Composition dependent characterization of copper indium diselenidethin film solar cells synthesized from electrodeposited binary selenideprecursor stacks

Johannes Fischer a, Jes K. Larsen a,1, Jerome Guillot b, Yasuhiro Aida c, Tobias Eisenbarth a,David Regesch a, Valerie Depredurand a, Nicole Fevre a, Susanne Siebentritt a,Phillip J. Dale a,n

a Laboratory for Photovoltaics, University of Luxembourg, 41, rue du Brill, L-4422 Belvaux, Luxembourgb Centre de Recherche Public Gabriel Lippmann, 41, rue du Brill, L-4422 Belvaux, Luxembourgc TDK corporation, Device Development Center, Ichikawa, Chiba 272-8558, Japan

a r t i c l e i n f o

Article history:Received 11 July 2013Received in revised form23 January 2014Accepted 22 March 2014

Keywords:ElectrodepositionCIGS solar cellsIndium selenideCopper selenidePhotoluminescenceInterface recombination

a b s t r a c t

CuInSe2 solar cells were prepared from electrodeposited and annealed stacks of indium selenide andcopper selenide on molybdenum (Mo/In2Se3/Cu–Se). In comparison to a simultaneous electrodepositionof all elements at once the electrodeposited binary selenide stack leads to larger grains and latticecoherence length if annealed under the same conditions (30 min at 550 1C). Absorbers with atomic Cu/Inratios o1, 1 and 41 and their corresponding solar cells were characterized. An incomplete reaction incase of the Cu-poor precursor caused a Cu-deficient layer at the back contact of the absorber leading tothe formation of a reverse electronic barrier reducing fill factor and short circuit current and thus thesolar power conversion efficiency. Photoluminescence measurements showed a strongly compensatedsemiconductor for the Cu-poor absorber and an incomplete charge carrier collection was identified byquantum efficiency measurements under reverse bias. The reverse electronic barrier, the highcompensation and the incomplete carrier collection could be avoided by Cu-rich growth conditions.The appearance of an excitonic transition in photoluminescence indicated a high semiconductor qualityin this case. It was linked with a high quantum efficiency resulting in a local short-current density of38.3 mA/cm2. Temperature-dependent JV measurements identified interface recombination as thelimiting recombination loss mechanism for the Cu-rich grown solar cell reducing the open-circuitpotential and decreasing the conversion efficiency. The solar cell prepared from the precursor with Cu/InE1 had similar properties as the Cu-poor device without the reverse electronic barrier at the backcontact but instead dominating interface recombination. It reached a best solar power conversionefficiency of 5.5%.

& 2014 Elsevier B.V. All rights reserved.

1. Introduction

The highest performance of thin film solar cells has beenreached with the compound semiconductor Cu(In,Ga)Se2 [1].Electrodeposition followed by selenization is a low-cost andmaterial-efficient method for the fabrication of this semiconductorand its underlying compound CuInSe2 (CIS) [2,3]. In principal thefollowing three routes can be applied for the electrodeposition ofCIS precursors [4]: (i) co-electrodeposition of all elements (Cu, In,Se) at once in the deposition process [5]; (ii) electrodeposition of a

precursor stack consisting of indium selenide and copper selenide(In–Se, Cu–Se) [6–8]; (iii) electrodeposition of a metal(-alloy)precursor stack (Cu, In, Cu–In) [9,10]. A combination of theseroutes is also possible (e.g. a Cu/In–Se stack) [11]. The co-electrodeposited precursor in route (i) is nano-crystalline. Previousexperiments have shown that its recrystallization during theannealing process can be insufficient and the grain size in theabsorber stays below 100 nm [12]. Such a small grain size isundesirable, because the numerous grain boundaries can act asnon-radiative recombination centers reducing the solar cell per-formance. Absorbers prepared from metal stacks in route (iii) canpeel during the selenization process due to the profound volumeexpansion of the layer with the incorporation of selenium [13]. Anelectrodeposited precursor stack of binary selenides according toroute (ii) is expected to solve the issues of low recrystallization

Contents lists available at ScienceDirect

journal homepage: www.elsevier.com/locate/solmat

Solar Energy Materials & Solar Cells

http://dx.doi.org/10.1016/j.solmat.2014.03.0450927-0248/& 2014 Elsevier B.V. All rights reserved.

n Corresponding author. Tel.: !352 46644 6279; fax: !352 466444 6602.E-mail address: phillip.dale@uni.lu (P.J. Dale).1 Present address: Institute of Energy Conversion, University of Delaware,

Newark, USA.

Solar Energy Materials & Solar Cells 126 (2014) 88–95

and peeling. Experiments on the selenization of sputtered pre-cursors have also resulted in a considerable shorter selenizationtime if the precursor does not consist of metal layers exclusivelybut contains a layer of indium selenide [14]. In co-electrodepositedprecursors the thermodynamically stable CIS phase is alreadypresent and the reduction of the number of grain boundaries isthe only driving force for recrystallization. In binary selenideprecursors the favorable thermodynamic reaction between theprecursor phases provides an additional driving force. The den-sities of the binary selenide are close to the density of CIS and thevolume expansion during annealing is reduced in comparison tometal stack precursors.

Hermann et al. studied the synthesis of CIS from an electro-deposited stack of Mo/Cu/In–Se/Cu–Se [6–8]. They successfullyprepared smooth, single-phase CIS absorber layers. No solar cellprepared from an electrodeposited stack of In–Se and Cu–Se hasbeen reported so far.

Only a few solar cells prepared from an In–Se/Cu–Se precursor bythe use of vacuum methods have been published. Anderson et al.deposit the precursor by Migration Enhanced Epitaxy and anneal itunder Se flux [15]. The finished solar cell has a power conversionefficiency of 5.1%. Park et al. prepared the precursor stack by physicalvapor deposition and obtained a best efficiency of 5.4% [16].

This work presents the first CIS solar cell prepared from anelectrodeposited precursor of In–Se and Cu–Se. Different growthcompositions are studied and their limitations are given.

2. Material and methods

2.1. Sample preparation

The precursors were prepared by electrodeposition from aqueoussolutions. The indium selenide layer was electroplated on Mo-coatedsoda-lime glass substrates (2.5"2.5 cm2) from an electrolyte con-taining 1.89 mmol/kg (of solvent) In2(SO4)3, 1 mmol/kg H2SeO3, and100 mmol/kg K2SO4. The pH value of the bath was stabilized at 2 by abuffer prepared from potassium hydrogen phthalate and sulfamicacid. The deposition was carried out under nitrogen atmosphere at80 1C and #0.60 V (versus a 3 M KCl Ag|AgCl reference electrode) ona rotating disc electrode (RDE) with a rotation speed of 400 rpm.These conditions result in a smooth, shiny gray, and amorphousindium selenide film with an atomic Se/In ratio of 1.5 [17]. Thecopper selenide layer was plated on top of this film from anelectrolyte containing 2.6 mmol/kg CuCl2, 5.9 mmol/kg H2SeO3, and232 mmol/kg LiCl. The pH value was buffered at 3 with potassiumhydrogen phthalate and sulfamic acid. The deposition was done atroom temperature and #0.30 V on a RDE at a rotation speed of100 rpm. The plated copper selenide film is black and opaque withan atomic Cu/Se ratio of 0.76 determined by inductively coupledplasma –mass spectroscopy (ICP-MS). In the case of a “stoichiometricprecursor” a charge densities of 3.63 C/cm2 and 3.10 C/cm2 have beenplated for indium selenide and copper selenide, respectively. It hasbeen verified by ICP-MS composition measurements that this ratio ofcharge densities results in a composition with an atomic Cu/In ratioof 1. Different compositions of the precursor stack were prepared byadjusting the charge density of the copper selenide layer.

For reference purpose a co-electrodeposited precursor has beenelectroplated at room temperature on a stationary electrode froma bath containing 2.6 mM CuCl2, 9.6 mM InCl3, 5.5 mM H2SeO3,and 240 mM LiCl in a pH 3 buffered solution [12].

The precursors were annealed for 30 min at 550 1C in a graphitebox with an inner volume of about 20 cm3 inside a resistance heatedquartz tube furnace. 100 mg Se powder was added to the graphitebox and the furnace was filled with a mixture of 10% H2 in N2 at10 mbar.

Before processing the absorber to a solar cell, possiblecopper selenide phases on its surface were removed by etchingwith a 5% potassium cyanide (KCN) solution for 2 min [18].Directly afterwards a 60 nm thick CdS layer was deposited bychemical bath deposition [19]. By mistake the CdS layer of the Cu-rich grown sample is slightly thinner (ca. 50 nm). The solar cellwas finished by adding a 130 nm intrinsic ZnO layer and a 540 nmthick aluminum doped ZnO layer by sputtering and evaporating aNi–Al grid for electrical contact.

The electrical performance of the solar cells is compared to aCuInSe2 absorber grown by physical vapor deposition (PVD). A typicalCu-poor three-stage coevaporation process was applied [20,21].

2.2. Characterization techniques

The thin films were characterized by scanning electron micro-scopy (SEM), energy dispersive spectroscopy (EDS), X-ray diffraction(XRD), Auger electron spectroscopy (AES), and photoluminescence(PL). The EDS spectrumwas recorded with a primary electron energyof 20 keV in a system calibrated with a stoichiometric CuInSe2standard. The XRD diffractograms were recorded with a Cu tube(Kα1 and Kα2 lines) in a grazing incidence geometry at an incidenceangle of 41. The XRD peak width was determined from highlyresolved diffractograms recorded in a theta–2 theta geometry witha parallel beam. AES measurements were combined with Ar!

sputtering (3 kV, 3 μA, 451 incidence angle) of the samples to obtainthe composition depth profile. The AES spectra were recorded with aprimary electron beam of 10 kV and 1 nA on a system calibrated witha stoichiometric CuInSe2 standard. PL measurements were per-formed at 10 K with the 514.5 nm line of an argon ion laser and anexcitation power between 1 mW and 50 mW. After calibrating thesystem with a lamp of known intensity the same set up was used todetermine a quantitative PL spectrum at room temperature.

The finished solar cell with an area of 0.5 cm2 was character-ized by (temperature dependent) current density–voltage (JV)measurements and (voltage dependent) external quantum effi-ciency (QE) measurements. JV measurements were recorded in ahome-built set up using a halogen lamp with a dichroic reflector asillumination source being adjusted to 100 mW/cm2 with a certi-fied reference solar cell. In a similar set up the solar cell was placedin a cryostat to control its temperature between 100 K and roomtemperature. QE measurements were recorded on a 1–2 mm2 spotin the middle of the solar cell illuminated with monochromaticlight and the intensity calibrated with a certified reference diode.The spectral response was measured by a lock-in amplifier. Anapparent QE was measured with a reverse voltage bias appliedbetween front and back contact of the solar cell.

2.3. Formulae used for the interpretation of data

2.3.1. Scherrer formulaThe Scherrer formula is used to determine the lattice coherence

length D of a polycrystalline material from the width of its XRDreflections [22]: D$ λ= βsize%2θ& cos %θ&

! "where D is the average

coherence length, λ is the wavelength of the X-ray radiation, and θis the diffraction angle. The integral breadth βobs of the XRD reflectionfor a Pseudo-Voigt profile is determined by the integral breadth βsizecaused by a finite coherence length and the integral breadth βinstrcaused by the instrument [23]: βsize ' βobs#β2

instr=βobs.

2.3.2. Photoluminescence measurements of highly compensatedsemiconductors

Highly compensated CIGS absorbers show a broad asymmetricPL spectrum [24,25]. The energy of the PL emission maximumdepends strongly on the laser excitation intensity Iexc . The increase

J. Fischer et al. / Solar Energy Materials & Solar Cells 126 (2014) 88–95 89

of the PL energy EPL can be described by EPL%Iexc& $ EPL%I0&!β log 10%Iexc=I0&, with I0 being a reference intensity. If the coeffi-cient β exceeds about 10 meV/decade it can be an indication for acompensated semiconductor [26].

2.3.3. Temperature-dependent JV-characteristicsThe activation energy Ea of the dominant recombination losses

in a solar cell can be identified by temperature-dependent mea-surements of the open-circuit potential VOC [27,28]: eVOC ' Ea#AkBT ln%J00=Jill& where e is the electron charge, A the diode factor,kB the Boltzmann constant, T the temperature, Jill the photocurrentdensity, and J00 corresponds to the saturation current density J0divided by an Arrhenius term exp%#Ea=AkBT&. The activationenergy Ea can be obtained from the extrapolation of the tempera-ture dependence of VOC for T-0 K. If the activation energy Ea issmaller than the band gap energy Eg (at room temperature) it canbe concluded that interface recombination is dominating in thesolar cell device [27–29].

2.3.4. Band gap determination from quantitative PLThe photon yield of the PL spectrum YPL ω% & can be described

by Planck's generalized law [30]: YPL%ω& $ CA%ω&ω2(exp%%ℏω#μ&=kBT&#1)#1; Cis a constant, A%ω& is the absorptivity, ℏω is the photonenergy, and μ is the splitting of the quasi-Fermi levels of the electronsand holes, respectively. If the absorptivity has been obtained by fittingPlanck's generalized law to the PL spectrum, the band gap Eg $ ℏωg

can be derived from ωg at A%ωg& $ 1=e [31].

3. Results and discussion

3.1. Absorber properties

This study focuses on three CISe absorber layers (and theircorresponding solar cells) which were prepared from precursors thatwould give the final absorber layer a Cu/In ratio of o1, 1, and 41.Table 1 lists the composition of the precursor, the annealed absorber,and the KCN etched absorber. The excess Se being present in theprecursor evaporates in the annealing process. KCN etching of theannealed absorber removes copper selenide phase and thus sample Cbecomes stoichiometric afterwards. The XRD diffractograms of theCu-poor sample A and the near-stoichiometric sample B show onlyCuInSe2 reflections together with Mo and MoSe2 reflections of thesubstrate. In case of the Cu-rich grown absorber C a shoulder appearsat the high angle side of the main CuInSe2 reflections and can beattributed to Cu2#xSe (Fig. 1) [32]. The shoulder disappears if thesample is KCN etched (not shown).

The binary selenide precursor has been motivated by anexpected improvement of recrystallization in the annealing process.SEM cross-sections allow a comparison of grain size before and

after annealing of the binary selenide absorber and the co-electrodeposited one (Fig. 2). The binary selenide precursor showsa structureless and amorphous In2Se3 layer covered by randomlystacked copper selenide platelets. The co-electrodeposited precur-sor consists of densely packed, fine grains. After annealing thesegrains have grown but their average size is still below 200 nm. Incontrast the binary selenide stack annealed under the same condi-tions shows micron-sized grains partly extending from the Mo backcontact to the surface of the layer. Although the surface is rougher,the morphology is comparable to a PVD-grown sample.

While a small grain size is typical for a co-electrodepositedprecursor, its recrystallization during the selenization can bedifferent from the behavior observed in Fig. 2. Several authorsreport a compact absorber consisting of micron-sized grains afterannealing the co-electrodeposited precursor in a selenium vaporatmosphere [34,35]. Thus the insufficient recrystallization, visua-lized in Fig. 2, is probably an effect of non-optimized selenizationconditions, likely to be due a too low selenium partial pressure.Guillemoles et al. propose liquid phase sintering involving ele-mental selenium as the important mechanism of recrystallization[36]. The larger grain size observed for the absorber layer formedfrom the binary selenide precursor over the co-electrodepositedprecursor might then be explained by the surplus of selenium inthe binary selenide precursor (Table 1) which could increase thepartial pressure of selenium in the oven during annealing. Anotherconsideration is that the grain growth mechanisms are different.For the binary selenide precursor, the grains are grown by thediffusion of metals through a single interface between the copperand indium selenide starting phases. For the co-electrodepositedprecursor, already consisting of nano sized CuInSe2 grains, graingrowth proceeds only due to the minimization of surface area, therate of which appears to depend on the selenium partial pressurein the annealing atmosphere.

In addition to grain size the lattice coherence length is anothermeasure for the crystallinity of a layer (Section 2.3.1). Table 2shows the lattice coherence lengths measured for the annealedco-electrodeposited sample and the absorbers A, B and C preparedfrom binary selenide precursors. The lattice coherence lengths ofthe absorbers A, B and C are significantly larger than for theannealed co-electrodeposited sample. This confirms that binaryselenide precursors experience an improved crystallization duringannealing in comparison to a co-electrodeposited sample annealedunder the same conditions.

Hermann et al. reported a superior recrystallization after annealingfor a binary selenide precursor in comparison to a co-electrodepositedprecursor [8]. They concluded this from the narrower line width of the

Table 1Composition (atomic ratios) of precursor, absorber and KCN etched absorber for thestudied growth conditions. The precursor Se/(Cu! In) ratios were calculated fromthe electrodeposited charges and the Se/In and Cu/Se atomic ratio of the indiumselenide and copper selenide film determined by ICP-MS. This calculation has beenverified by ICP-MS for a similar precursor. The atomic ratios of the absorber beforeand after KCN etching were measured by EDS. The error in the precursor atomicratio is about 70.08 and in the absorber atomic ratio 70.05 if not statedotherwise.

Sample Precursor After annealing After KCN etching

Se/(Cu! In) Cu/In Se/(Cu! In) Cu/In Se/(Cu! In)

A 1.45 0.92 1.03 0.91 1.02B 1.44 0.98 0.99 0.98 0.97C 1.43 1.4770.12 0.92 1.02 1.00

Fig. 1. Grazing-incidence XRD diffractograms of the as-deposited absorbers A, B,and C. CuInSe2-reflections are exemplary labeled with their Miller indices (hkl) forsample C [33]. For samples A and B only the maxima not corresponding to CuInSe2are marked. The Cu-rich grown sample C has overlapping reflections of Cu2#xSe.

J. Fischer et al. / Solar Energy Materials & Solar Cells 126 (2014) 88–9590

XRD reflections in case of the absorber annealed from the binaryselenide precursor. However this comparison is not representative,because the co-electrodeposited sample showed a very poor recrys-tallization (FWHM of (112) reflection in XRD above 11).

PL measurements allow conclusions on the optoelectronicquality of the absorbers (Fig. 3). The Cu-poor grown absorbershows a broad asymmetric maximum in its PL spectrum (Fig. 3,sample A). The shape is common for Cu-poor grown CuInSe2 andcan be attributed to a highly compensated semiconductor as it wasreported by Dirnstorfer et al. [24] (Section 2.3.2). The energy ofthe PL maximum intensity profoundly increases with the laser

excitation intensity (β$19 meV/decade) giving further evidencefor a highly compensated semiconductor.

The PL spectrum of the near-stoichiometric grown absorber(Fig. 3, sample B) has a similar shape as the Cu-poor grown one. Incombination with the measured coefficient β$13 meV/decade thisabsorber is again identified as a highly compensated one, althoughthe degree of compensation is lower than for the Cu-poor grown

Fig. 2. SEM pictures on cross-sections of (a) a Mo/In2Se3/Cu–Se precursor stack, (b) the same precursor after annealing at 550 1C for 30 min; (c) a co-electrodepositedprecursor, (d) the same precursor after annealing at 550 1C for 30 min; (e) a CuInSe2 absorber grown by PVD. The absorber prepared from the binary selenide precursor stackshows larger grains than the one prepared from a co-electrodeposited precursor (consider the different scale in (a)).

Table 2Integral breadth βobs(2θ) and coherence length measured by high resolution XRD onthe annealed and KCN etched absorbers. βobs(2θ) was extracted from a Pseudo-Voigtfit of the XRD reflection with the software WINPLOTR [37]. The correspondingcoherence length is calculated by the Scherrer formula (Section 2.3.1). Theinstrument resolution has been measured on the (200)-reflection of a GaAs wafer.

Sample (112) Reflection

!obs(2") (1) Coherence length (nm)

A 0.086 150720B 0.068 260750C 0.074 210740Co-electrodeposited 0.189 5075Instrument 0.047

Fig. 3. Photoluminescence spectra of the absorbers A, B and C measured at 10 Kand a laser excitation power of 50 mW. The inset shows the magnified high energyside of the spectrum for sample C. The high energy part of the spectrum has beenapproximated by two Gaussian functions to visualize the signal caused by theexciton transition.

J. Fischer et al. / Solar Energy Materials & Solar Cells 126 (2014) 88–95 91

sample. Since the degree of compensation is correlated to theCu-poorness of grown absorbers, the lower compensation confirmsthe EDS measurements showing that the Cu/In ratio of sample B ishigher than sample A. Although the PL spectrum of the Cu-richgrown absorber (Fig. 3, sample C) is again very broad, this time itis not a sign of a highly compensated semiconductor, becauseβ$3 meV/decade. Therefore the spectrum must be explained bybroad, overlapping donor–acceptor transition peaks due to broadenergy distributions of defects. The main peak position at 0.97 eVcorresponds to the donor–acceptor transition DA2 which is typicallyobserved for Cu-rich grown absorbers [25]. The Cu-rich grownabsorber also exhibits a weak maximum at 1.05 eV which corre-sponds to the energy of a free exciton in CuInSe2 (Fig. 3, inset) [25].The energy of the maximum is independent on the laser excitationenergy being another indication for an exciton. The presence ofexcitonic transitions is usually a sign of a high semiconductorelectronic quality [38].

3.2. Solar cell properties

Fig. 4 shows the JV-characteristics of the solar cells preparedfrom the absorbers A, B, and C and Table 3 lists the correspondingparameters. The highest solar power conversion efficiency of 5.5%was reached for the near-stoichiometric grown absorber B. Com-pared to a co-electrodeposited solar cell or a typical three-stageprocess co-evaporated CuInSe2 solar cell the main performanceloss can be attributed to the low VOC. This means that the solarcells prepared from annealed electrodeposited binary selenideprecursors show higher recombination losses than CuInSe2 solarcells prepared by other methods. Although it has been shown inTable 2 that the lattice coherence length of the binary selenideabsorber is larger than the one of a co-electrodeposited absorber,the latter shows less recombination. This indicates that grainboundaries are not a dominant loss mechanism at the discussed

performance level. A low electronic activity of grain boundaries istypical in CIGS solar cells [39].

The solar cell prepared from the Cu-poor grown absorber Ashows an additional performance loss due to its low fill factor. TheJV-characteristic of this solar cell does not show the exponentialtrend expected from the diode model of a solar cell but has a “roll-over”-current for forward currents. A “roll-over”-current in the JV-characteristic can be a sign for a reverse barrier in the electronicbands of the semiconductor device [29].

The composition depth profile measured by AES (Fig. 5) revealsthe reason for the reverse barrier. At the Mo back contact of thesolar cell the In and Se content increases while the Cu contentdecreases. It is likely that an ordered copper vacancy phase ispresent in this region. At the depth where the In concentrationreaches its maximum the atomic ratio (Cu:In:Se) of the elements is1:3.3:5.8 being close to the composition of a CuIn3Se5 phase. Dueto the high annealing temperature it is expected that all phases arepresent in a crystalline modification. Nevertheless CuIn3Se5 phasecannot be identified in the XRD diffractogram, because its mainpeaks are close to CuInSe2. The intensity of the unique peaks is lowand taking into account that the phase is only present in a thinlayer at the back of the absorber they cannot be identified in thediffractogram. Thin films of the ordered vacancy compounds aretypically n-doped [40] causing the formation of reverse pn-junction at the back contact of the absorber. If a forward currentpasses the solar cell the reverse junction at the back contact willblock the current and cause the “roll-over” behavior. At voltagesbelow VOC the reverse junction causes a voltage drop and thusreduces the fill factor of the device. Since the precursor stack hasthe structure Mo/In2Se3/Cu–Se it can be assumed that the orderedvacancy phase at the back contact results from an incomplete

Fig. 4. JV-characteristics of solar cells prepared from absorbers A, B and C (Table 1).

Table 3JV parameters of the solar cells prepared from absorbers A, B, and C (Fig. 4). Solarpower conversion efficiency η, open circuit voltage VOC, short circuit current densityJsc, and fill factor FF. Jsc and η refer to total cell area. A co-electrodeposited solar cellgrown under comparable conditions and with a similar morphology as the samplein Fig. 2 was taken from [12]. Sample PVD is a typical Cu-poor CuInSe2 solar cellprepared by a three-stage physical vapor co-evaporation process in our group[20,21].

Sample η (%) VOC (V) Jsc (mA/cm²) FF

A 3.3 0.33 28.3 0.35B 5.5 0.34 30.9 0.49C 4.1 0.26 32.2 0.52Co-electro 6.7 0.40 31.2 0.52PVD 13.5 0.49 39.7 0.70

Fig. 5. AES composition depth profiles of (a) the Cu-poor grown absorber A, (b) theCu-rich grown absorber C; both absorbers have been etched in KCN before toremove possible copper selenide phases.

J. Fischer et al. / Solar Energy Materials & Solar Cells 126 (2014) 88–9592

reaction of the precursor in the annealing process. This is anunexpected result, because Koo et al., reported uniform composi-tion depth profiles for a similar binary selenide precursor afterannealing for 2 min at 550 1C [47]. The Mo/(In1#xGax)2Se3/CuSeprecursor stack was prepared by physical vapor deposition with anoverall Cu/(In!Ga) ratio between 0.80 and 0.95. The differencebetween the PVD precursor and our electrodeposited precursorcan probably be explained by the different morphology of thedeposited copper selenide layer. While Koo et al. showed a smoothand compact copper selenide layer, the electrodeposited layer isvery porous (Fig. 2a). Thus the interface area between the indiumselenide and copper selenide layer is tiny and limits the inter-diffusion of ions during the reaction.

If Cu excess is provided in the growth process (absorber C), thereaction is complete and no ordered vacancy phase is detected atthe back contact (Fig. 5b).

The reasonwhy the performance of the Cu-rich grown solar cellstays behind the performance of the near-stoichiometric grownone is a lower VOC caused by higher recombination losses (Fig. 4).The origin of these recombination losses can be identified bytemperature-dependent JV measurements. In Fig. 6 the VOC hasbeen plotted versus temperature for the solar cells A, B, and C. Thelinear trend is extrapolated to T-0 K in order to obtain theactivation energy Ea of the recombination process (Section 2.3.3).Although sample A has a reverse diode at the backside of theabsorber, VOC is not effected, if no photocurrent is collected at thereverse junction. Since there is no current flow under open circuitconditions the voltage drop over the reverse diode is 0. Thus theextrapolation towards T-0 K is still allowed to obtain Ea. Thedeviation of the low temperature data points from the linear trendindicates that the simple model of a reverse diode is not longersufficient at low temperatures.

While the band gap energy Eg (0.96–1.02 eV) is similar for allsolar cells the activation energy decreases profoundly with the Cu-content present in the absorber growth process. For the Cu-poorgrown solar cell Ea is 0.94 eV, i.e. close to the band gap. In case ofthe Cu-rich grown solar cell Ea is 0.63 eV and clearly below theband gap, thus suggesting strong interface recombination. Thistrend is in agreement with the results reported by Turcu et al. onco-evaporated Cu(In,Ga)(Se,S)2 solar cells [41,42]. Cu-rich grownsolar cells are typically dominated by interface recombinationwhich lowers the VOC as observed in Table 3. On the contrary theelectronic loss in high performing solar cells grown under Cu-poorconditions is dominated by bulk recombination. PL has shown aCu-poor signature for sample B (Fig. 3). Nevertheless the activationenergy of 0.80 eV is still significantly lower than Eg. Thus it must

be considered that beside the composition there are still furtherdefects at the interface of the near-stoichiometric grown sample B.

In comparison to a solar cell prepared from a physical vapordeposited absorber the cells prepared from electrodepositedabsorbers in Table 3 are also lacking short circuit current densityJsc. The Jsc of a solar cell is dependent on its band gap energy Eg,which has been determined for the absorbers in Table 3 byquantitative PL measurements (Section 2.3.4). The band gapenergy is 0.97 eV for the Cu-poor grown absorber A, 0.99 eV forthe near-stoichiometric grown absorber B, and 1.02 eV for the Cu-rich grown absorber C. The results are similar if the band gap isdetermined from the QE spectra (Fig. 7) by a plot of [ln(1#QE)]2

versus photon energy. Sample A has a band gap energy of 0.98 eV,sample B of 0.98 eV and sample C of 1.00 eV. The increase of Egwith the Cu-content during growth is in agreement with resultsreported on PVD CuInSe2 samples [43]. The different band gapdoes not explain the trend in Jsc observed in Table 3, since Jsc isrising with Cu-content, i.e. higher Eg.

Current loss mechanisms in solar cells can be determined by QEmeasurements (Fig. 7) [44,45]. The QE loss between 400 nm and500 nm is mainly caused by the optical absorption of the CdS bufferlayer. Since the Cu-rich grown sample was prepared with a slightlythinner CdS layer (Section 2.1), its QE is slightly higher in this range.Integrating over the QE spectrum of the Cu-rich sample C results in ashort-current density of 38.3 mA/cm2 (active area) being larger thanthe one recorded by JVmeasurements. The difference cannot be solelyassigned to the grid shading (about 5% of the cell area) but must beexplained by a non-uniform performance. While the QEmeasurementprobes a 1–2mm2 broad spot in the middle of the sample, the JV setup measures the current of the whole cell. The shape of the Cu-poor

Fig. 6. Temperature dependence of the open circuit potential for solar cellsprepared from absorbers A, B, and C. A linear fit is applied to the results andextrapolated to T-0 K. The horizontal lines mark the energy of the band gap atroom temperature measured by quantitative PL.

Fig. 7. External quantum efficiency measurements of solar cells prepared from aCu-poor grown absorber (A), a near-stoichiometric grown absorber (B), and a Cu-rich grown absorber (C).

Fig. 8. Ratio of the apparent quantum efficiency measured with an applied voltagebias of –0.5 V (reverse bias) and without bias.

J. Fischer et al. / Solar Energy Materials & Solar Cells 126 (2014) 88–95 93

grown sample and the near-stoichiometric grown sample is similar,although the absolute QE value is lower in the first case. In contrastto the Cu-rich grown sample these samples show a pronounced QEloss in the range 1000–1200 nm, which is usually attributed to aninsufficient absorption or carrier collection.

An insufficient carrier collection can be identified by measuringthe apparent QE with a reverse voltage bias applied to the solarcell. The reverse voltage bias widens the space charge region (SCR)and therefore improves carrier collection if the diffusion length ispoor. The improvement increases for longer wavelengths pene-trating deeper into the absorber. This behavior is observed for thesolar cells prepared from Cu-poor and near-stoichiometric grownabsorbers (Fig. 8). Thus incomplete carrier collection seems to bean important loss mechanism to their Jsc and is more pronouncedin case of the Cu-poor grown absorber. Nevertheless it cannotcompletely explain the behavior observed in Fig. 8, because also atshorter wavelengths around 700 nm an increase of apparent QE isobserved under reverse voltage bias. For these wavelengths mostphotons are absorbed in the top 150 nm of the absorber [46].Assuming a typical width of 230 nm for the SCR in an electro-deposited CuInSe2 absorber [12] the excited carriers are createdinside the SCR and thus do not benefit from the widening of theSCR by a reverse voltage bias. Besides an incomplete carriercollection another current loss mechanism being voltage depen-dent but wavelength independent must exist in the solar cellsprepared from Cu-poor and near-stoichiometric grown absorbers.Possible mechanisms are interface recombination or a photocur-rent barrier in the conduction band [29]. Guillemoles et al.measured a wavelength independent decrease of spectral responsefor co-electrodeposited CuInSe2 solar cells and assigned it tointerface recombination [36]. It was shown in Fig. 6 that interfacerecombination is also present in CuInSe2 solar cells prepared froma near-stoichiometric binary selenide stack. In case of the solar cellprepared from the Cu-poor grown absorber the reverse pn-junction at the back contact (Fig. 5) will also provide a wavelengthindependent loss mechanism for Jsc, because it causes a voltagedrop. The voltage drop decreases for a reverse voltage bias (beingin forward direction for the reverse diode at the back contact) andthus explains the increase of apparent QE measured in Fig. 8.

In summary the Cu-rich grown solar cell shows a good mino-rity carrier collection but is limited by interface recombinationreducing VOC. The open-circuit voltage is better for the near-stoichiometric grown solar cell, which on the other hand experi-ences higher current losses due to an incomplete carrier collectionand partly also because of interface recombination. The Cu-poorgrown solar cell behaves similar to the near-stoichiometric grownone. Interface recombination does not seem dominant any morebut a reverse junction at the backside of the absorber reduces fillfactor and Jsc.

4. Conclusions

The first CuInSe2 solar cell prepared from an electrodepositedbinary selenide stack has been presented reaching a best solarpower conversion efficiency of 5.5%. In comparison to a co-electrodeposited precursor annealed under the same conditionsthe binary selenide precursor leads to a larger grain size and latticecoherence length.

All growth conditions have been successful in producing work-ing devices. Each condition has specific benefits and disadvan-tages. The absorber formation of the Cu-poor precursor wasincomplete causing a reverse pn-junction at the back contact thatreduces fill factor and Jsc. Besides the reverse junction, theelectrical performance of the Cu-poor grown device is also limited

by an incomplete charge carrier collection. It is either caused by ashort electron diffusion length or a narrow space charge width.

The Cu-rich grown solar cell shows remarkable charge carriercollection. This can be the result of improved semiconductorquality which is concluded from the appearance of an excitonictransition in PL. The electrical performance of the solar cell islimited by a low VOC caused by interface recombination which is atypical problem with Cu-rich grown devices [41,42].

The best efficiency has been obtained for a near-stoichiometricprecursor, because it balances the benefits and disadvantages ofCu-poor and Cu-rich growth conditions. PL measurements identi-fied the absorber as being actually still Cu-poor. Despite the Cu-poor composition interface recombination is still dominant and afurther optimization of the interface necessary. Like the Cu-poorgrown solar cell also the near-stoichiometric one loses currentbecause of incomplete charge carrier collection. It might beimproved if the Cu content is slightly increased and the actualcomposition is closer to stoichiometry.

Further challenges for the solar cell synthesis from electrodepos-ited binary selenide precursor stacks are the electrodeposition of asmooth and compact copper selenide layer, the incorporation ofgallium in the indium selenide layer, and the optimization of theannealing conditions.

Acknowledgments

The authors would like to thank Michael Kirsch and co-workersat Helmholtz Zentrum Berlin (HZB) for completing the solar celldevices (TCO and grid). They also acknowledge ICP-MS composi-tion measurements by Johanna Ziebel and Cedric Guignard at theCentre de Recherche Public Gabriel Lippmann (CRP-GL). SEM/EDSmeasurements have been contributed by Jean-Christophe Lam-brechts (CRP-GL). Susan Schorr (HZB) provided help with Scherreranalysis of the XRD diffractograms.

This work was funded by Fonds National de la RechercheLuxembourg (project number C08/MS/02), TDK Corporation andthe University of Luxembourg.

References

[1] P. Jackson, D. Hariskos, E. Lotter, S. Paetel, R. Wuerz, R. Menner, et al., Newworld record efficiency for Cu(In,Ga)Se2 thin-film solar cells beyond 20%, Prog.Photovolt. 19 (2011) 894–897.

[2] R.N. Bhattacharya, Solution growth and electrodeposited CuInSe2 thin films, J.Electrochem. Soc. 130 (1983) 2040–2042.

[3] R.N. Bhattacharya, A.M. Fernandez, M.A. Contreras, J. Keane, A.L. Tennant,K. Ramanathan, et al., Electrodeposition of In–Se, Cu–Se, and Cu–In–Se thinfilms, J. Electrochem. Soc. 143 (1996) 854.

[4] D. Lincot, J. Guillemoles, S. Taunier, D. Guimard, J. Sicx-Kurdi, A. Chaumont,et al., Chalcopyrite thin film solar cells by electrodeposition, Sol. Energy 77(2004) 725–737.

[5] M.E. Calixto, K.D. Dobson, B.E. McCandless, R.W. Birkmire, Controlling growthchemistry and morphology of single-bath electrodeposited Cu(In,Ga)Se2 thinfilms for photovoltaic application, J. Electrochem. Soc. 153 (2006) G521–G528.

[6] A.M. Fernandez, M.E. Calixto, P.J. Sebastian, S.A. Gamboa, A.M. Hermann, R.N. Noufi, Electrodeposited and selenized (CuInSe2) (CIS) thin films for photo-voltaic applications, Sol. Energy Mater. Sol. C 52 (1998) 423–431.

[7] A.M. Hermann, R. Westfall, R. Wind, Low-cost deposition of CuInSe2 (CIS) filmsfor CdS/CIS solar cells, Sol. Energy Mater. Sol. C 52 (1998) 355–360.

[8] A.M. Hermann, M. Mansour, V. Badri, B. Pinkhasov, C. Gonzales, F. Fickett,et al., Deposition of smooth Cu(In,Ga)Se2 films from binary multilayers, ThinSolid Films 361–362 (2000) 74–78.

[9] V.K. Kapur, B.M. Basol, E.S. Tseng, Low cost methods for the production ofsemiconductor films for CuInSe2/CdS solar cells, Sol. Cells 21 (1987) 65–72.

[10] M.E. Calixto, P.J. Sebastian, CuInSe2 thin films formed by selenization of Cu–Inprecursors, J. Mater. Sci. 33 (1998) 339–345.

[11] C. Guillen, J. Herrero, Reaction pathways to CuInSe2 formation from electro-deposited precursors, J. Electrochem. Soc. 142 (1995) 1834–1838.

[12] P.J. Dale, A.P. Samantilleke, G. Zoppi, I. Forbes, L.M. Peter, Characterization ofCuInSe2 material and devices: comparison of thermal and electrochemicallyprepared absorber layers, J. Phys. D Appl. Phys. 41 (2008) 085105.

J. Fischer et al. / Solar Energy Materials & Solar Cells 126 (2014) 88–9594

[13] R. Caballero, C. Guillen, CuInSe2 formation by selenization of sequentiallyevaporated metallic layers, Sol. Energy Mater. Sol. C 86 (2005) 1–10.

[14] N. Romeo, A. Bosio, S. Mazzamuto, D. Menossi, A. Romeo, CIGS thin filmsprepared by sputtering and selenization by using In2Se3, Ga2Se3 and Cu assputtering targets, in: Proceedings of the 2010 35th IEEE PhotovoltaicSpecialists Conference (PVSC), 2010, pp. 000786–000788.

[15] T.J. Anderson, S.S. Li, O.D. Crisalle, V. Craciun, Fundamental Materials Researchand Advanced Process Development for Thin-Film CIS-Based Photovoltaics(Final Technical Report), 2 October 2001–30 September 2005, NationalRenewable Energy Laboratory (NREL), Golden, CO, 2006.

[16] S.C. Park, D.Y. Lee, B.T. Ahn, K.H. Yoon, J. Song, Fabrication of CuInSe2 films andsolar cells by the sequential evaporation of In2Se3 and Cu2Se binary com-pounds, Sol. Energy Mater. Sol. C 69 (2001) 99–105.

[17] J. Fischer, S. Siebentritt, P.J. Dale, Synthesis and characterization of CuInSe2 thinfilms by annealing of electrodeposited In2Se3/Cu and In2Se3/CuxSey Stacks, in:ECS Transactions, vol. 25, issue 15, October 4–October 9, 2009, Vienna, Austria,2010, pp. 129–142.

[18] R. Klenk, R. Menner, D. Cahen, H.W. Schock, Improvement of Cu(Ga,In)Se2based solar cells by etching the absorber, in: Proceedings of 1990 21st IEEEPhotovoltaic Specialists Conference (PVSC), 1990, pp. 481–486.

[19] M.A. Contreras, M.J. Romero, B. To, F. Hasoon, R. Noufi, S. Ward, et al.,Optimization of CBD CdS process in high-efficiency Cu(In,Ga)Se2-based solarcells, Thin Solid Films 403–404 (2002) 204–211.

[20] V. Depredurand, Y. Aida, J. Larsen, T. Eisenbarth, A. Majerus, S. Siebentritt,Surface treatment of CIS solar cells grown under Cu-excess, in: Proceedings ofthe 37th IEEE Proceeding, 2011, pp. 337.

[21] A.M. Gabor, J.R. Tuttle, S.A. David, M.A. Contreras, R. Noufi, A.M. Hermann,High-efficiency CuInxGa1#xSe2 solar cells made from (InxGa1#x)2Se3 precur-sor films, Appl. Phys. Lett. 65 (1994) 198–200.

[22] A.L. Patterson, The Scherrer formula for X-ray particle size determination,Phys. Rev. 56 (1939) 978–982.

[23] N.C. Halder, C.N.J. Wagner, Separation of particle size and lattice strain inintegral breadth measurements, Acta Crystallogr. 20 (1966) 312–313.

[24] I. Dirnstorfer, M. Wagner, D.M. Hofmann, M.D. Lampert, F. Karg, B.K. Meyer,Characterization of CuIn(Ga)Se2 Thin Films, Phys. Status Solidi (a) 168 (1998)163–175.

[25] S. Siebentritt, N. Rega, A. Zajogin, M.C. Lux-Steiner, Do we really need anotherPL study of CuInSe2? Phys. Status Solidi (c) 1 (2004) 2304–2310.

[26] S. Siebentritt, U. Rau, Wide-Gap Chalcopyrites, 1st ed., Springer, Heidelberg,Berlin, 2006.

[27] U. Rau, H.W. Schock, Electronic properties of Cu(In,Ga)Se2 heterojunction solarcells – recent achievements, current understanding, and future challenges,Appl. Phys. A – Mater. 69 (1999) 131–147.

[28] V. Nadenau, U. Rau, A. Jasenek, H.W. Schock, Electronic properties of CuGaSe2-based heterojunction solar cells. Part I. Transport analysis, J. Appl. Phys. 87(2000) 584.

[29] R. Scheer, H.-W. Schock, Chalcogenide Photovoltaics: Physics, Technologies,and Thin Film Devices, 1st ed., Wiley-VCH, Weinheim, 2011.

[30] P. Würfel, Physics of Solar Cells: From Principles to New Concepts, Wiley-VCH,Weinheim, 2007.

[31] L. Gütay, G. Bauer, Spectrally resolved photoluminescence studies on Cu(In,Ga)Se2 solar cells with lateral submicron resolution, Thin Solid Films 515 (2007)6212–6216.

[32] L.M. Fearheiley, The phase relations in the Cu, In, Se system and the growth ofCuInSe2 single crystals, Sol. Cells 16 (1986) 91–100.

[33] K.S. Knight, The crystal structures of CuInSe2 and CuInTe2, Mater. Res. Bull. 27(1992) 161–167.

[34] J.F. Guillemoles, A. Lusson, P. Cowache, S. Massaccesi, J. Vedel, D. Lincot,Recrystallization of electrodeposited copper indium diselenide thin films in anatmosphere of elemental selenium, Adv. Mater. 6 (1994) 376–379.

[35] O. Volobujeva, J. Kois, R. Traksmaa, K. Muska, S. Bereznev, M. Grossberg, et al.,Influence of annealing conditions on the structural quality of CuInSe2 thinfilms, Thin Solid Films 516 (2008) 7105–7109.

[36] J.F. Guillemoles, P. Cowache, A. Lusson, K. Fezzaa, F. Boisivon, J. Vedel, et al.,One step electrodeposition of CuInSe2: improved structural, electronic, andphotovoltaic properties by annealing under high selenium pressure, J. Appl.Phys. 79 (1996) 7293–7302.

[37] T. Roisnel, J. Rodríquez-Carvajal, WinPLOTR: a windows tool for powderdiffraction pattern analysis, Mater. Sci. Forum 378 (2001) 118–123.

[38] J.I. Pankove, Optical Processes in Semiconductors, 2nd ed., Dover Publications,New York, 2010.

[39] U. Rau, K. Taretto, S. Siebentritt, Grain boundaries in Cu(In,Ga)(Se,S)2 thin-filmsolar cells, Appl. Phys. A: Mater. Sci. Process. 96 (2009) 221–234.

[40] I.V. Bodnar, V.F. Gremenok, Y.A. Nikolaev, V.Y. Rud, Y.V. Rud, E. Terukov,Photosensitivity of thin-film structures based on CuIn3Se5 and CuIn5Se8ternary semiconductor compounds, Tech. Phys. Lett. 33 (2007) 111–113.

[41] M. Turcu, O. Pakma, U. Rau, Interdependence of absorber composition andrecombination mechanism in Cu(In,Ga)(Se,S)2 heterojunction solar cells, Appl.Phys. Lett. 80 (2002) 2598–2600.

[42] M. Turcu, Compositional trends of defect energies, band alignments, andrecombination mechanisms in the Cu(In,Ga)(Se,S)2 alloy system, Thin SolidFilms 431–432 (2003) 158–162.

[43] L. Gütay, D. Regesch, J.K. Larsen, Y. Aida, V. Depredurand, A. Redinger, et al.,Feedback mechanism for the stability of the band gap of CuInSe2, Phys. Rev. B86 (2012) 045216.

[44] S.S. Hegedus, W.N. Shafarman, Thin-film solar cells: device measurements andanalysis, Prog. Photovolt. 12 (2004) 155–176.

[45] S. Siebentritt, What limits the efficiency of chalcopyrite solar cells? Sol. EnergyMater. Sol. C 95 (2011) 1471–1476.

[46] P.D. Paulson, R.W. Birkmire, W.N. Shafarman, Optical characterization of CuIn[1#x]Ga[x]Se2 alloy thin films by spectroscopic ellipsometry, J. Appl. Phys. 94(2003) 879.

[47] J. Koo, S.C. Kim, H. Park, W.K. Kim, Cu(InGa)Se2 thin film photovoltaic absorberformation by rapid thermal annealing of binary stacked precursors, Thin SolidFilms 520 (2011) 1484–1488.

J. Fischer et al. / Solar Energy Materials & Solar Cells 126 (2014) 88–95 95