important role in many moderntechnologies. In the electronics industry electrochemicaland electroless deposition are widely used for applicationssuch as copper printed circuit boards, through-holeplating, multilayer read/write heads and thin film magneticrecording media [1, 2]. Surprisingly, there have beenrelatively few reports on electrochemical deposition ofmetals onto semiconductors despite the technologicalimportance of metal/semiconductor contacts for Schottkyjunctions and metallization. The recent drive towardsreplacing aluminium metallization by copper in silicondevice technology has led to a renewal of interest inelectrochemical deposition of metals onto silicon as wellas onto various barrier materials.

The thermodynamics and kinetics of deposition ofmetals onto semiconductor surfaces are complicated bya number of factors. The interaction energy betweenmany semiconductors and metals is relatively weak so thatdeposition of metals onto semiconductor surfaces usuallyfollows a 3D island growth (Volmer–Weber) mechanism.The deposition process may occur via the conduction band,via the valence band or via surface states. For mostsemiconductors, open circuit (electroless) deposition occursvia the valence band and involves holes, which may leadto oxidation of the semiconductor surface. Electrochemicaldeposition onto n-type semiconductors involves electronsin the conduction band and is usually preferred since thenucleation and growth processes are dependent on theapplied potential and can be controlled externally. The rateof charge transfer at semiconductor surfaces is often slower

than that at metal surfaces of due to the lower surfaceelectron density which can significantly affect propertiessuch as the adhesion and texture of deposits.

In this paper, we review the thermodynamic and ki-netic aspects of deposition of metals onto semiconductors.We discuss the potential distribution at the semiconduc-tor/solution interface, mechanisms for transfer of chargefrom semiconductors to metal ions in solution, semiconduc-tor/metal contacts and the mechanisms of nucleation andgrowth. In addition, we summarize recent results on thedeposition of copper, gold and platinum onto silicon.

2. Semiconductor/solution interfaces

Figure 1 shows an energy band diagram for an n-typesemiconductor in contact with a solution. For convenience,the redox couple is considered as donor and acceptor stateswith a Gaussian distribution of energy levels. The mid-point between the acceptor and donor states correspondsto the equilibrium energy,EO/R. Under equilibriumconditions the Fermi energy in the semiconductor is alignedwith the equilibrium energy of the solution, resulting inthe formation of a space charge layer at the semiconductorsurface and a counteracting charge in the solution. Thebuild up of charge on both sides of the interface leadsto a potential drop over the space charge layer in thesemiconductor and over the Helmholtz layer on the solutionside of the interface. The potential drop across the spacecharge layer is equivalent to the built-in potential formedat a Schottky junction. In most cases of practical interest,under equilibrium conditions the surface is depleted ofmajority carriers and a depletion layer is formed.

0022-3727/98/161927+23$19.50 c© 1998 IOP Publishing Ltd 1927

G Oskam et al

Figure 1. The energy band diagram for an n-typesemiconductor in contact with a solution, illustrating theband bending in the semiconductor and the temporaldistribution of energy levels of the electron donor andacceptor states in solution. The mid-point between theacceptor and donor states corresponds to the equilibriumpotential of the redox couple, EO/R .

The theory of metal-on-metal deposition is wellestablished [3–8]; however, the theory of deposition ofmetals onto semiconductors has not been well developed.In the context of the Gerischer model for the transferof charge at semiconductor/electrolyte solution interfaces[9, 10] several situations can be identified. Figure 2illustrates possible mechanisms for deposition of metalsonto an n-type semiconductor. Figure 2(a) shows thesituation for a metal/metal ion redox couple with asufficiently negative equilibrium potential (for example atransition metal) so that the acceptor levels have a largeoverlap with the conduction band edge. If the surfaceelectron concentration in the conduction band is sufficientlyhigh, electrons can be transferred to the metal ion insolution, resulting in deposition of the metal. In figure 2(b),electrons are transferred to the metal ion from surface statesin the band gap of the semiconductor. Figure 2(c) showsthe situation for a metal/metal ion couple with a positiveequilibrium potential (for example a noble metal), in whichthe acceptor states overlap with the valence band of thesemiconductor. In this case, holes can be injected directlyinto the valence band, resulting in deposition of the metal.In the next section, we review the potential distributionat the semiconductor/solution interface and in subsequentsections we analyse the cases identified in figure 2.

2.1. Partitioning of the applied potential

The potential distribution at the semiconductor/solutioninterface is more complicated than that at the metal/solutioninterface since the applied potential is partitioned betweenthe space charge layer of the semiconductor and theHelmholtz layer in solution. Under equilibrium conditionswith no applied potential the Fermi levels of the

semiconductor and the solution are equal, and the exchangecurrent between the semiconductor and solution is due tothe reversible redox process. If a reversible redox coupleis not present in the solution, the partial oxidation andreduction reactions may be derived from different processesand, hence, define steady state conditions (at open circuit).In the absence of any other potential drops in the system(such as across an interfacial layer such as an oxide oracross a diffuse double layer in solution), the total potentialdrop across the interface,1φtotal , is given by

1φtotal = 1φsc +1φH (1)

where1φsc is the potential drop across the space chargelayer of the semiconductor and1φH is the potential dropacross the Helmholtz layer.1φsc and1φH are defined as

1φsc = φssc − φbsc (2a)

1φH = φsoln − φssc (2b)

whereφssc is the electrostatic potential at the semiconductorsurface,φbsc is the potential in the bulk of the semiconductorandφsoln is the potential of the bulk solution. Hence, forthe situation shown in figure 1,1φsc < 0 since the chargeon the semiconductor is due to (positive) ionized donoratoms (depletion). In the absence of an interfacial layerand adsorption of charged species,1φsc and1φH have thesame sign. The band bending is defined ase1φsc, wheree is the electronic charge. The energy of an electron inthe space charge layer is given by−eφ(x) where φ(x)corresponds to the electrostatic potential as a function ofthe distance from the surface.

We note that charged surface states, adsorption ofcharged species and storage of charge in interfacial (forexample oxide) layers can significantly influence thepotential distribution at the interface. For example, formany metal oxide or covalent semiconductors the H+/OH−

adsorption equilibrium results in a Nernstian dependenceof the position of the band edges on the pH [11]. Inthis section we analyse the case in which adsorption andthe presence of interfacial layers or surface states can beneglected; the analysis can be extended to include theseeffects but they are not discussed here.

In semiconductor electrochemistry, the applied poten-tial, U , is generally referenced to the potential at whichthe bands are flat,U0

f b. For the case considered here, atan applied potentialU = U0

f b, the potential drops acrossthe space charge layer and the Helmholtz layer are equalto zero,1φsc = 1φH = 0. Note that, for a solid statemetal/semiconductor Schottky junction, the applied poten-tial is referenced to the built-in potential. The appliedpotential is partitioned over the space charge layer andHelmholtz layer and can be related to the electrostatic po-tential drops at the interface as follows:

U − U0f b = φbsc − φsoln = −1φtotal . (3)

Note that, in the presence of an interfacial layer, surfacestates or adsorbed species,1φtotal 6= 0 at U = U0

f b.The partitioning of the applied potential over the spacecharge layer and Helmholtz layer is dependent on the

1928

Electrochemical deposition of metals onto silicon

Figure 2. Energy band diagrams illustrating possible mechanisms of deposition of a metal onto an n-type semiconductor.(a) Transfer of electrons from the conduction band to a metal/metal ion couple with a sufficiently negative equilibriumpotential, (b) transfer of electrons from surface states and (c) injection of holes into the valence band from a metal/metal ioncouple with a sufficiently positive equilibrium potential.

relative magnitudes of the differential capacitance of thespace charge layer,Csc, and the differential capacitanceof the Helmholtz layer,CH , as shown in figure 3. Forpotentials at which the semiconductor is in deep depletion(figure 3(a)),Csc is usually much smaller thanCH . As aconsequence, when the potential is shifted to more negativevalues, any change in the applied potential is dropped overthe space charge layer. In this case, the potential drop overthe Helmholtz layer is independent of the applied potentialand the positions of the band edges at the surface are fixed(figure 3(b)). When the magnitude of the space charge layercapacitance approaches that of the Helmholtz layer, whichoccurs in weak depletion or in accumulation, a change inthe applied potential is partitioned between the space chargelayer and the Helmholtz layer (figure 3(c)). The fraction ofthe applied potential dropped across the space charge layercan be obtained in the following way. Under steady-stateconditions at any applied potential, conservation of chargeat the semiconductor/solution interface (assuming that thedensity of charged surface states is sufficiently low) can beexpressed as

Qsc −QH = 0 (4)

whereQsc is the charge in the space charge layer andQH

is the charge in the Helmholtz layer. The charge in theHelmholtz layer can be obtained from the Helmholtz layercapacitance and the potential drop over the Helmholtz layer.By definition,CH = |dQH/d1φH | and, if it is assumed thatthe Helmholtz capacitance is independent of1φH [11, 12];then

QH = CH1φH = −CH [(U − U0f b)+1φsc]. (5)

The charge in the semiconductor space charge layer isproportional to the electrical field at the surface,ξ s , and canbe obtained from Gauss’ law and the Poisson–Boltzmannequation [13]:

Qsc = εε0ξs = ∓(2kT εε0ND)

1/2

×[

exp

(e1φsc

kT

)− e1φsc

kT− 1

]1/2

(6)

where ε0 is the permittivity of free space,ε is therelative permittivity,ND is the donor density (for an n-typesemiconductor assuming complete ionization),k is theBoltzmann constant andT is the temperature. The minussign corresponds to accumulation (1φsc > 0) and thepositive sign corresponds to depletion (1φsc < 0).

Substituting equations (5) and (6) into equation (4)gives

−CH [(U − U0f b)+1φsc] ∓ (2NDεε0kT )

1/2

×[

exp

(e1φsc

kT

)− e1φsc

kT− 1

]−1/2

= 0. (7)

The dependence of the band bending,1φsc, on the appliedpotential,U , can be obtained by solving equation (7) fora given value ofCH . It is convenient to define the dcpotential drop over the space charge layer as a function ofthe applied dc potential in terms of the coefficientγdc:

γdc = −1φscU − U0

f b

(8)

whereγdc is between 0 and 1. The flat band potential,U0f b,

corresponds to the condition1φsc = 0. In practice, the flatband potential is usually obtained by extrapolation fromcapacitance measurements in deep depletion. However,in some cases, the value for the flat band potential,Ufb,obtained by this method is not equal toU0

f b (that is1φsc 6= 0 at U = Ufb) due to unpinning of the bandedges. Band edge unpinning generally occurs in the weakdepletion regime as a result of surface reactions (suchas conversion of a hydrogen-terminated to a hydroxyl-terminated surface or vice versa [14]), filling and emptyingof surface states [15–17], charging of an interfacial layer[18, 19] or partitioning of the applied potential as themagnitude ofCsc approaches that ofCH . In this paper,we consider only the last of these mechanisms.

The experimentally measured differential capacitance,C = |dQ/dU |, is usually determined as a function of theapplied potential. The capacitance is generally measuredusing either a small signal transient technique or byrecording the response of the current to a small periodic

1929

G Oskam et al

Figure 3. The energy band diagram for an n-type semiconductor in contact with a solution. (a) The potential distribution atthe semiconductor/solution interface. Any change in the applied potential may result in a change of the potential drop acrossthe space charge layer, 1φsc , which is illustrated in (b), or across the Helmholtz layer, 1φH , as shown in (c). In case (c), theband bending is unchanged but the band edges at the surface are shifted to a more negative potential (corresponding tohigher energy).

potential perturbation (for example, in electrochemicalimpedance spectroscopy). At any applied potential,U , achange in the applied potential, dU , is partitioned over thespace charge layer and the Helmholtz layer. Neglectingadsorption, interfacial layers and surface states, we canwrite

−dU = d1φtotal = d(1φsc)+ d(1φH ). (9)

The partitioning of the applied potential, dU , between thespace charge layer and the Helmholtz layer is determinedby the relative magnitudes of the differential capacitancesof the two regions. The non-steady-state partitioning canbe described in terms of the fraction of the applied potentialchange, dU , dropped across the space charge layer,γac:

γac = −d1φscdU

(10a)

1− γac = −d1φHdU

(10b)

where γac is between 0 and 1. Using the definitionsfor the differential capacitances of the Helmholtz layer(CH = |dQH/d1φH |) and the space charge layer(Csc = |dQsc/d1φsc|), it can easily be shown that

γac = CH

CH + Csc . (11)

The relationship betweenγac and γdc can be obtained bydifferentiation of equation (8):

γac = γdc + (U − U0f b)

dγdcdU

. (12)

Both γac andγdc are important parameters for analysisof the charge transfer kinetics at the semiconductor/solutioninterface since they determine how the potential ispartitioned across the interface. The parameterγac isimportant in the analysis of measurements involving a smalltransient or a periodic perturbation, such as electrochemicalimpedance spectroscopy. In addition,γac determines, inpart, the slope of current–potential curves (see the next

Figure 4. The fraction of the applied potential droppedover the space charge layer both for dc conditions, γdc , andfor ac conditions, γac , versus the applied potential (withrespect to the potential at which the bands are flat). Theresults shown are for CH = 1 µF cm−2 and ND = 1015 cm−3.

section). The parameterγdc is important in analysing thekinetics of reactions under steady state conditions, wherethe rate of reaction may be dependent on the magnitude ofthe band bending or the overpotential.

Figure 4 is a plot ofγac and γdc versus the appliedpotential for n-type silicon with a donor density of1015 cm−3 calculated for a Helmholtz layer capacitanceof 1 µF cm−2. We note that values forCH of about1–3µF cm−2 have been reported in the literature [18–22].From figure 4 it can be seen thatγdc decreases stronglyat applied potentials negative of the flat band potential,illustrating that the band edges become unpinned and mostof the applied potential is dropped over the Helmholtz layer.

Figure 5 shows the band bending as a function of theapplied potential for various values ofCH illustrating thatthe band bending saturates at sufficiently negative potentialsas the band edges become unpinned. It can be seen that,

1930


Figure 5. The band bending, 1φsc , as a function of theapplied potential for three values of the capacitance of theHelmholtz layer: (a) 10 µF cm−2, (b) 3.5 µF cm−2 and(c) 1 µF cm−2. The results shown are for n-type silicon withND = 1015 cm−3.

for CH = 1 µF cm−2, the maximum band bending in theaccumulation regime is limited to about 300 mV.

The differential capacitance of the space charge layeris defined asCsc = |dQsc/d1φsc|. A general expressionfor Csc can be obtained by differentiation of equation (6):

Csc =(e2NDεε0

2kT

)1/2∣∣∣∣[ exp

(e1φsc

kT

)− 1

]∣∣∣∣×[

exp

(e1φsc

kT

)− e1φsc

kT− 1

]−1/2

. (13)

Under deep depletion conditions such that1φsc � 0,Csc ∝ 1/

√1φsc. For a moderately doped semiconductor,

Csc � CH and the applied potential is mainly droppedacross the semiconductor space charge layer (γac = 1 andγdc = 1). In this potential range,1φsc can be replaced byU − Ufb:

1

C2=(

2

eNDεε0

)(U − Ufb − kT

e

)(14)

where C is the measured capacitance which, in thispotential regime, is equal to the capacitance of the spacecharge layer,Csc. Equation (14) is often used for thedetermination of the flat band potential,Ufb, the donordensity,ND, and the position of the band edges.

In the accumulation regime at sufficiently large bandbending (1φsc > 3kT /e), equation (13) reduces to

Csc =(e2NDεε0

2kT

)1/2

exp

(e1φsc

2kT

). (15)

Equation (15) is applicable as long as the surface is non-degenerate; that is, as long as1φsc is not too large andNDis relatively small. For larger values of1φsc (typically1φsc > 300 mV for ND ≈ 1015 cm−3), the Fermi

distribution function must be used in solving the Poissonequation [23]. Note that1φsc reaches a magnitude of only300 mV at applied potentials more negative than−0.5 V(versusU0

f b), depending on the value used forCH (seefigure 5).

At potentials close to the flat band potential,Cscapproaches the value ofCH and the applied potential ispartitioned between the two double layers (γdc < 1). Underthese conditions,γdc (and hence1φsc) must be known inorder to obtainCsc from equation (13). In semiconductorelectrochemistry, the electrode potential is often givenwith respect to the flat band potential determined in deepdepletion using equation (14); however, we emphasizethat the flat band potential,Ufb, is a potential dependentparameter since any shift in the band edges, due to surfacereactions or partitioning of the applied potential, results ina shift in the flat band potential.

2.2. Charge transfer at semiconductor surfaces

In this section, we discuss the influence of the partitioningof the applied potential on the charge transfer kineticsat semiconductor surfaces for the three cases shown infigure 2.

2.2.1. Case 1: reduction of metal ions via the conductionband. We begin by considering the two limiting cases ofspace charge layer control (γac = 1), in which the rate ofreaction is controlled by the potential drop in the spacecharge layer, and Helmholtz layer control (γac = 0), inwhich the rate is controlled by the potential drop in theHelmholtz layer. We then derive general expressions forthe influence of the partitioning of the applied potential onthe charge transfer kinetics at semiconductor surfaces.

2.2.1.1. Space charge layer control. For the case inwhich a change in the applied potential is dropped acrossthe space charge layer of the semiconductor (γac = 1), therate of reaction is determined by the overlap integral of thedensity of states of the electron acceptor in solution andthe electron concentration in the conduction band at thesurface of the semiconductor. The current due to majoritycarriers in the conduction band for an n-type semiconductor,icb, for a single-electron charge transfer process is given by[9, 10]

icb = ek+cbNCNred − ek−cbnsNox (16)

where k+cb and k−cb are the rate constants for oxidationand reduction, respectively,NC is the effective density ofstates in the conduction band,Nred is the density of theelectron donors in solution,ns is the electron density at thesemiconductor surface andNox is the density of electronacceptors in solution. The concentration of electrons at thesurface,ns , is given by

ns = n0 exp

(e1φsc

kT

)(17)

where n0 is the bulk electron concentration (n0 = NDfor completely ionized donors). If the rate of theoxidation reaction (ek+cbNCNred ) is sufficiently small, then

1931

G Oskam et al

the conduction band current is determined by the rate ofthe reduction reaction:

icb = −ek−cbNoxn0 exp

(e1φsc

kT

). (18)

If the applied potential,U , is dropped over thesemiconductor and the positions of the band edges at thesurface are fixed then the potential drop over the spacecharge layer,1φsc, can be replaced byU − Ufb, whereUfb is the flat band potential. In this potential range,γacand γdc are equal to 1 andns increases with an inverseslope of 60 mV per decade according to equation (17).Therefore, for a charge transfer reaction under space chargelayer control, the inverse slope of a plot of log(i) versusthe potential is 60 mV per decade. This is expected to bethe case when a depletion layer is formed at the surface(1φsc � 0), namely at potentials positive to the flat bandpotential for an n-type semiconductor.

2.2.1.2. Helmholtz layer control. For the case in whichthe applied potential is dropped over the Helmholtz layer(γac = 0), the reaction rate for a single-electron transferprocess can be described by the Butler–Volmer equation[11]:

i = −i0[

exp

(− αceη

kT

)+ exp

((1− αc)eη

kT

)](19)

whereαc is the cathodic transfer coefficient andη is theoverpotential. For negative overpotentials, at which thereduction reaction dominates, equation (19) reduces to

i = −i0 exp

(− αceη

kT

). (20)

The exchange current density,i0, is given by

i0 = eNAvk0(Nox)αc (Nred)

1−αc (21)

whereNAv is Avogadro’s number,k0 is the rate constantat the standard equilibrium potential andNox and Nredare the density of the electron donor and acceptor levels,respectively.

The overpotential,η, is defined as

η = 1φH −1φeqH (22)

where 1φeqH is the electrostatic potential drop in theHelmholtz layer at the equilibrium potential,Ueq (=EO/R,see figure 1). For many redox couples at metal electrodesαc is found to be about 0.5, leading to an inverse slope on aplot of log(i) versus potential of about 120 mV per decade.

2.2.1.3. The general case.The current–potential curvefor a semiconductor in contact with a solution is given bythe total differential of the current density,i, with respectto the applied potential [16, 17]:

di

dU= ∂i

∂1φsc

∂1φsc

∂U+ ∂i

∂1φH

∂1φH

∂U. (23)

Under depletion conditions, the second term on the right-hand side in equation (23) can be neglected. This situation

corresponds to space charge layer control. In weakdepletion or accumulation, both terms have to be taken intoaccount.

Using equations (10), (18) and (20), the inverse slope,dU/d log(i), is given by

dU

d log(i)= 2.303

kT

e

1

γac + αc(1− γac) . (24)

In deep depletionγac = 1 and the inverse slope is 60 mVper decade, corresponding to the potential dependenceof the density of electrons at the surface, given by theBoltzmann equation (see equation (17)). In the limitingcase in whichγac = 0 (CH � Csc), the inverse slope ofthe log(i) versus potential plot is equal to 2.303kT /(eαc) asdefined in the Butler–Volmer equation (see equation (19)).For intermediate cases, the inverse slope increases from60 mV per decade (γac = 1) to 120 mV per decade (forαc = 0.5). For example, in a situation in which half ofthe applied potential is dropped over the Helmholtz layer(γac = 0.5) andαc = 0.5, the inverse slope is equal to80 mV per decade. At potentials close toUfb, γac is astrong function of the applied potential so that the inverseslope is expected to be potential dependent. However, for(U − U0

f b) < −0.5 V, γac is weakly dependent on thepotential, as shown in figure 4, so that a plot of log(i)

versusU is expected to be linear.This analysis shows that in order to determine ki-

netic parameters associated with reactions at semiconduc-tor/solution interfaces, the partitioning of the applied po-tential across the space charge and Helmholtz layers mustbe known. For the specific case in which a change in ap-plied potential is dropped across the space charge layer, theMarcus–Gerischer model may be used to determine the rateconstants if the following conditions are satisfied: (i) thesurface electron concentration can be calculated from anal-ysis of the position of the band edges in deep depletion (forexample Mott–Schottky analysis); (ii) the energy of the re-dox couple is such that the current–voltage curves can beanalysed in weak depletion (usually at least 200–300 mVfrom the flat band potential); (iii) the density of surfacestates is sufficiently low; and (iv) the surface is stable inthe solution. As a consequence, quantitative determinationof rate constants at semiconductor/solution interfaces hasbeen limited to specific cases; see for example [24–28].

Equation (24) is a general expression for charge transferat semiconductor electrodes involving outer sphere redoxcouples. For inner sphere reactions such as reduction ofmetal ions, the situation is more complicated since surfaceinteractions must be taken into account. Nonetheless, sincedeposition of metals onto semiconductors usually occursunder conditions of weak depletion or accumulation, thepartitioning of the applied potential is expected to have asignificant effect on the rate of reduction of metal ions.Other factors, such as the mechanism of nucleation andgrowth, are discussed in more detail in section 4.

2.2.2. Case 2: reduction of metal ions via surfacestates. Figure 6 is an energy band diagram for depositionof a metal onto an n-type semiconductor via electrically

1932


Figure 6. The energy band diagram for deposition of ametal via surface states at an n-type semiconductor.

active, monoenergetic surface states. If the surface statesare located at an energy above midgap, interaction withthe valence band can be neglected. The occupancy of thesurface states by electrons is dependent on the position ofthe Fermi level and hence on the applied potential. Thecapture of electrons from the conduction band (CB) bysurface states, thermal de-trapping and transfer of electronsto the solution can be described by the following equations:

SS0+ e−(CB)k1→ SS− (25a)

SS−k2→ SS0+ e−(CB) (25b)

SS− +Oxk3→ SS0+ Red (25c)

where k1 and k2 are the rate constants for filling andemptying the surface states, respectively, andk3 is the rateconstant for transfer of electrons to the solution.

As the potential is shifted in the negative directionand surface states become filled, an additional charge ofmagnitude−es− appears at the semiconductor surface. Asa consequence, under weak depletion conditions the totalpositive charge on the n-type semiconductor is reduced.In order to maintain electroneutrality, the charge onthe solution also decreases, resulting in a decrease inthe absolute magnitude of the potential drop across theHelmholtz layer. This, in turn, causes a shift of the bandedges to higher energy. The change in the potential dropacross the Helmholtz layer,1φssH , can be estimated from

1φssH =es−

CH. (26)

For example, for a Helmholtz capacitance of 1.6 µF cm−2

and a density of filled surface states of 1010 cm−2, the shiftof the band edges is 1 meV and can be neglected. However,for surface state densities of 1012 cm−2 or more, the shiftof the band edges is of the order of 100 meV and mustbe to taken into account in the analysis of the potentialdistribution at the interface and the charge transfer kinetics[16, 17].

The time dependence of the density of filled surfacestates,s−, can be described by

ds−(t)dt= k1s

0(t)ns(t)− k2s−(t)− k3s

−(t)Nox (27)

Figure 7. The density of surface states as a function of pHfor n-Si(111) with ND = 4× 1013 cm−3 in 1 M fluoridesolutions.

wheres0 is the density of empty states. Under steady-stateconditions it follows that

s− = k1ns

k1ns + k2+ k3Noxstot (28)

where stot is the total density of interface states(stot = s− + s0).

For deposition of metals via surface states, thenucleation process involves transfer of electrons fromoccupied surface states to the metal ion in solution with arate proportional tok3s

−Nox . As a result, this mechanismis expected to be important for deposition under conditionswhere the surface states are filled, corresponding to thesituation in which the semiconductor Fermi level is abovethe energy of the surface states. At potentials close to theflat band potentialk1ns is large (see equation (17)) andas long ask1ns � k2 and k1ns � k3Nox then all surfacestates are filled (s− ≈ stot ) and the electron transfer stepis the rate limiting step. If the rate of transfer of chargefrom the surface states is sufficiently large then the surfacestates may act as nucleation sites for deposition of a metal;this would imply that the maximum nucleus density fordeposition of a metal via surface states is limited bystot .

For n-type silicon in aqueous fluoride solutions,monoenergetic surface states are present at an energy ofabout 0.4 eV below the conduction band edge at the surface[15, 22, 29]. The density of surface states is dependent onthe pH as shown in figure 7. At low pH the density ofsurface states is of the order of 2× 1010 cm−2, implyinga maximum nucleus density of the order of 1010 cm−2.The ability of surface states to act as nucleation sitesfor deposition of a metal is dependent on the electronicand chemical properties of the states. To the bestof our knowledge, there have been no reports showingunambiguously that deposition occurs via surface states.

In photocatalysis, metal islands on p-type semiconduc-tors [30–41] have been shown to catalyse charge transfer

1933

G Oskam et al

processes such as the reduction of water. In this case themetal islands can be viewed as electronic surface states.Similarly, after nucleation of metal islands on a semicon-ductor surface it is likely that further growth takes placeat the existing islands which can be interpreted as chargetransfer through surface states.

2.2.3. Case 3: reduction of metal ions via the valenceband. For noble metal/metal ion couples with highlypositive equilibrium potentials, the energy states of themetal ion are expected to overlap with the valence bandand deposition can occur in the dark through injection ofholes from the metal ion in solution, both on p-type andon n-type silicon. For a p-type semiconductor, the injectedholes can either be transported to the external circuit, givingrise to a direct current, or they can be involved in anoxidation process. For an n-type semiconductor, holes canrecombine with electrons in the conduction band, resultingin a deposition current (if the band bending is not too large),or, under conditions of deep depletion, the holes can beinvolved in an oxidation process such as etching of thesemiconductor. If the positions of the band edges remainfixed during the deposition process, then, neglecting theback reaction, the deposition current is independent of theapplied potential and is given by

ivb = ekvbNred (29)

whereivb is the current via the valence band andkvb is therate constant.

For the case of deposition of a metal via the valenceband, if the holes injected from the metal ion canbe consumed by an oxidation process, deposition canoccur without an externally applied bias. The possiblemechanisms for zero-current or electroless deposition areshown in figure 8: (i) the injected holes are consumed byoxidation of the semiconductor surface so that substrateatoms are replaced by metal atoms (displacement plating);(ii) the holes injected during deposition of metal aretransferred to an electron donor in the solution; and (iii) themetal ion and the reducing agent in the solution reactdirectly at catalytic sites at the surface without involvementof the substrate in the charge transfer process.

In the first case (figure 8(a)), the oxidation productsmust be soluble in the solution in order to prevent theformation of a passivating layer. For silicon this is possiblein low-pH fluoride solutions. In principle, the thicknessof the metal film is limited since the deposition processcannot continue once complete coverage is achieved. Themorphology and adhesion of the deposit may be poor dueto the simultaneous dissolution of the semiconductor anddeposition of the metal. The second case (figure 8(b))is not often encountered for the deposition of metalsonto semiconductors since the rate of transfer of holesto the solution is usually slower than the trapping ofholes at surface atoms, which is the first step in theoxidation mechanism. This situation is similar to thestabilization of semiconductor surfaces under illuminationin photoelectrochemical energy conversion [12, 42, 43].The third case (figure 8(c)) is often used in electroless

Figure 8. Schematic illustrations of the three mechanismsof electroless metal deposition: (a) holes injected frommetal ions (M+) are consumed by oxidation of thesemiconductor and, if the products (Sn+) are soluble,dissolution of the semiconductor (displacement plating);(b) injected holes are captured by an electron donor in thesolution; and (c) metal ions are catalytically reduced at anexisting metal nucleus involving electrons donated by areducing agent in the solution.

deposition of metals onto semiconductors. In general, smallclusters of palladium or some other activating metal aredeposited by the displacement mechanism (figure 8(a)) andthen act as catalytic sites for the reduction of the metal ionand oxidation of the reducing agent in solution. For thismethod to be successful and result in a continuous film, thedensity of catalytic centres should be high.

Electroless deposition of metals onto silicon is ofinterest in the electronics industry as a method fordeposition of metals, but also because metal ion impuritiesin wafer cleaning solutions may result in unwanteddeposition of metal nuclei that induce surface and electronicdefects. For example, trace concentrations of metalion impurities (such as Cu2+) in HF solutions can leadto electroless deposition of metal clusters that severelydegrade the performance of gate oxides and dynamic RAMcells [44–46]. The deposition usually proceeds throughinjection of holes into the valence band and simultaneous,often localized, dissolution of silicon (mechanism (i)) [47–49]. From figure 2(c) it can be seen that electrolessdeposition of metal impurities can occur only when themetal ion donor states overlap with the valence band and,as a result, this process is usually limited to noble metals.

Electroless deposition of copper onto silicon has beenreported in the literature [48, 49], although the mechanismis more complicated insofar as the copper ion levelsdo not exhibit good overlap with the valence band.Consequently, low levels of illumination dramatically

1934


increase the nucleation of copper both on n-type and onp-type silicon [49].

Electroless deposition of gold films onto III–Vsemiconductors (GaAs, InP, GaP and AlxGa1−xAs) hasbeen achieved on surfaces activated with palladium nuclei,where the palladium was deposited by electroless deposition(mechanism (i)) [50]. In these cases, deposition of goldonto the palladium clusters involved the simultaneousoxidation of KBH4 according to mechanism (iii). Filmsdeposited onto silicon using the same technique, however,exhibited poor adhesion [50]. Electroless nickel filmscan be deposited directly onto silicon without activation[51, 52], although pre-roughening of the surface has beenfound to be essential to obtain good adhesion. Electrolessdeposition of gold, copper, tin, platinum and palladiumonto p-type silicon through the displacement mechanismdoes not yield adherent metal films [53, 54]. There is alsoevidence for formation of silicides in some cases [54]. Ithas recently been shown that epitaxial copper films can beprepared by electroless deposition onto a silicon surfacecovered with an evaporated epitaxial copper seed layer[55, 56].

These results illustrate the difficulty in using electrolessplating for the deposition of thin metal films onto silicon.In the following sections we focus on the electrochemicaldeposition mechanism since it allows greater controlover the deposition process and may be more successfulfor the deposition of thin continuous metal films ontosemiconductor surfaces.

3. Silicon/metal contacts

Metal/semiconductor junctions are essential building blocksfor microelectronic devices. The properties of Schottkyjunctions produced by sputter deposition or evaporationhave been studied extensively over the past 20 years[57]. In contrast, there have been very few reportson the electrochemical deposition of Schottky junctionsand most of these have been related to II–VI and III–Vsemiconductors [58–62]. In many cases, the barrier heightsof electrochemically fabricated Schottky junctions havebeen found to be higher than those prepared by physicaldeposition techniques due to the presence of thin interfaciallayers (for example, of oxide).

The quality of a Schottky diode is determined by thebarrier height and the ideality factor. The barrier heightis, in principle, determined by the difference between thework functions of the metal and the semiconductor butfor covalent semiconductors it is generally found that thebarrier height is determined by interface states that pin theFermi level [57]. The method of deposition can stronglyinfluence the interfacial chemistry and, hence, the barrierheight.

Experimentally, the barrier height can be determinedfrom the flat band potential, which can be obtained fromimpedance measurements in the deep depletion regime(reverse bias) according to the Mott–Schottky relation(equation (14)). The barrier height,8B , is given by

8B = |eUfb| + ξ + kT (30)

where ξ is the energy gap between the conduction bandedge (for an n-type semiconductor) and the Fermi level.

The barrier height can also be obtained fromcurrent–potential measurements in the weak depletionregime (forwards bias). According to the thermionicemission model [57], the relationship between the currentdensity,i, and the applied potential,U , is given by

i = −i0 exp

(− eUnkT

)[1− exp

(eU

kT

)](31)

where

i0 = A∗∗T 2 exp

(−8B

kT

)(32)

where i0 is the saturation current density andA∗∗ isthe Richardson constant. The ideality factor,n, givesan indication of the electronic quality of the interface.For an ideal Schottky diode (n = 1), a plot ofln{−i/{1− exp[eU/(kT )]}} versusU is linear with aninverse slope of 25.7 mV. The saturation current,j0, isobtained from the intercept and the barrier height can bedetermined from equation (32). A review of the electronicproperties of silicon/metal junctions can be found in [57].

4. Nucleation and growth

The mode of growth for deposition of a metal onto a foreignsubstrate is dependent on the interaction energy between theadsorbed metal atom and the substrate and the differencein interatomic spacing between the bulk metal phase andthe substrate [63]. In general, for any growth process thegrowth mechanism and the structure of the deposited filmare determined by the relative rates of the particle fluxand surface diffusion. Three different modes of growthcan be identified: layer by layer (Frank–van der Merwegrowth), 3D island formation (Volmer–Weber growth) and2D layer deposition followed by the growth of 3D islands(Stranski–Krastanov growth). In many cases, deposition ofmetals onto semiconductors follows a 3D island formationmechanism of growth due to the weak interaction energybetween the adsorbed metal atom and the semiconductor.

4.1. Mechanisms of nucleation and growth

The 3D island growth (Volmer–Weber) mechanism hasbeen exploited in the deposition of metal islands ontosemiconductor surfaces for photocatalysis. For example,platinum islands deposited onto p-InP and p-Si have beenshown to enhance the rate of evolution of hydrogen andphotoelectrochemical cells with efficiencies up to 12% havebeen fabricated [30–33]. The density of electrochemicallydeposited metal islands on semiconductor surfaces istypically of the order of 107–1011 cm−2 [64–68].

The rate law for growth of 3D islands duringelectrochemical deposition is dependent on the mechanismof nucleation and growth. Models for electrochemicaldeposition onto a foreign substrate usually assume thatnucleation occurs at certain specific sites on the surface[1, 3] and the nucleation mechanism is generally describedin terms either of instantaneous or of progressive

1935

G Oskam et al

nucleation. If the rate of nucleation is fast in comparisonwith the subsequent rate of growth, then nuclei are formedat all possible growth sites within very short times andnucleation is considered instantaneous. Conversely, if therate of nucleation is slow, then nucleation will continueto occur at the surface while other clusters are growingand nucleation is considered progressive; for the simplecase of first-order nucleation kinetics, the number ofnuclei increases linearly with time. We note, however,that, although experimental data have exhibited excellentagreement with these models, the nature of the nucleationsites has not been determined.

4.1.1. The potential dependence of nucleation. For 3Disland growth, electrochemical deposition of metals ontoeither semiconductor or metal surfaces proceeds throughnucleation of metal clusters and subsequent growth of a film[63]. The thermodynamics of electrochemical nucleation issimilar to that of nucleation from the gas phase. In bothcases, the formation of thermodynamically stable nucleioccurs in the supersaturation regime. For electrochemicalnucleation, the supersaturation can be described in terms ofthe difference between the electrochemical potentials of thesolid metal,µM(s), and the dissolved metal ions in solution,µMn+(aq). Experimentally,1µ can be directly controlled bythe applied potential and the following relation holds:

1µ = µM(s) − µMn+(aq) = −ze|η| (33)

wherez is the number of charges involved in the electro-chemical process and|η| is the absolute overpotential; notethat η is negative for (overpotential) metal deposition. Forsemiconductor electrodes, the term overpotential is gener-ally not used since the applied potential is partitioned acrossthe space charge layer and the Helmholtz layer. As a con-sequence, the potential dependences of the charge transferprocesses are significantly different from those at metal sur-faces, as was described in section 2.2.1. We will addressthis issue in more detail in section 6.3.4.

In general, a nucleus needs to attain a critical size sothat further growth is energetically more favourable thandissolution. The Gibbs free energy for the formation of acluster ofN atoms,1G(N), is given by

1G(N) = N1µ+8(N) (34)

where 8(N) is an excess energy term associated withthe formation of the new interfaces and depends on thesurface energies of the substrate and the cluster.8(N)

may also contain terms associated with internal strain due tothe lattice mismatch between the substrate and the cluster.Both terms in equation (34) are functions of the numberof atoms in the cluster. For small clusters,8(N) usuallydominates and, as a consequence,1G(N) increases withincreasingN . The formation of a cluster can, therefore,only take place through thermal fluctuations. As thecluster size increases,N1µ dominates so that the slopeof the1G(N) versusN curve becomes negative and thecluster will grow spontaneously. The critical cluster size,Ncrit , is defined such that d1G(N)/dN = 0. Assumingthat 8(N) is determined by the surface energy terms

and considering an arbitrary three-dimensional geometricalform, equation (34) can differentiated with respect toN .The overpotential dependence of the critical cluster size,Ncrit , for a single-electron-transfer deposition process isgiven by the following relation [63]:

Ncrit = 8BV 2mσ

3

27(e|η|)3 (35)

whereB is a geometrical factor (36π for a sphere, 63 fora cube),Vm is the atomic volume andσ is the averagespecific surface energy. The associated critical energy forgrowth of a nucleus,1Gcrit , is given by

1Gcrit = 4BV 2mσ

3

27(e|η|)2 . (36)

The rate of nucleation can be represented by the classicVolmer–Weber equation [69]:

Jnucl = A3D exp

(−1Gcrit

kT

)(37)

where the pre-exponential factorA3D is only weaklydependent on the overpotential [63]. Combiningequations (36) and (37) for three-dimensional nucleationleads to the following relation between the rate ofnucleation and the overpotential:

Jnucl = A3D exp

(− 4BV 2

mσ3

27kT (e|η|)2). (38)

This equation is used in the analysis of experimentalresults by plotting ln(Jnucl) versus |η|−2. In general,the overpotential dependence ofA3D is neglected in thisanalysis. The overpotential dependence of the rate ofnucleation is characterized by a threshold overpotential;as long as the absolute overpotential is smaller than thiscritical overpotential, the rate of nucleation is negligible.The critical overpotential for nucleation is often observedexperimentally. Another useful relation is obtained bycombining equations (35)–(37) and neglecting the potentialdependence ofA3D:

d ln(Jnucl)

d|η| = − 1

kT

d1Gcrit

d|η| =e

kTNcrit . (39)

Hence, by plotting ln(Jnucl) versus the overpotential thenumber of atoms required to form the critical nucleus sizecan be obtained from the slope; note that this is onlypossible for small overpotential regimes becauseNcrit isa function of the overpotential.

The classical approach described above is based onthe bulk properties of the deposited metal. Various othertheoretical treatments based on the kinetic approach firstsuggested by Becker and Doering [70] have been proposed.In these models, the formation of a cluster is treatedas a sequence of attachment and detachment steps. Inequilibrium, the rates of attachment and detachment areequal, whereas supersaturation leads to an increase in therate of attachment and growth of the cluster. The small-cluster model developed by Walton [71, 72] and Stoyanov

1936


[73, 74] uses an atomistic probability approach in which nomacroscopic quantities such as surface energy and volumeare involved. The result of this theoretical analysis is anexpression very similar to that obtained from the classicalmodel [63]:

Jnucl = A∗3D exp

(βe|η|kT

)exp

(Ncrit e|η|kT

). (40)

The pre-exponential factorA∗3D is independent of thepotential as long asNcrit is potential independent; the factorβ depends on the mechanism of attachment. The totalpotential dependence of the rate of nucleation in anoverpotential range in whichNcrit is constant accordingto the atomistic model is thus given by

d ln(Jnucl)

d|η| = e

kT(β +Ncrit ). (41)

The value ofβ in equations (40) and (41) results fromthe probability of attachment of an atom to a cluster,thus forming a cluster of critical size. If the attachmentoccurs through direct transfer of metal ions to the clusterand simultaneous charge transfer, thenβ = αc, whereαcis the cathodic charge transfer coefficient. This can bedirectly obtained from the cathodic term in the Butler–Volmer equation (for a single-electron-transfer process)as d ln(i)/d|η| = eαc/(kT ) (see equation (20)). If themechanism is attachment of adatoms to the cluster, thenβ = 1. This stems from the overpotential dependence ofthe concentration of adatoms [63, 75]. Hence, by plottingln(Jnucl) versus|η| the value ofNcrit can be determined;the result for a large critical cluster size is the same as thatobtained from the classical theory sinceβ is negligible inthis case.

Various experimental studies have been performed todetermine the size of the critical nucleus for deposition ofmetals onto metals [75–78], and a few for depositionof metals onto semiconductors [67, 68]. For depositionof metals onto metals, the critical nucleus size determinedfrom equation (39) or (41) is usually found to be fewer thanten atoms. This result suggests that the critical energy forformation of a stable cluster is very small. For depositionof metals onto semiconductors, the critical nucleus size hasbeen reported to be close to unity, implying a barrierlessnucleation process. However, the determination of|η| inequations (39) and (41) for semiconductor surfaces is notstraightforward. The implications of this will be discussedin section 6.3.4.

4.1.2. The time dependence of nucleation and growthprocesses. The nucleation mechanism and its rate areusually determined using current transient techniques.Upon applying a potential step from an initial potentialat which the rate of nucleation is negligible to a fixedoverpotential, the formation of stable nuclei and theirgrowth can be observed directly by monitoring the current[79]. From the shape of the resulting current transient,information on the rate of nucleation, the density of nucleiand the mechanism of growth can be obtained.

In general, nucleation of a metal on a foreign substrateis assumed to take place at active sites on the surface, suchas steps, kinks or other surface defects. The density ofactive sites,N0, represents the total number of possiblesites for nucleation and may be potential dependent. Thedensity of nuclei as a function of time,N(t), is usuallydescribed in terms of a linear growth law with a nucleationrate constant,A:

N(t) = N∞[1− exp(−At)] (42)

whereN∞ is the final density of nuclei (the total densityof utilized active sites) under the experimental conditions.From these definitions,N∞ ≤ N0 at any given potential.From equation (42) two limiting cases can be identified. IfA is large andAt � 1 at short times thenN(t) = N∞immediately after the pulse. Conversely, ifA is small andAt � 1 at short times thenN(t) = AN∞t and the densityof nuclei increases linearly with time. These two casescorrespond to instantaneous nucleation and progressivenucleation, respectively.

The growth of nuclei can be kinetically limited,diffusion limited or under mixed control. In many cases, theoverpotential used for nucleation and growth is relativelylarge, so that diffusion of metal ions to the surface is oftenrate limiting [80–87]. At very short times, growth may bekinetically limited due to the large diffusion flux to smallnuclei [88]. As the growth becomes diffusion limited, thediffusion zones around individual nuclei will start to overlapand linear diffusion to the surface may occur before nucleiimpinge on each other. If the nucleation is progressive, thedevelopment of diffusion fields may block active sites fornucleation, leading to the situation in whichN∞ < N0.In addition to the potential dependence ofN0, N∞ islikely to be a function of the potential. In general, it isobserved that the density of nuclei increases with increasingoverpotential, which is a consequence of these two effects.

For progressive nucleation, the rate of nucleationcan be determined from current transients; assuming thatN(t) = AN∞t , the rate of nucleation is simply given by

Jnucl = dN(t)

dt= AN∞. (43)

Therefore, by determiningAN∞ as a function of the appliedpotential, the number of atoms required to form a criticalnucleus can be obtained using equation (39) or (41).

Experimentally, the mechanism of nucleation andgrowth can be obtained from analysis of current transientsat different deposition potentials. Models for currenttransients are based on various simplifications, althoughrecent modifications have included a correction forshielding effects due to growing nuclei and mathematicaltreatments for other shapes and distributions of nuclei[86, 87]. In general, these refinements do not significantlyaffect the form of the transients for instantaneousnucleation, but can have an effect on the progressivenucleation model. In the following, we restrict thediscussion to the simple cases.

The time-dependent deposition current density (nor-malized for the geometrical surface area),i(t), for

1937

G Oskam et al

instantaneous nucleation followed by three-dimensionaldiffusion-limited growth is [80, 81]

i(t) = zFD1/2c

π1/2t1/2

{1−exp

[−N∞πDt

(8πcM

ρ

)1/2]}(44)

whereD is the diffusion coefficient,c is the concentrationof metal ions in the bulk solution,M is the molar weightof the deposit andρ is the density of the film. Notethat, for instantaneous nucleation,N∞ should be equalto N0 since all active sites are utilized immediately afterthe potential pulse. For progressive nucleation, the time-dependent deposition current density is given by [80, 81]

i(t) = zFD1/2c

π1/2t1/2

{1−exp

[− 2

3AN∞πDt2

(8πcM

ρ

)1/2]}.

(45)In both cases, the current initially increases with time dueto 3D diffusion to an increasing surface area and thendecreases as the diffusion zones around the growing nucleistart to overlap, resulting in a 1D diffusion-limited current.In some cases, the time required for the formation of nucleiof critical size may be significant, leading to an inductiontime.

The deposition mechanisms can be compared directlyby rewriting the growth laws in terms of the maximumcurrent,imax , and the time at which the maximum currentis observed,tmax . For instantaneous nucleation

i2

i2max= 1.9542

(tmax

t

)[1− exp

(−1.2564

t

tmax

)]2

(46)

and for progressive nucleation

i2

i2max= 1.2254

(tmax

t

)[1− exp

(−2.3367

t2

t2max

)]2

.

(47)The time,t , in equations (44)–(47) is the time with respectto the onset of the deposition current; that is,t is correctedfor the induction time,t0.

Verification of the mechanisms of nucleation andgrowth can be obtained through determination of thediffusion coefficient and comparison with known values.The diffusion coefficient of the metal ion,D, canbe obtained from the producti2maxtmax [80, 81]. Forinstantaneous nucleation

D = i2maxtmax

0.1629(zFc)2. (48)

For progressive nucleation:

D = i2maxtmax

0.2598(zFc)2. (49)

The diffusion coefficient can also be obtained fromanalysis of the decay of the current at long times after thecurrent peak. At sufficiently long times, linear diffusionto the electrode surface dominates the deposition rate andthe current is described by the Cottrell equation; fromequations (44) and (45) it can be seen that

D1/2 = π1/2

zFc

di

dt−1/2. (50)

Note that 1D diffusion-limited growth becomes dominantafter the current peak at the point at which the exponentialterm in equations (44) and (45) become negligible. Forinstantaneous nucleation the growth current is 95% of thelinear diffusion-limited current at a timet = 2.39tmax . Forprogressive nucleation the growth current is 95% of thelinear diffusion-limited current att = 1.13tmax .

For instantaneous nucleation, the nucleus density,N∞,can be determined directly from the current transients usingthe following equation:

N∞ = 0.065

(8πcM

ρ

)−1/2(zFc

imaxtmax

)2

. (51)

For progressive nucleation, the nucleation rate,AN∞, canbe obtained from the maximum in the current transients:

AN∞ = 0.2898

(8πcM

ρ

)−1/2(zFc)2

i2maxt3max

. (52)

The nucleus density,N∞, can be measured by depositionof sufficient material to image the nuclei by a suitabletechnique, such as SEM or TEM. For progressivenucleation and 3D diffusion-limited growth,N∞ reachesa limiting value at t/tmax � 1 at which the diffusionfields for the nuclei overlap and transport of metal ionsto the surface is one dimensional. Hence, in order todetermineN∞, the deposition time needs to be sufficientlylong in order to ensure that the final nucleus density isattained. These analyses can be used to quantitativelycompare experimental results with the theoretical models,which we will show in section 6.3.

4.2. Deposition of continuous films onto semiconductorsurfaces

The deposition of continuous metal films onto semicon-ductor surfaces can be difficult to achieve for various rea-sons. The interaction energy between metal adatoms andthe semiconductor surface is generally small and depositionusually follows the 3D island growth mechanism. In orderto obtain a continuous film, the density of nuclei must besufficiently high for the islands to coalesce during furtherdeposition of the metal. If the growth process is diffusionlimited, however, any instabilities will lead to very roughmorphologies. Hence, deposition of a continuous metalfilm requires a high density of nuclei followed by growthunder kinetic or mixed control (namely at relatively low de-position currents). For the deposition of continuous metalfilms onto n-type semiconductors, the electron transfer stepmay be faster at metal nuclei and further film growth maybe expected to occur preferentially at pre-existing metalislands. Indeed, this is the basis for photoelectrochemi-cal cells in which a p-type electrode is covered with metalislands in order to increase the rate of evolution of hydro-gen by utilizing photogenerated conduction band electrons[30–36]. For the deposition of continuous films onto p-typesemiconductors through charge transfer via the conductionband, a high intensity of light may be required in order toobtain a sufficiently high surface electron density both forthe nucleation and for growth processes. Note that deposi-tion of thicker films onto p-type surfaces under illumination

1938


Figure 9. Schematic illustrations showing (a) the initialstages of Volmer–Weber growth of a metal on silicon and(b) deposition of a metal onto silicon at later stages when acontinuous metal film is deposited.

is complicated by the reflection from the deposited metallayer.

As a result, strategies for the deposition of continuousmetal films onto semiconductor surfaces generally includean initial potential step involving the formation of a highdensity of nuclei, followed by growth at a low overpotentialunder conditions such that the flux of metal ions to thesurface is not rate limiting.

4.3. Deposition of metals onto semiconductor/metalfilms

During deposition of metals onto semiconductor surfacesunder conditions where the density of nuclei is sufficientlyhigh, coalescence of the nuclei will result in the formationof a continuous film. In this case the semiconductoris isolated from the solution and subsequent growth isdependent on the properties of the metal film and thepotential distribution across the interfaces. Figure 9(a)illustrates the steps involved in electrochemical depositionof a metal onto the semiconductor surface. Initially ametal ion reaches the surface and accepts electrons fromthe semiconductor; the adsorbed metal atom can thendiffuse along the surface before becoming incorporatedinto a nucleus. Figure 9(b) shows the steps involvedfor deposition of a metal onto an existing continuousmetal film. In this case the energetics and kinetics ofthe deposition reaction are very different because thesemiconductor surface is no longer involved in the process.

As soon as metal nuclei are deposited onto asemiconductor surface, the energetics and kinetics of thesystem may change dramatically, depending on the natureof the semiconductor/metal contact. For example, if thesurface coverage of the metal on the semiconductor is low,then the band bending may be spatially inhomogeneous

[32, 58]. If the semiconductor/metal film results in theformation of a Schottky barrier, electrochemical strippingof the metal layer from an n-type semiconductor is notpossible since electrons in the metal film have to be excitedto the conduction band for dissolution to occur. An exampleof this behaviour is gold on n-type silicon (see figure 19later). In some cases stripping can be achieved underopen-circuit conditions when a suitable oxidizing agent ispresent [67]. In contrast, if an ohmic contact is formedon an n-type semiconductor, the dissolution of the metalby excitation of metal electrons to the conduction bandcan be very fast and a stripping or oxidation peak canbe observed in the current–potential curve; an exampleof this mechanism is indium on n-type InP [89]. In thecase of deposition of metals onto p-type semiconductors,a stripping peak is usually observed due to the transfer ofvalence band holes, which are the majority carriers, to themetal at sufficiently positive potentials [90]. In additionto these energetic considerations, the kinetics of reactionsat metal islands and semiconductor surfaces usually differconsiderably, leading to a spatial dependence of rates ofreaction at surfaces covered with metal nuclei.

From the above discussion it can be seen thatthe energetics and kinetics of redox reactions atmetallized semiconductor surfaces are very complicated[32, 58, 61, 62] and the presence of two interfaces hasimportant consequences. Under equilibrium conditions,the Fermi energies of the semiconductor, metal and redoxcouple in the solution are the same. The metal layer is notdirectly connected to the external circuit and, consequently,upon applying a potential the Fermi level of the metal layercan remain aligned with that of the semiconductor or withthat of the redox couple, or it can be at some intermediateposition. Kinetically, this can be viewed as being dependenton the rates of transfer of charge between either the solutionand the metal or the semiconductor and the metal.

5. Silicon

5.1. Silicon surfaces

In any deposition process, the surface chemistry andmorphology play important roles in determining thestructure and properties of the deposit. For depositionof metals onto silicon, the experimental conditions shouldbe such that oxide formation is avoided and the surfaceshould be stable during the deposition process. This can beachieved by careful selection of the deposition solution andby avoiding strongly oxidizing conditions. The most widelyused technique is hydrogen passivation of the surface.

The hydrogen-passivated silicon surface is obtained byimmersion in HF/NH4F solution, in which the final stepafter dissolution of oxide is the formation of Si–Hx surfacebonds [91–93]. The morphology and surface chemistryare dependent on the crystal orientation and the pH of theetching solution. The hydrogen-terminated (111) surface ischaracterized by a very low density of surface states and alow surface recombination velocity [22, 91, 94] and is stableunder open-circuit conditions up to about pH 8 and abovepH 14. In fluoride solutions at pH< 6 [15, 92–97] the

1939

G Oskam et al

Figure 10. The effective equilibrium potentials for variousmetal/metal ion complexes with respect to the position ofthe band edges for silicon surfaces in aqueous solution atpH 2 and pH 14.

hydrogen-passivated surface is stable over a wide potentialrange; however, at positive potentials local etching leads tothe formation of a porous layer [98]. In NaOH solutions(pH 14) the surface is stable at potentials negative relativeto the open-circuit potential [99–101].

5.2. The energy of the electron acceptor

From the preceding discussion it is clear that the mechanismof the deposition process is dependent on the energeticposition of the acceptor states in solution with respect tothe band edges of the semiconductor. Consequently, themechanism of deposition can be controlled by modifyingthe position of the band edges or the equilibrium potentialof the metal/metal ion couple. In one approach, theenergetics of the metal/metal ion couple can be tunedby complexation of the metal ion. Figure 10 showsthe equilibrium potential for various metal ion complexes[102, 103] with respect to the position of the band edges insilicon. In the second approach, the position of the bandedges of the semiconductor can be shifted by adsorptionor derivatization of the surface; for example, for somesemiconductors, the flat band potential exhibits a Nernstiandependence on pH due to adsorption of H+/OH− [12].Figure 10 shows the energetic position of the band edgesof silicon for a 1 M fluoride solution at pH 2 and for 1 MNaOH, of which the pH is 14.

6. Deposition of copper, gold and platinum ontosilicon

6.1. Copper on silicon

Figure 11 shows a current–potential curve for (100) orientedn-type silicon, ND = 1015 cm−3 (ρ = 4 � cm) in7.5 mM CuCO3·Cu(OH)2 with 20 g l−1 H3BO3 and 9.6 gl−1 HBF4 at pH 1.4. The open-circuit potential in thissolution was−0.17 V (relative to Ag/AgCl). Duringthe first cycle, the onset of reduction of Cu2+ occurs at−0.3 V (Ag/AgCl) with a characteristic diffusion-limitedgrowth peak at−0.38 V (Ag/AgCl). After the deposition

Figure 11. The current–potential curve for n-Si(100),ND = 1015 cm−3, in 7.5 mM CuCO3·Cu(OH)2 with 20 g l−1

H3BO3 and 9.6 g l−1 HBF4 (full line) at a scanning rate of10 mV s−1. Also shown is a current–potential curve forsilicon in the same solution without Cu2+ (broken line).

peak, the current again increases at a potential of about−0.8 V (Ag/AgCl) due to the reduction of water at thecopper clusters on the surface. Also shown in figure 11is a current–potential curve for n-type silicon in the samesolution in the absence of copper ions, illustrating that theonset of the hydrogen evolution current occurs at about−0.9 V (Ag/AgCl). In the potential range where evolutionof hydrogen occurs, however, the current density at asurface with copper clusters is higher than that at the siliconsurface.

The reverse scan in figure 11 shows a deposition currentof about 0.8 mA cm−2 due to the continued growth atcopper clusters. The potential difference between thecurrent onset in the forwards and reverse scans is related tothe nucleation barrier. The reverse scan does not exhibit ananodic peak, which confirms that electrochemical strippingof the copper layer is not significant in the dark due to thelarge barrier height (about 0.6 eV) of the n-Si/Cu junction.In order to dissolve the copper, electrons must be excitedfrom the Fermi level of copper into the conduction band;however, this is a slow process. In some cases, dependingon the metal and the composition of the solution, the metallayer can be dissolved chemically. For example, Scherband Kolb [67] used reflectance spectroscopy in conjunctionwith electrochemical measurements to show that copperelectrodeposited onto n-GaAs is chemically dissolved in achloride solution whereas this does not occur in a sulphatesolution. For the solution used here, this effect is notexpected and we did not observe chemical dissolution ofdeposited copper layers.

Figure 12 shows a series of deposition transients fordeposition of copper from 7.5 mM CuCO3·Cu(OH)2. Allthe transients exhibit an initial increase in current due tonucleation followed by a decrease in current associatedwith diffusion-limited growth. In this solution, nucleation

1940


Figure 12. Current–time curves for the deposition of copperonto n-Si(100), ND = 1015 cm−3. In all cases the potentialwas stepped from −0.2 V (Ag/AgCl), to (a) −0.8 V,(b) −0.7 V, (c) −0.6 V, (d) −0.55 V and (e) −0.35 V.

is relatively slow and the peak current is well separatedfrom double layer charging.

Figure 13 shows the deposition transient at−0.8 V(Ag/AgCl) plotted in dimensionless form, indicating thegood agreement with the model for instantaneous nucleationfollowed by 3D diffusion-limited growth. Figure 14 is adimensionless plot of the transient at−0.35 V (Ag/AgCl)which is consistent with the progressive nucleation model.Analysis of transients at potentials between these two valuesrevealed behaviour intermediate between the two limitingcases. This result suggests that, at deposition potentialsnegative relative to−0.8 V (Ag/AgCl), the nucleation sitesbecome saturated after short times in comparison with thetransient peak.

The density of nuclei can be calculated fromtmax andimax for the case of instantaneous nucleation observed at−0.8 V (Ag/AgCl) using equation (51). The density ofnuclei was determined to be 3× 107 cm−2 which is of thesame order of magnitude as values reported in the literaturefor nucleation of copper on GaAs [67, 68].

The value of the diffusion coefficient,D, wasobtained from the producti2maxtmax . For the depositiontransient at−0.8 V, we obtainD = 1.7× 10−5 cm2 s−1

and for the deposition transient at−0.35 V, we obtainD = 0.61× 10−5 cm2 s−1. Figure 15 shows the decayportion of the deposition transient at−0.8 V (Ag/AgCl)plotted asi versust−1/2. Figure 15 shows a linear regionfrom about 50 ms up to the end of the data record afterabout 2 s. In this case, the current peak occurred at19 ms, indicating that 1D diffusion-limited growth becomesdominant at about 2.5tmax , which is in agreement with theprediction for instantaneous nucleation. From the linearregion shown in figure 15, the diffusion coefficient wascalculated to be 2.1×10−5 cm2 s−1, in good agreement withthe value of 1.7× 10−5 cm2 s−1 obtained from the currentpeak. For the deposition transient at−0.35 V (Ag/AgCl)

Figure 13. The current transient for deposition of copperonto n-Si(100), ND = 1015 cm−3, at −0.8 V (Ag/AgCl)plotted in dimensionless form. Also shown are thetheoretical curves for progressive nucleation (broken line)and instantaneous nucleation (full line).

Figure 14. The current transient for deposition of copperonto n-Si(100), ND = 1015 cm−3, at −0.35 V (Ag/AgCl)plotted in dimensionless form. Also shown are thetheoretical curves for progressive nucleation (broken line)and instantaneous nucleation (full line).

the current peak occurred at 2.16 s and a linear region inthe i(t) versus t−0.5 plot was observed after about 4 s,corresponding to about 1.8tmax , in good agreement withthe progressive nucleation model. From the slope of thelinear region the diffusion coefficient was calculated to be0.52× 10−5 cm2 s−1, in good agreement with the value of0.61× 10−5 obtained from the current peak. The valuesfor the diffusion coefficient determined from analysis ofthe current–time transients are in good agreement with theliterature value of about 0.6× 10−5 cm2 s−1 [105], which

1941

G Oskam et al

Figure 15. A Cottrell plot for growth limited by lineardiffusion to a planar surface for the deposition of copperonto n-Si(100), ND = 1015 cm−3, at −0.8 V (Ag/AgCl) (seefigure 12, curve (a)).

confirms that the models of nucleation and growth used areapplicable in this case. The slightly higher value found at−0.8 V is probably due to co-reduction of H+ to H2.

6.2. Platinum on silicon

We have deposited platinum films onto n-type silicon (100),ND = 5× 1014 cm−3 (ρ = 8.2 � cm). The films weredeposited from 10 mM(NH4)2PtCl6 in 0.41 M HNa2PO4

adjusted to pH 4 by adding 49 wt% HF. The presence ofHF eliminates formation of an interfacial oxide as longas the applied potential is sufficiently negative. Analysisof the mechanism of deposition was not possible sincethe platinum deposition current could not be deconvolutedfrom the hydrogen evolution current (evolution of hydrogenis strongly catalysed on platinum). Slow deposition ofplatinum onto p-type silicon in the dark was observed,indicating that platinum can be deposited by hole injection.Platinum can also be deposited onto n-type silicon bythe displacement method (case (i) in figure 8), which isconsistent with the energetic situation shown in figure 10.However, at sufficiently negative potentials, the densityof electrons at the surface is high and injected holesmay recombine quickly with conduction band electrons,resulting in a deposition current. In addition, the transferof conduction band electrons to the solution may be able tocompete with the hole injection process. In both cases, theoxidation of the silicon surface is minimized. This situationis similar to the case for deposition of platinum onto n-typeGaAs [66].

The deposition of thin, continuous platinum films onton-type silicon was achieved using two potential steps: first,a 20 ms potential pulse from the open-circuit potential(about−0.9 V (relative to Pt)) to−2.7 V (Pt); and second,a pulse of about 10 min to−1.0 V (Pt). The first step

Figure 16. The SEM cross section of a 10 µm thickelectrochemically deposited platinum film on n-Si(100),ND = 1015 cm−3.

Figure 17. A Mott–Schottky plot for a solid staten-Si(100)/Pt junction fabricated by electrochemicaldeposition. The frequency of the ac signal was 10 kHz andthe amplitude was 10 mV (RMS).

ensures a large density of nuclei and during the secondstep the layer was grown at a kinetically limited rate. Theplatinum films deposited by this method were uniform andsmooth and had good adhesion to the silicon substrate, asshown in figure 16. In contrast, platinum films depositedonto n-type silicon by electroless deposition exhibit pooradhesion [54].

Figure 17 is a Mott–Schottky plot for an electrochemi-cally fabricated n-Si/Pt contact. The donor density obtainedfrom the slope of the Mott–Schottky plot in figure 17 cor-responds to the value obtained from resistivity measure-ments. From the intercept with the potential axis and thedonor density a barrier height of 0.81 eV is obtained. Thisvalue is in good agreement with values reported for junc-tions fabricated by vapour deposition of the platinum layer[57, 104].

Figure 18 shows the forwards current–voltage curveon a semi-logarithmic plot. The plot is linear with aninverse slope of 27.3 mV, corresponding to an idealityfactor of 1.06. The low ideality factor indicates a high-quality junction with a low density of interface states. Theintercept at zero bias gives the saturation current densityand, by using equation (32), a barrier height of 0.82 eV isfound, which is in good agreement with the value obtainedfrom the impedance measurements. The current in the

1942


Figure 18. The current–voltage curve for a solid staten-Si(100)/Pt junction fabricated by electrochemicaldeposition in the forwards bias regime. The full linerepresents an inverse slope of 27.3 mV, corresponding toan ideality factor of 1.06.

reverse bias regime (not shown) was within a factor of twoof the saturation current density up to a voltage of 1.5 V,indicating that the interface is of high electronic quality.

6.3. Gold on silicon

The standard equilibrium potential for the AuCl−4 /Au redox

couple is 0.76 V (Ag/AgCl) so that the AuCl−4 acceptorlevels are expected to have good overlap with the valenceband of silicon. In order to achieve conduction bandcontrol of the deposition process, the equilibrium potentialcan be shifted to more negative values by complexing thegold (I) ion. The effective equilibrium potential for theAu(CN)−2 /Au couple is about−0.85 V (Ag/AgCl) andthe conduction band edge for silicon at pH 14 is about−1.2 V (Ag/AgCl), as shown in figure 10. As a result, theAu(CN)−2 states are expected to have a large overlap withthe conduction band.

6.3.1. Current–potential curves. Figure 19 shows acurrent–potential curve for (100) oriented n-type silicon,ND = 1015 cm−3 (ρ = 4 � cm) in 10 mM KAu(CN)2with 1 M KCN (pH 14) in the dark. Also shown is acurrent–potential curve for 1 M KCN (pH 14). In theKCN solution, the onset of evolution of hydrogen occursat about−1.7 V (Ag/AgCl). In KAu(CN)2 solution, thecurrent increases sharply at about−1.3 V (Ag/AgCl) dueto the nucleation and growth of gold on silicon. Thecurrent goes through a maximum as diffusion of Au(CN)−2to the surface becomes rate limiting and, at more negativepotentials, evolution of hydrogen is observed. The reversescan implies a nucleation barrier of about 0.3 V. The reversescan does not exhibit a stripping peak due to the largebarrier height for the silicon/gold contact. After multiple

Figure 19. Current–potential curves of n-Si(100),ND = 1015 cm−3, in 1 M KCN (pH 14) (broken line) and in10 mM KAu(CN)2 with 1 M KCN (pH 14) (full line); thesweeps were started at −0.5 V (Ag/AgCl) and the arrowsshow the direction of scanning. The scanning rate was10 mV s−1.

cycling, a gold film was present on the surface, however,the adhesion was generally poor. Deposition of goldonto p-type silicon from this solution occurs only underillumination, confirming that deposition proceeds throughtransfer of electrons from the conduction band.

6.3.2. Current transients. Figure 20 shows a seriesof current transients at potentials in the range−1.45 V(Ag/AgCl) to −1.70 V (Ag/AgCl). The experimentswere performed by stepping the potential from−1.1 V(Ag/AgCl), at which no deposition occurs on the siliconsurface. The transients show a charging peak att < 1 mswhich is followed by the nucleation and growth current.For clarity, the charging peaks are not shown in figure 20.For deposition potentials in the range−1.30 V (Ag/AgCl)to −1.60 V (Ag/AgCl) the current att > 1 s is limited bydiffusion of Au(CN)−2 to the surface, whereas at potentialsmore negative than−1.60 V (Ag/AgCl) an additionalcurrent due to evolution of hydrogen is observed. Afternucleation, the current corresponding to deposition of goldis kinetically limited in the potential range from−1.10 V(Ag/AgCl) to −1.25 V (Ag/AgCl).

Figure 21 shows the transients at growth potentials of−1.45 V (Ag/AgCl) and−1.7 V (Ag/AgCl) in reducedform, illustrating that the nucleation of gold on silicon inthe KAu(CN)2 with KCN solution follows a progressivenucleation mechanism. Figure 22 shows a SEM image ofgold nuclei on silicon prepared in a 50 mM KAu(CN)2with 1 M KCN solution at pH 14. The deposition potentialwas−1.26 V (Ag/AgCl) which is about 40 mV positivewith respect to the potential of the current maximumin the current–potential curve (at 10 mV s−1) for thissolution. The current transient was in agreement with theprogressive nucleation model and the total deposited chargewas 20 mC cm−2. The SEM image clearly shows a large

1943

G Oskam et al

Figure 20. Current–time curves for the deposition of Auonto n-Si(100), ND = 1015 cm−3, in 10 mM KAu(CN)2 with1 M KCN (pH 14). The potential was stepped from −1.05 V(Ag/AgCl) to (a) −1.7 V, (b) −1.65 V, (c) −1.6 V,(d) −1.55 V, (e) −1.5 V and (f) −1.45 V. The increases incurrent at longer times for curves (a) and (b) are due toevolution of hydrogen.

distribution in nucleus sizes which is in agreement with theprogressive nucleation mechanism. The density of nucleiunder these experimental conditions is about 5×108 cm−2,but it should be noted that smaller clusters not observed atthis magnification may be present. The density of nuclei isrelatively low at this potential; at potentials negative withrespect to the peak potential in the current–potential curve,the density of nuclei increases to about 4× 109 cm−2 andsaturates at this value [106].

Figure 23 shows the dependences oftmax and imaxon the deposition potential on a semi-logarithmic plot,illustrating that both parameters are exponential functionsof the potential. The inverse slopes of the log(tmax) andlog(imax) versus potential plots are 140 mV per decade and−280 mV per decade, respectively. The producti2maxtmaxfor these two slopes is therefore potential independent,implying that the diffusion coefficient for Au(CN)−2 is alsoconstant (equation (49)). From this value the diffusioncoefficient for Au(CN)−2 was determined to be about 2×10−5 cm2 s−1. The expressions fortmax and imax forprogressive nucleation are obtained from equation (45):

tmax =(

3.505

AN∞πc1/2D( 8πMρ)1/2

)1/2

(53)

imax = 0.4959zFD3/4c9/8

(8πM

ρ

)1/8

(AN∞)1/4. (54)

It can be seen from equations (53) and (54) that the potentialdependences both oftmax and of imax in this analysis aredetermined by the potential dependence ofAN∞ since allother terms are independent of the potential. In accordancewith the experimental results, the magnitude of the slope of

Figure 21. Current transients for deposition of Au onto n-Si(100): (♦), at −1.7 V (Ag/AgCl); and (◦), at −1.45 V(Ag/AgCl) plotted in dimensionless form (see figure 20,curves (a) and (f)). Also shown are the theoretical curvesfor progressive nucleation (broken line) and instantaneousnucleation (full line).

Figure 22. A SEM image of electrochemically depositedgold nuclei on silicon. The nuclei were deposited from50 mM KAu(CN)2 with 1 M KCN (pH 14) at a pulsepotential of −1.26 V (Ag/AgCl) for 15 s (Q = 20 mC cm−2).The reduced parameter plot (see figure 21) was inagreement with progressive nucleation. The image showsthat there is a large distribution in nucleus sizes, whichconfirms that progressive nucleation occurs.

the log(imax) versus potential plot should be twice as largeas and of opposite sign to that of the log(tmax) versus pulsepotential plot.

Figure 24 shows ln(AN∞) versus the pulse potential,where AN∞ was obtained using equation (52). Astraight line is obtained in agreement with equations (39)and (41), which relate the rate of nucleation to theoverpotential according to the classical and atomistictheories, respectively. The relationship is linear over apotential range of 0.25 V, which is somewhat surprising

1944


Figure 23. The time and the current at the maximum of the current transients shown in figure 20, measured in 10 mMKAu(CN)2 with 1 M KCN (pH 14). Both tmax (�) and imax (◦) are exponentially dependent on potential.

sinceNcrit should decrease with increasing overpotentialaccording to equation (35). However, differentiation ofequations (53) and (54) shows that the slopes of the plotsshown in figure 23 are consistent with the slope of theln(AN∞) versus potential plot. The slope in figure 24is −32.7 V−1, which corresponds toNcrit = 0.84 inthe classical nucleation model andNcrit = 0 in theatomistic model (β is expected to be between 0.5 and1). This result suggests that there is no barrier fornucleation in this potential range. However, nucleationof metals on semiconductor substrates may present variouscomplications which we will discuss in section 6.3.4. Itshould be noted that there is no additional information inthe potential dependence oftmax andimax if the analysis interms of the conventional theories for deposition of metalsonto metals is used.

6.3.3. Properties of Si/Au dry contacts and Si/Auelectrodes in solution. In order to obtain continuous,adherent films, special care must be taken to ensure a largedensity of nuclei and a kinetically limited deposition ratefor the growth of the film. Using this method, high-qualitygold films were obtained. Measurements of silicon/gold drycontacts showed that the barrier height is about 0.82 eV,which is consistent with the observation that there is nostripping peak in the current–potential curve in this solution(see figure 19). Current–potential measurements on thedry junctions gave the same barrier height with an idealityfactor of about 1.15–1.2, indicating high electronic quality.

In order to study the potential distribution at gold-covered silicon electrodes, the potential of the gold film canbe measured independently versus the reference electrodeas a function of the potential applied to the silicon.Figure 25(a) shows a current–potential curve for n-typesilicon with a 150 nm thick gold layer in a solution of0.1 M K4Fe(CN)6 plus 0.5 M KCl. The current–potential

Figure 24. A plot of ln(AN∞) versus potential calculatedusing equation (52) for the results obtained in 10 mMKAu(CN)2 plus 1 M KCN (pH 14).

curve of the silicon/gold film is similar to that of siliconsurfaces: at positive potentials the anodic current is smallsince the density of holes (minority carriers) is low, whereasat negative potentials a cathodic current is seen due tothe reduction of water through electron transfer from theconduction band via the gold layer. The small plateaucurrent is probably due to reduction of oxygen and/orgenerated Fe(CN)3−6 .

Figure 25(b) is a Mott–Schottky plot for the gold-covered silicon electrode. The plot is linear between0 and 1 V (Ag/AgCl) and the flat band potential is−0.52 V (Ag/AgCl), close to the value for a silicon

1945

G Oskam et al

Figure 25. (a) The current–potential curve for n-Si(100),ND = 1015 cm−3, with a 150 nm thick layer of gold in 0.1 MK4Fe(CN)6 with 0.5 M KCl; (b) the correspondingMott–Schottky plot (f = 10 kHz, amplitude 5 mV (RMS));and (c) the potential of the gold layer versus the referenceelectrode. All experiments were performed simultaneously,with silicon as the working electrode.

surface at the same pH [18]. Figure 25(c) shows thepotential of the gold film recorded independently duringa potential sweep from positive to negative potentials.At positive potentials, for which Mott–Schottky behaviourwas observed, the potential of the gold layer is constantat about 0.07 V (Ag/AgCl) corresponding to the Nernstpotential of the Fe(CN)3−6 /Fe(CN)4− redox couple. At

negative potentials between 0 and−0.4 V (Ag/AgCl), thecapacitance–potential plot has a plateau, indicating that theband bending does not change with the applied potential.From figure 25(c) it can be seen that in this potential range,the gold potential changes linearly with the applied potentialwith a slope of unity.

These results can be explained as follows. At positivepotentials, electrons from the gold layer must be excitedinto the conduction band of the silicon for the Fermilevels of the gold and silicon to remain aligned. Theexchange current of the redox couple is much faster thaninjection of electrons into the semiconductor and, hence,the redox couple determines the Fermi level of gold. TheMott–Schottky plots show that all the applied potential isdropped over the silicon space charge layer and hence thepotential drop over the Helmholtz layer between the goldand the solution is independent of the applied potential.

At negative potentials, figure 25(c) shows thatthe potential of the gold layer is determined by thesemiconductor. In this case, the band bending is sufficientlysmall that a large forwards current can maintain the Fermilevel of gold equal to that of the silicon. As a result,all the applied potential is dropped across the Helmholtzlayer at the gold/electrolyte interface and the band bendingis independent of the applied potential. Hence, in theabsence of interfacial layers and surface states, the potentialdistribution at metal-covered semiconductor electrodescan be analysed using the description for the potentialdistribution at semiconductor/solution interfaces discussedin section 2.1. The presence of the metal layer provides anopportunity to determine the dependence of the potentialdrop over the Helmholtz layer at the metal/solutioninterface, which is not possible for bare semiconductorsurfaces.

The maximum current that can flow through thesilicon/gold junction can be estimated from the thermionicemission current. The band bending in the potential rangefor which the applied potential is dropped across thegold/electrolyte interface (U < 0 V (Ag/AgCl)) can beestimated from the flat band potential and figure 25(c) tobe about 0.52 eV. The forwards current on a dry n-typesilicon/gold Schottky junction at this band bending is about2.6 mA cm−2 (see equations (31) and (32); usingU =−0.3 V andn = 1.15) which is sufficiently large to reducewater.

6.3.4. Discussion of nucleation and growth of gold onsilicon. In order to apply the analysis of the transientparameters as a function of the deposition potential in termsof Ncrit for semiconductor electrodes, the overpotentialneeds to be defined as a function of the applied potential.The overpotential is equal to the potential differencebetween the Fermi level of the semiconductor or metal andthe Nernst potential of the metal/metal ion redox couple.At semiconductor surfaces, the kinetics of deposition atsmall overpotentials depend on the band bending in thesemiconductor under equilibrium conditions as shown infigure 1. As discussed in section 2.2.1, the rate of chargetransfer may be controlled either by the space charge layeror by the Helmholtz layer. In the case of space charge

1946


layer control, the built-in potential determines the densityof conduction band electrons at the surface and; hence,the rate of deposition. As a consequence, for the case ofdirect transfer the value ofβ in equation (41) need not bedetermined by the Butler–Volmer equation (equation (20))insofar as the rate-limiting step for attachment of the atomconverting the cluster to a critical cluster may be determinedby the potential dependence of the electron density at thesurface,ns . This is especially true whenNcrit < 1, sincethe attachment of the first atom has to correspond to thetransfer of an electron from the silicon to the solution.For the case in which the applied potential,U , is droppedentirely over the semiconductor and(U − Ueq) = η, thefollowing relation holds:

d ln(i)

d|η| =d ln(ns)

d|η| =e

kT(55)

wherei is the current density for the mechanism of directelectron transfer from the conduction band to the solution.As a consequence, for the case in which the mechanism ofattachment is direct electron transfer,β = 1 (instead ofαc).

In general, however, the potential distribution atthe semiconductor/solution interface is more complicated.At negative potentials, the semiconductor capacitanceincreases upon shifting the potential to more negativevalues. Consequently, the applied potential may bepartitioned over the space charge layer and the Helmholtzlayer, as discussed in section 2.1. In the previous section itwas shown that, for a silicon surface covered with a goldfilm in solution, the applied potential is dropped across theHelmholtz layer at the gold/solution interface at sufficientlynegative potentials. Similarly, since nucleation occurs atlarge negative potentials, the applied potential is likely to bepartitioned, which would result in a mixing in of the transfercoefficientαc through the parameterγac, as described insection 2.1 and equation (12). In this case, the value ofβ

would be determined by the value ofγac.As a consequence of the built-in band bending at

equilibrium, the kinetics for charge transfer and, hence,nucleation are generally slow. It has been shown that theexchange current density at semiconductor surfaces is oftenorders of magnitude lower than that at metal substrates[16]. As a result, electrochemical deposition of metals isusually characterized by a large critical overpotential fornucleation. This is also observed for nucleation of gold onsilicon, for which the nucleation overpotential was foundto be about 300 mV. For deposition of metals onto metalsubstrates, the critical overpotential is usually on the orderof 3–10 mV [63] and deposition potentials are usually in therange 5–100 mV. Figure 20 shows that the current transientsrecorded for the nucleation of gold were in the overpotentialrange 0.4–0.8 V. Equation (35) shows that the number ofatoms required to form a critical nucleus is proportionalto |η|−3, hence, it is not surprising thatNcrit may be verysmall for deposition onto semiconductors.

A further complication arises from the semiconductingproperties of the substrate. For example, electrontransfer from a metal nucleus to an n-type semiconductor,corresponding to dissolution of the nucleus, may be veryslow due to the absence of holes in the valence band.

For deposition of metals onto metal substrates, an ohmiccontact is always formed; this is not generally true forsemiconductors. Gold, platinum and, to a lesser extent,copper form rectifying contacts on n-type silicon, as wasshown in previous sections. Hence, in order for an electronto be injected from the metal into the conduction band, alarge energy barrier has to be overcome. As a consequence,the deposition reaction is not reversible; when an atomof metal is deposited onto a semiconductor substrate, itcannot be dissolved electrochemically [68]. This makesthe concept of a critical nucleus size somewhat questionableinsofar as it assumes equilibrium between the reduction andoxidation processes.

As a final remark, the electrical properties of metal filmson semiconductors are fairly well documented; however,the properties of a semiconductor surface with small metalclusters are largely unknown. The properties of thesemiconductor/metal cluster/solution system may be veryimportant in determining the mechanisms of nucleation andgrowth and thus the ability to fabricate adherent, continuousmetal films on semiconductors.

7. Summary

In this paper, the thermodynamics and kinetics of depositionof metals onto semiconductor surfaces from solution arediscussed. We have analysed the potential distribution atthe semiconductor/solution interface and the influence ofthe band structure of semiconductors on charge transfermechanisms is described.

Metal deposition can be achieved by three processes:electron transfer from the conduction band to metal ionsin solution (n-type semiconductors or illuminated p-typesemiconductors), electron transfer from surface states to thesolution and electron transfer from the valence band to thesolution, namely injection of holes. The advantages anddisadvantages of these mechanisms have been discussed.In some cases, the mechanism can be tailored by choiceof the solution: the positions of the band edges of manysemiconductors shift in a Nernstian fashion with the pHand the potential of the metal ion/metal redox couple can beadjusted by complexation of the ion. As an example, it wasshown that gold can be deposited onto n-type silicon eithervia the valence band, using AuCl−4 , or by the conductionband, using Au(CN)−2 .

As soon as metal nuclei are deposited onto asemiconductor surface, the energetics and kinetics of furtherdeposition change. Various effects may play a role: themetal nuclei can act as catalytic sites for charge transferprocesses and, hence, further deposition may be morefavourable at existing nuclei. It has been shown previouslythat this can be interpreted as charge transfer via surfacestates. In addition to this enhancement of kinetics, theenergetics of the interface also change: depending on theparticular system, the metal deposit may form either anohmic contact or a Schottky barrier. This has a largeeffect on the deposition characteristics of a metal layerand complicates the application of conventional theories toelectrochemical deposition of metals onto semiconductors.

1947

G Oskam et al

The properties of metal-coated semiconductor elec-trodes determine the energetics and kinetics of electroplat-ing of thicker films and direct measurements of the potentialof the metal during experiments provide useful information.Since there is no external contact to the metal layer underthe experimental conditions, the potential of the metal layeris determined by the relative rates of charge transfer at thesemiconductor/metal interface and at the metal/solution in-terface. An example of this situation was presented forn-Si/Au electrodes in K4Fe(CN)6 solution. It was shownthat, at positive potentials, the potential of the gold filmsis pinned to the equilibrium potential of the redox cou-ple, whereas at negative potentials the potential of the goldlayer is equal to that of the silicon. These results could beexplained by invoking the high Schottky barrier betweenn-type silicon and gold.

The theory of nucleation and growth for depositionof metals onto metals or semiconductors was reviewed indetail. Nucleation and growth onto semiconductors usuallyfollow either instantaneous or progressive nucleation andsubsequent diffusion-limited growth of the nuclei. Wediscussed experimental methods to determine the rateof nucleation for progressive nucleation and the densityof nuclei for instantaneous nucleation. The potentialdependence of these parameters was discussed in theframework of several models. It was shown that, forthe solutions used in this work, deposition of gold onton-type silicon follows progressive nucleation, whereas fordeposition of copper instantaneous nucleation is observed atsufficiently negative potentials. The nucleation of gold wasanalysed in detail and it was found that the critical nucleussize is smaller than one atom: the physical interpretationof this result and possible explanations were discussed.

The results of the experimental studies on depositionof gold, copper and platinum were used to determine theconditions required to deposit adherent, continuous metalfilms on silicon forming a contact of high electronic quality.It was shown that a large density of nuclei is needed and thatthe rate of growth should be kinetically limited in order forthe nuclei to coalesce. This method proved to be successfulin the three cases presented here.

References

[1] Romankiw L T and Palumbo T A 1988ElectrodepositionTechnology, Theory and Practiceed L T Romankiwand D R Turner (Pennington, NJ: ElectrochemicalSociety) p 13

[2] Searson P C and Moffat T P 1994 Electrochemicalsurface modification and materials processingCriticalRev. Surf. Chem.3 171

[3] Budevski E B 1983Comprehensive Treatise ofElectrochemistryvol 7, ed B E Conwayet al(New York: Plenum) p 399

[4] Staikov G, Budevski E, Hopfner M, Obretenov W,Juttner K and Lorenz W J 1993Surf. Sci.248 234

[5] Sonnenfeld R 1986Appl. Phys. Lett.49 1172[6] Gao X, Hamelin A and Weaver M J 1991Phys. Rev. Lett.

67 618[7] Manne S, Hansma P K, Massie J, Elings V B and

Gewirth A A 1991 Science251 183[8] Magnussen O M, Hotlos J, Nichols R J, Kolb D M and

Behm R J 1990Phys. Rev. Lett.64 2929

[9] Gerischer H 1970Physical Chemistry: An AdvancedTreatisevol 9A, ed H Eyring (New York: Academic)p 463

[10] Gerischer H 1990Electrochim. Acta35 1677[11] Bard A J and Faulkner L R 1980Electrochemical

Methods: Fundamentals and Applications(New York:Wiley)

[12] Morrison S R 1980Electrochemistry at Semiconductorand Oxidized Metal Electrodes(New York: Plenum)

[13] Many A, Goldstein Y and Grover N B 1965Semiconductor Surfaces(Amsterdam: North-Holland)

[14] Schroder K and Memming R 1985Ber. Bunsenges. Phys.Chem.89 385

[15] Oskam G, Hoffmann P M, Schmidt J C and Searson P C1996J. Phys. Chem.100 1801

[16] Vanmaekelbergh D 1997Electrochim. Acta42 1121[17] Vanmaekelbergh D 1997Electrochim. Acta42 1135[18] Oskam G, Schmidt J C, Hoffmann P M and Searson P C

1996J. Electrochem. Soc.143 2531[19] Oskam G, Schmidt J C and Searson P C 1996

J. Electrochem. Soc.143 2538[20] Schefold J 1992J. Electroanal. Chem.341 111[21] de Wit A R, Vanmaekelbergh D and Kelly J J 1992

J. Electrochem. Soc.139 2508[22] Oskam G, Hoffmann P M and Searson P C 1996Phys.

Rev. Lett.76 1521[23] Dewald J F 1960Bell Syst. Tech. J.39 615[24] Koval C A and Howard J N 1992Chem. Rev.92 411[25] Smith B B and Nozik A J 1996J. Chem. Phys.205 47[26] Miller R J D, McLendon G L, Nozik A J, Schmickler W

and Willig F 1995Surface Electron Transfer Processes(New York: VCH)

[27] Fajardo A M and Lewis N S 1996Science274 969[28] Memming R 1997Solid State Ionics94 131[29] Madou M J, Loo B H, Frese K W and Morrison S R 1981

Surf. Sci.108 135[30] Heller A, Aharon-Shalom E, Bonner W A and Miller B

1982J. Am. Chem. Soc.104 6942[31] Miller B, Heller A, Menezes S and Lewerenz H J 1980

Faraday Discuss. Chem. Soc.70 1[32] Nakato Y, Ueda K, Yano H and Tsubomura H 1988

J. Phys. Chem.92 2316[33] Szklarczyk M and Bockris J O’M 1984J. Phys. Chem.88

1808[34] Chandra N, Wheeler B L and Bard A J 1985J. Phys.

Chem.89 5037[35] Yoneyama H, Azuma H and Tamure H 1985

J. Electroanal. Chem.186 247[36] Cook R L, Dempsey P F and Sammels A F 1985

J. Electrochem. Soc.132 1315[37] Chazalviel J-N and Stefenel M 1983Surf. Sci.134 865[38] Allongue P and Souteyrand E 1989Electrochim. Acta34

1027[39] Allongue P, Souteyrand E and Allemand L 1989

J. Electrochem. Soc.136 1027[40] Jung Ch and Kolb D M 1990Ber. Bunsenges. Phys.

Chem.94 861[41] Oskam G, Vanmaekelbergh D and Kelly J J 1993

Electrochim. Acta38 301[42] Gomes W P, Lingier S and Vanmaekelbergh D 1989

J. Electroanal. Chem.269 237[43] Vanmaekelbergh D, Gomes W P and Cardon F 1982

J. Electrochem. Soc.129 546[44] Lin P S D,Marcus R B and Sheng T T 1983

J. Electrochem. Soc.130 1878[45] Verhaverbeke S, Meuris M, Mertens P W, Heyns M M,

Philipossian A, Graf D and Snegg A 1991TechnicalDigest of the Int. Electron Devices Meetingp 71

[46] Sze S M 1981Physics of Semiconductor Devices(New York: Wiley) pp 21, 35–38

[47] Hsu E, Parks H G, Craigin R, Tomooka S, Ramberg J Sand Lowry R K 1992J. Electrochem. Soc.139 3659

1948


[48] Morinaga H, Suyama M and Ohmi T 1994J. Electrochem. Soc.141 2834

[49] Norga G J, Platero M, Black K A, Reddy A J, Michel Jand Kimerling L C 1997J. Electrochem. Soc.1442801

[50] D’Asaro L A, Nakahara S and Okinaka Y 1980J. Electrochem. Soc.127 1935

[51] Ting C H and Paunovic M 1989J. Electrochem. Soc.136456

[52] Furukawa S and Mehregany M 1996Sensors ActuatorsA56 261

[53] Nagahara L A, Ohmori T, Hashimoto K and Fujishima A1993J. Vac. Sci. Technol.A 11 763

[54] Gorostiza P, Servat J, Morante J R and Sanz F 1996ThinSolid Films275 12

[55] Chang C-A 1989J. Appl. Phys.67 566[56] Li J and Sacham-Diamand Y 1992J. Electrochem. Soc.

139 L37[57] Rhoderick E H and Williams R H 1978

Metal–Semiconductor Contacts(New York: OxfordUniversity Press)

[58] Bindra P, Gerischer H and Kolb D M 1977J. Electrochem. Soc.124 1012

[59] Reineke R and Memming R 1987Surf. Sci.192 66[60] Allongue P and Souteyrand E 1987J. Vac. Sci. Technol.B

5 1644[61] Oskam G, Bart L, Vanmaekelbergh D and Kelly J J 1993

J. Appl. Phys.74 3238[62] Oskam G, Vanmaekelbergh D and Kelly J J 1993

Electrochim. Acta38 1115[63] Budevski E, Staikov G and Lorenz W J 1996

Electrochemical Phase Formation and Growth(Weinheim: VCH)

[64] Allongue P and Souteyrand E 1989J. Electroanal. Chem.269 361



[67] Scherb G and Kolb D M 1995J. Electroanal. Chem.396151

[68] Vereecken P M, Strubbe K and Gomes W P 1997J. Electroanal. Chem.433 19

[69] Volmer M and Weber A 1926Z. Phys. Chem.119 277[70] Becker R and Doering W 1935Ann. Phys., Lpz.24 719[71] Walton D 1962J. Phys. Chem.37 2182[72] Walton D 1969Nucleationed A C Zettlemoyer

(New York: Marcel Dekker)[73] Stoyanov S 1978Current Topics in Materials Science

vol 3, ed E Kaldis (Amsterdam: North Holland)[74] Stoyanov S 1973Thin Solid Films18 91[75] Milchev A and Vassileva E 1980J. Electroanal. Chem.

107 323[76] Milchev A, Vassileva E and Kertov V 1980

J. Electroanal. Chem.107 337[77] Milchev A and Stoyanov S 1976J. Electroanal. Chem.72

33[78] Gunawardena G, Hills G and Montenegro I 1982

J. Electroanal. Chem.138 241

[79] Southampton Electrochemistry Group 1990InstrumentalMethods in Electrochemistry(New York: EllisHorwood)

[80] Gunawardena G, Hills G, Montenegro I and Scharifker B1982J. Electroanal. Chem.138 225

[81] Scharifker B R and Hills G J 1983Electrochim. Acta28879

[82] Bosco E and Rangarajan S K 1982J. Electroanal. Chem.134 213

[83] Abyaneh M Y 1982 Electrochim. Acta27 1329[84] Milchev A 1983Electrochim. Acta28 947[85] Sluyters-Rehbach M, Wijenberg J H O J,Bosco E and

Sluyters J H 1987J. Electroanal. Chem.236 1[86] Mirki n M V and Nilov A P 1990J. Electroanal. Chem.

283 35[87] Milchev A, Kruijt W S, Sluyters-Rehbach M and

Sluyters J H 1993J. Electroanal. Chem.350 89[88] Jacobs J W M 1988J. Electroanal. Chem.247 135[89] Gagnaire A, Joseph J, Etcheberry A and Gautron J 1985

J. Electrochem. Soc.132 1655[90] Oskam G, Long J G, Nikolova M and Searson P C 1997

Electrochemical Synthesis and Characterization ofMaterials ed P C Andricacoset al (New York:Materials Research Society) p 257

[91] Higashi G S and Chabal Y J 1993Handbook ofSemiconductor Wafer Cleaning Technology: Science,Technology and Applicationsed W Kern (Ridge, NJ:Noyes) pp 433–96

[92] Ubara H, Imura T and Hiraki A 1984Solid StateCommun.50 673

[93] Graf D, Grundner M and Schulz R 1989J. Vac. Sci.Technol.A 7 808

[94] Yablonovitch E, Allara D L, Chang C C, Gmitter T andBright T B 1986Phys. Rev. Lett.57 249

[95] Tomita E, Matsuda N and Itaya K 1990J. Vac. Sci.Technol.A 8 534

[96] Watanabe S, Nakayama N and Ito T 1991Appl. Phys.Lett. 59 1458

[97] Hessel H E, Feltz A, Reiter M, Memmert U andBehm R J 1991Chem. Phys. Lett.186 275

[98] Searson P C 1995Advances in Electrochemical Scienceand Engineeringvol 4, ed H Gerischer andC W Tobias (Weinheim: VCS) p 69

[99] Allongue P, Brune H and Gerischer H 1992Surf. Sci.275414

[100] Allongue P, Costa-Kieling V and Gerischer H 1993J. Electrochem. Soc.140 1009

[101] Allongue P, Costa-Kieling V and Gerischer H 1993J. Electrochem. Soc.140 1021

[102] Bard A J, Parsons R and Jordan J (eds) 1985StandardPotentials in Aqueous Solutions(New York: MarcelDekker)

[103] Beltowska-Brzezinska M, Dutkiewicz E and Lawicki W1979J. Electroanal. Chem.99 341

[104] Wolf S 1995Silicon Processing for the VLSI Eravol 1(Sunset Beach, CA: Lattice)

[105] Quickenden T I and Qingzhong Xu 1996J. Electrochem.Soc.143 1248

[106] Oskam G and Searson P C 1998 in preparation

1949

Date post:	06-Oct-2020
Category:	Documents
Upload:	others
View:	5 times
Download:	0 times

REVIEW ARTICLE Electrochemical deposition of metals onto ......In semiconductor electrochemistry,...

Documents