Influence of Substrate Temperature during InxSy Sputtering on Cu(In,Ga)Se2/Buffer Interface Properties and Solar Cell Performance

Indium sulfide (InxSy)—besides CdS and Zn(O,S)—is already used as a buffer layer in chalcopyrite-type thin-film solar cells and modules. We discuss the influence of the substrate temperature during very fast magnetron sputtering of InxSy buffer layers on the interface formation and the performance of Cu(In,Ga)Se2 solar cells. The substrate temperature was increased from room temperature up to 240 °C, and the highest power conversion efficiencies were obtained at a temperature plateau around 200 °C, with the best values around 15.3%. Industrially relevant in-line co-evaporated polycrystalline Cu(In,Ga)Se2 absorber layers were used, which yield solar cell efficiencies of up to 17.1% in combination with a solution-grown CdS buffer. The chemical composition of the InxSy buffer as well as of the Cu(In,Ga)Se2/InxSy interface was analyzed by time-of-flight secondary ion mass spectrometry. Changes from homogenous and stoichiometric In2S3 layers deposited at RT to inhomogenous and more sulfur-rich and indium-deficient compositions for higher temperatures were observed. This finding is accompanied with a pronounced copper depletion at the Cu(In,Ga)Se2 absorber surface, and a sodium accumulation in the InxSy buffer and at the absorber/buffer interface. These last two features seem to be the origin for achieving the highest conversion efficiencies at substrate temperatures around 200 °C.


Introduction
Indium sulfide (In 2 S 3 ) is one of the materials, besides CdS and Zn(O,S), which is already used as a buffer layer in chalcopyrite-type solar modules [1,2]. Due to environmental concerns and the political decision to avoid Cd in solar modules, In 2 S 3 could be a candidate of choice to apply as a buffer layer for Cu(In,Ga)Se 2 (CIGS) thin-film solar cells instead of the established CdS. In x S y layers for the application as CIGS buffers can be grown by almost all available deposition techniques, including atomic layer deposition (ALD), thermal evaporation, chemical bath deposition (CBD), ion layer gas reaction (ILGAR), spray pyrolysis, and sputtering [3][4][5]. A value of 18.2% (with anti-reflective coating) was reported by Spiering et al. for a CIGS cell with thermal evaporated In x S y buffer [6], 17.9% (certified) was achieved with a 30 cm × 30 cm module with thermal evaporated In x S y :Na buffer by the company Avancis [7], 16.4% were published for a cell with an ALD In x S y buffer [4], 16.8% (in-house) and 16.1% (certified) were achieved by Saez-Araoz et al. with an ILGAR In x S y :Cl buffer [8], and 14.3% for a 225 cm 2 cell with a reactively sputtered In x (O,S) y buffer fabricated by the company Midsummer [2].
The sputtering approach can be easily implemented in an industrial environment, similar to the sputtering of the molybdenum back contact, high-resistive layers and transparent conductive oxides used as front contact. Furthermore, the method would be favorable compared to ALD, thermal evaporation, or growth by CBD, due to the very fast deposition times of approximately one minute or less; compared to ten or even more minutes to deposit a 20-60 nm thick In x S y buffer layer. This "dry" sputtering process could be a candidate for an application in a roll-to-roll coater with a flexible substrate or the implementation in an in-line deposition machine without breaking the vacuum after CIGS absorber deposition.
We present our development of very fast rf-magnetron sputtered In x S y buffer layers deposited at different substrate temperatures, show the efficiency results of In x S y -buffered cells with open-circuit voltage (V OC ) values close to reference cells with CBD-CdS buffer, and discuss the influence of substrate temperature during In x S y sputtering. In addition, a detailed materials characterization of our various In x S y buffer layers including chemical depth profiles for the analysis of the In x S y buffer and CIGS/In x S y interface reveal a sodium accumulation in the In x S y buffer and at the CIGS/buffer interface and a copper depletion of the CIGS surface with higher substrate temperatures.

Materials and Methods
Polycrystalline CIGS absorber layers with a thickness of 2.3 µm, a Ga/(Ga+In) ratio of 0.3 and a Cu/(Ga+In) of 0.8 were grown in an industrially relevant in-line multi-stage co-evaporation process on Mo-coated soda-lime glass [9]. No alkali post-deposition treatment was applied prior to the buffer deposition, which can significantly improve the device performance mainly due to an increased V OC .
In the next step, In x S y as the buffer layer was rf-magnetron sputtered in a VON ARDENNE (Dresden, Germany) high-vacuum sputtering system of type CS 730S from a commercially available In 2 S 3 target with a diameter of 20 cm in a pure argon atmosphere (for further details see [10]). The substrate temperature during sputtering of the In x S y layers was increased from room temperature (RT) up to 240 • C at low power densities of approximately 1 W/cm 2 compared to high power densities around 10 W/cm 2 typically used for ZnO:Al sputtering [11]. Thicknesses of In x S y buffer layers for solar cells were in the range of 30-60 nm for deposition times of 40-80 s, resulting in a sputter deposition rate of approximately 0.8 nm/s. In addition, around 500 nm thick In x S y layers were deposited on highly transparent suprasil quartz glass substrates for transmittance and gracing incident X-ray diffraction (GIXRD) measurements. The thicknesses of the thin In x S y layers were determined on CIGS/In x S y cross-sections with a scanning electron microscope and for the thick layers by evaluation of the transmittance data. The commonly used CdS buffer with approximately 50 nm thickness was grown by CBD with a thiourea-based process as reference material.
Subsequently, rf-sputtered undoped ZnO (i-ZnO) was applied as high-resistive layer and dc-sputtered ZnO:Al served as front contact with Ni/Al/Ni grid fingers deposited by e-beam evaporation on top. The total cell area was 0.5 cm 2 . Figure 1a illustrates the stacking sequence as used in this work with a sputtered In x S y buffer in combination with i-ZnO. Alternatively we tried different stacking sequences with rf-sputtered Zn 0.75 Mg 0.25 O as high-resistive layer instead of i-ZnO and without any high-resistive layer with ZnO:Al deposited directly on top of In x S y . Both approaches turned out fine but could not significantly improve the cell efficiencies compared to our standard stacking sequence with i-ZnO, which we used for the present work.
Current-voltage (I-V) curves were measured with a WACOM (Saitama, Japan) AM1.5G solar simulator with four-point geometry at standard testing conditions with a silicon reference solar cell for calibration. External quantum efficiency (EQE) measurements were performed with a BENTHAM (Berkshire, UK) PVE 300 setup with an additional light bias. For I-V and EQE measurements the CIGS cells were in the as-grown state (no further light-soaking or post-annealing procedures were applied). The transmittance of 500 nm thick In x S y layers on suprasil quartz glass substrates was measured with a Perkin Elmer (Waltham, MA, USA) UV/VIS/NIR Lambda 900 spectrometer and GIXRD patterns for structural characterization were obtained with a PANalytical (Almelo, The Netherlands) Empyrian diffractometer with Cu K α radiation with an incident angle of 0.5 • . We performed time-of-flight secondary ion mass spectroscopy (ToF-SIMS) depth profiles with a ToF 5 -SIMS instrument from IONTOF (Münster, Germany). A Bi+ liquid metal gun running at 30 keV generates the secondary ions, which are recorded by a time-of-flight detector. We choose an oxygen gas gun as sputtering source to detect sodium in a good sensibility. To acquire quantified depth profiles of sulfur, oxygen, and CIGS matrix elements, we selected a cesium metal gun. Both sputter guns worked at an energy of 2 keV. The size of the sputter area with 130 × 130 µm 2 was larger than the analysis area of 80 × 80 µm 2 to get rid of crater edge information. For the quantified measurement, we chose an analysis area of 80 × 80 µm 2 and a sputter area of 500 × 500 µm 2 to increase the number of data points. Figure 1 depicts the I-V and EQE curves of the best In x S y -and CdS-buffered CIGS solar cells fabricated within this experimental campaign. The corresponding solar cell parameters are listed in the inset of Figure 1b. With 15.3%, the efficiency of the In x S y -buffered cell is reduced compared to the reference cell with CdS buffer with 17.1% efficiency-mainly due to a lower fill factor (FF). Nevertheless, the V OC is only slightly lower with 661 mV compared to 680 mV of the reference cell. There is no gain in short-circuit current density (J SC ) for the In x S y -buffered cell as there is an absorption in the short wavelength region in the EQE (see Figure 1c). This feature is the result of a too-low bandgap energy and parasitic absorption due to the pronounced thickness of the In x S y layer. In addition, the EQE values of the In x S y -buffered cell are slightly below the CdS reference in the region between 500 and 1100 nm. The gradual decrease in EQE between 700 and 400 nm would tend to an indirect bandgap energy for In x S y . We observed no significant effect, neither positive nor negative, with cold-light soaking under AM1.5G conditions on solar cell parameters of CIGS solar cells with sputtered In x S y buffer. Figure 2 shows the dependency of solar cell parameters on substrate temperature during sputtering of the In x S y buffer layer. Relatively low efficiency values resulted for cells with In x S y sputtered at RT whereas there is an efficiency plateau for a substrate temperature around 200 • C. For significantly higher substrate temperatures such as 240 • C there is a decrease in the efficiency values again. The difference in efficiency is mainly affected by the V OC whereas FF and J SC have minor influence. The increase in the scattering of efficiency data for solar cells fabricated with In x S y sputtered at temperatures lower and higher than 220 • C is mainly driven by the FF.  Figure 3a depicts the GIXRD patterns obtained on 500 nm thick In x S y layers on glass substrates. Both layers deposited at RT and 220 • C are crystalline and exhibit the tetragonal β-In 2 S 3 structure [PDF no. 96-400-0814]. It should be noted that the reflexes of the cubic high-temperature structure [PDF no. 96-151-8188] overlap with most of the tetragonal reflexes, and Soni et al. reported recently the formation of cubic In 2 S 3 by sputtering [12]. Therefore, a coexistence of this cubic α-phase could not be totally excluded in our case and probably this phase could occur in case there are locally very high temperatures above 717 K [13]. Even the sputtering power (density) [14] and thickness of the In x S y layers [15] could influence the structural properties of the deposited In x S y films. The high-temperature modification γ-In 2 S 3 which typically forms for temperatures above 1049 K, as reported by Pistor et al. [13] is very unlikely to occur during our sputtering conditions. Figure 3b illustrates the difference in transmittance of In x S y films deposited at RT and 220 • C under otherwise identical conditions on suprasil quartz glass substrates. The evaluated thicknesses from these transmittance measurements are 490 and 480 nm for RT and 220 • C, respectively. Thus, there is only a slight reduction in the sputter deposition rate with temperature. The main finding is that the optical absorption edge slightly shifts to higher energies for the sample with the In x S y buffer deposited at 220 • C. According to a model with an indirect bandgap energy (E g ) for In 2 S 3 the values of E g = 2.02 eV (RT) and E g = 2.08 eV (220 • C), respectively, were extracted by extrapolation of the linear region in the sqrt(α E)-E-plot, as illustrated in Figure 3c. In case a model with a direct bandgap E g is used for the evaluation (not shown here), higher E g values would result but with the same trend-a slight increase in E g with elevated temperature. However, the corresponding (α E) 2 -E-plots show no distinct linear shape as expected for a semiconductor with a direct bandgap. Generally, an increase in bandgap energy might be expected as a result of the incorporation of oxygen in In 2 S 3 [16]. In our case, this oxygen could probably stem from oxygen/water residual from the sputtering system, which could be more easily incorporated in the growing In x S y film deposited at higher substrate temperatures. Nevertheless, we could not clearly detect a higher amount of oxygen inside the In x S y layer grown at higher temperatures as reported in Section 3.2.2. A more likely explanation could be changes in chemical composition, i.e., in the sulfur/indium ratio as also shown Section 3.2.2. As a conclusion, only slight differences in structural and optical material properties between both In 2 S 3 layers sputtered at RT and 220 • C were observed, which is an indication that pronounced changes in solar cell performance could be more affected by the interface formation between CIGS absorber and In x S y buffer layer.

Chemical Composition of In x S y /CIGS Interface Region
Depth profiles were performed by ToF-SIMS in order to get an idea what influences the efficiency and V OC and to reveal changes in the chemical composition of the In x S y buffer and CIGS/buffer interface due to the different substrate temperatures during In x S y sputtering.
The quantified ToF-SIMS depth profiles shown in Figure 4 highlight changes in the chemical composition of the In x S y buffer layer deposited on CIGS. The chemical composition of In x S y grown at RT is stoichiometric, i.e., the layer has a homogeneous In 2 S 3 composition over the whole 60 nm thickness; whereas for substrate temperatures higher than 200 • C, the In x S y composition becomes more sulfur-rich and indium-deficient, and the layers tend to be more inhomogenous. ToF-SIMS depth profiles obtained on samples sputtered at 160 and 180 • C (omitted here for clarity) reflect the same trend as the RT sample.
Since pure argon without any addition of oxygen was used as sputtering gas, the oxygen concentration in all layers is expectedly very low, i.e., the S/(S+O) ratio is close to 1, indicating that no oxygen is incorporated during the sputtering process even at the higher substrate temperatures of 220 and 240 • C. Small amounts of oxygen are only detected at the In x S y surface.
In addition to changes in the composition of the In x S y buffer, there is a pronounced copper depletion of the CIGS surface, which is more obvious for layers deposited at higher substrate temperatures of 220 and 240 • C compared to RT, 160, 180, and 200 • C. In a more detailed consideration, the copper depletion at the CIGS surface was estimated to rise from around 10 nm at RT, up to approximately 70 nm at 240 • C. ToF-SIMS depth profiles were also performed on complete CIGS solar cells from ZnO:Al front contact down to the soda-lime glass substrate to reveal differences in chemical composition between CIGS solar cells with efficiencies of 7.1% and 14.6% with In x S y buffer layers sputtered at RT and at 220 • C, respectively ( Figure 5). The depth profile for the sample fabricated at higher substrate temperature exhibits a strong increase in the sodium signal inside the In x S y buffer layer and at the CIGS/buffer interface (note the logarithmic scale here). The sodium is originally diffusing from the soda-lime glass substrate through the molybdenum back contact into the CIGS layer at elevated deposition temperatures during the CIGS co-evaporation process but was not intentionally incorporated into the CIGS absorber or In x S y buffer layer. The observed sodium accumulation combined with a moderate copper depletion at the CIGS/In x S y interface region could be the most significant effect for an efficiency improvement of In 2 S 3 buffered CIGS-based solar cells.

Discussion
We have demonstrated that we could achieve suitable film quality for In x S y buffer layers with the very fast deposition method of rf-magnetron sputtering applied to CIGS thin-film solar cells to realize efficiencies above 15%. For our sputtered buffer layers, we applied high sputtering deposition rates around 0.8 nm/s; compared to slower deposition processes such as ALD, CBD, ILGAR, and thermal evaporation. Nevertheless, there is still an efficiency gap compared to the CdS-buffered reference cells, which was also observed in our previous works with sputtered In x S y buffers [10,17] and reported by other groups for CIGS cells with sputtered In x S y buffers [12,18,19], evaporated In x S y buffers [20,21], and our own cells with In x S y layers grown by metal-organic chemical vapor deposition [22].
For both 500 nm thick In x S y layers deposited at RT and 220 • C substrate temperature on glass substrates, we obtained crystalline structures without any huge difference between both layers. We indicated the tetragonal β-phase for In 2 S 3 . This result is in accordance with structural investigations on rf-sputtered In x S y layers reported by other groups [14,15,23], but also for structural investigations on evaporated In x S y layers [20,24]. In contrast, high-resolution transmission electron microscopy analyses revealed nanocrystalline structures for our oxygen-containing CBD-In x (O,S) y [25] and sputtered In x (O,S) y buffer layers deposited from an In 2 (O 0.25 S 0.75 ) 3 target [26] at the CIGS/buffer interface in the complete device.
The extracted bandgap energy (indirect) for our In x S y is in the range of 2.0-2.1 eV, thus not perfectly ideal for an application as a buffer in CIGS thin-film solar cells. With the addition of dopants like chlorine [8], oxygen [16], or sodium [1], the bandgap energy can be increased to suitable values to reduce parasitic absorption of the buffer and obtain high J SC . In addition, a beneficial band alignment with chalcopyrite-based absorber is necessary to obtain high V OC [27] and close the efficiency gap to the CdS-buffered reference cells. In this sense, the surface termination of the CIGS absorber could play an important role. The parasitic absorption can be also reduced by using thinner In x S y buffers with thicknesses below 20 nm. In this case the subsequent high-resistive layer becomes more important in terms of band alignment. It should be noted that the power density during sputtering could influence the bandgap value of In x S y layers as well [14]. Our extracted bandgap energy increases slightly around 0.06 eV with the substrate temperature during In x S y sputtering. This trend with increasing temperature was also observed by other groups for sputtered In x S y layers [17,18,28]. In contrast, Karthikeyan et al. reported an opposite trend as found for their In x S y deposited by pulsed direct magnetron sputtering [23]. From theoretical density calculations there is only a very small difference between direct and indirect transitions in β-In 2 S 3 [29,30].
The substrate temperature during In x S y sputtering has a major influence on the performance of the device. Solar cells with In x S y sputtered at RT result in poor efficiencies whereas a maximum seems to be around 200 • C. For higher temperatures >240 • C, there is a decrease in efficiency and V OC again. The strong dependency of solar cell efficiency on substrate temperature during sputtering and sample pre-conditioning was also observed in previous experiments at ZSW [10,17] and by other groups [31]. For higher substrate temperatures above 220 • C there seems to be a destruction of the important pn-junction at the CIGS/buffer interface resulting in deteriorated solar cell parameters.
In addition, the substrate temperature during In x S y sputtering has an influence on the chemical composition of the In x S y buffer and CIGS/buffer interface. In this work, we observed an increasing sulfur/indium ratio starting from a value of 3:2 (In 2 S 3 ) grown at RT with increasing substrate temperature during sputtering, similar to the trend in our previous experiments, but with a different starting ratio of around 1:1 [17], and similar to results obtained by D. Abou-Ras et al. [31].
The sodium accumulation in the In x S y buffer and at the CIGS/buffer interface and the copper-depleted CIGS surface seem to be the most likely candidates to increase the V OC of the 220 • C cell compared to the RT cell. Sodium addition to CIGS in general increases the V OC , e.g., in case a sodium-free substrate is used and sodium is intentionally incorporated with a precursor [32] or post-deposition process [33]. Sodium accumulation in the buffer and at the CIGS/buffer interface was also found after post-annealing processes for CBD Zn(O,S)-buffered devices [34]. A sodium accumulation on the nano scale was observed by atom probe tomography measurements at the CIGS/In x S y interface for samples annealed at temperatures above 225 • C compared to the RT samples without sodium accumulation [12]. The observed copper depletion at the CIGS surface could result in an ordered vacancy compound with a slightly higher bandgap due to a lowered valence band maximum. This feature could promote an improved band alignment in combination with the In x S y buffer and could be also a reason for improved V OC values as reported for evaporated In x S y :Na buffer on Cu(In,Ga)(S,Se) 2 [27]. The formation of such a Cu-poor CuIn 5 S 8 composition at the CIGS/sputtered In x S y interface was observed at relatively high temperatures around 340 • C [31]. In general, the integral copper content of the CIGS absorber can significantly influence the efficiency and the dominating recombination mechanism of In x S y -buffered devices [35].
The phenomena of copper diffusion at the CIGS/In 2 S 3 interface, when investigated, was mostly isolated from alkali elements in the past. Pistor et al. clearly showed with high kinetic energy X-ray photoelectron spectroscopy that above 200 • C, the copper diffusion from CIGS into In 2 S 3 is enhanced [36]. Recent calculations with density functional theory also took sodium into account and revealed an interplay between sodium and copper diffusion at the CIGS/In 2 S 3 interface, and the formation of Na-and Cu-containing phases is very likely due to strong negative driving force energies (energy releasing processes). The even lower driving force energy value for sodium implies that sodium diffusion is more favored than copper diffusion and would therefore be actuated earlier [37]. This difference could be one explanation for our observed results from the depth profiles: First, there is a sodium diffusion up to 200 • C with only small changes in copper. For higher temperatures >220 • C the copper diffusion is no longer hindered by the sodium and copper diffusion takes place. Following this argumentation, the important parameter for an increased efficiency is the sodium accumulation at the CIGS/buffer interface and a moderate copper depletion at the CIGS surface. An enhanced copper enrichment of the In 2 S 3 layer would be detrimental. It should be noted that, besides the copper depletion of the CIGS surface, an intermixing between In x S y buffer and CIGS absorber could also take place. A possible sulfur incorporation into the CIGS absorber, which we did not observe clearly in the ToF-SIMS profiles, would lead to a lowered valence band maximum and a different band alignment to In x S y .
Future research directions could address one or several of the following approaches: Sodium can be intentionally inserted in the In 2 S 3 target to increase the bandgap energy, similar to the sodium-containing In x S y buffer used by the company Avancis deposited by thermal evaporation [7]. In addition, post-annealing procedures or heat-light soaking could be an adequate approach to diffuse sodium into the In x S y buffer and to the CIGS/buffer interface, which was already successful applied to CIGS cells with CBD-Zn(O,S)/(Zn,Mg)O buffer system. Oxygen can be intentionally introduced via reactive sputtering with oxygen in the sputtering gas [2], via variation of the base pressure with water vapor as oxygen source [18], the use of mixed In 2 (O,S) 3 targets, or co-sputtering from pure In 2 S 3 and In 2 O 3 targets. Another approach could be the use of a very thin In x S y layer with a thickness of 20 nm or even less. In this case the subsequent high-resistive layer should be optimized, e.g., (Zn,Mg)O or ZnTiO x [38] could be candidates to replace i-ZnO. Furthermore, different window layers-such as InZnO x or InSnO x instead of ZnO:Al-can influence the device performance of CIGS solar cells in combination with sputtered In x (O,S) y buffers [2].

Conclusions
The substrate temperature during the sputtering of In x S y buffer layers significantly influences the performance of CIGS thin-film solar cells. The best power conversion efficiencies in the as-grown state were obtained at a temperature plateau around 200 • C with the best value of 15.3% compared to the reference cell with solution-grown CdS buffer with 17.1% efficiency. The difference in performance compared to cells with buffers deposited below 200 • C is mainly due to a reduced V OC . The chemical composition and homogeneity of In x S y changes from stoichiometric and homogenous In 2 S 3 for layers deposited RT to inhomogenous and more sulfur-rich and indium-deficient composition for higher temperatures in the range of 220-240 • C. In addition, there is a pronounced copper depletion at the CIGS absorber surface and a sodium accumulation in the In x S y buffer and at the CIGS/buffer interface for higher temperatures compared to In x S y layers deposited at RT. These two features could explain the higher V OC values of cells with In x S y buffers grown at higher substrate temperatures compared to RT. However, In x S y layers with a non-ideal bandgap energy of 2.0-2.1 eV (indirect) induce parasitic losses and a limited J SC . In future research, appropriate quantities of sodium and/or oxygen could be intentionally incorporated in the In x S y layer and-with very thin In x S y layers, in combination with an optimized high-resistive layer-the parasitic absorption could be further reduced and the J SC increased.