Passivating Surface Defects and Reducing Interface Recombination in CuInS2 Solar Cells by a Facile Solution Treatment

Interface recombination at the absorber buffer interface impedes the efficiency of a solar cell with an otherwise excellent absorber. The internal voltage or the quasi-Fermi level splitting (qFLs) measures the quality of the absorber. Interface recombination reduces the open circuit voltage (VOC) with respect to the qFLs. The present work explores a facile sulfur-based post-deposition treatment (S-PDT) to passivate the interface of CuInS2 thin films grown under Cu-rich conditions, which show excellent qFLs values, but much lower VOCs. The CuInS2 absorbers are treated in three different S-containing solutions at 80 oC. Absolute calibrated photoluminescence and current-voltage measurements demonstrate a reduction of the deficit between qFLs and VOC in the best S-PDT device by almost one third compared to the untreated device. Analysis of temperature dependence of the open-circuit voltage shows increased activation energy for the dominant recombination path, indicating less interface recombination. In addition, capacitance transient measurements reveal the presence of slow metastable defects in the untreated solar cell. The slow response is considerably reduced by the S-PDT, suggesting passivation of these slow metastable defects. The results demonstrate the effectiveness of solution based S-treatment in passivating defects, presenting a promising strategy to explore and reduce defect states near the interface of chalcogenide semiconductors.

untreated device. Analysis of temperature dependence of the open-circuit voltage shows increased activation energy for the dominant recombination path, indicating less interface recombination. In addition, capacitance transient measurements reveal the presence of slow metastable defects in the untreated solar cell. The slow response is considerably reduced by the S-PDT, suggesting passivation of these slow metastable defects. The results demonstrate the effectiveness of solution based S-treatment in passivating defects, presenting a promising strategy to explore and reduce defect states near the interface of chalcogenide semiconductors.

Introduction
The copper indium gallium disulfide Cu(In,Ga)S2 alloy system is a promising candidate for top cell in a thin film tandem solar cell. 1 So far, a stable efficiency of 15.5% has been achieved by growing absorber at a temperature above 550 o C. 2 CuInS2, the ternary compound allows to reduce additional effects due alloy disorder and band gap gradients introduced by addition of gallium. [3][4][5] The CuInS2 absorbers grown under Cu-excess conditions exhibit higher quasi-Fermi level splitting (qFLs) compared to the absorbers grown under Cu-deficient conditions. 6 The qFLs represents the open-circuit voltage the absorber itself can produce under illumination. This qFLs is still significantly lower (~700meV) than the bandgap (1.5eV) of the absorber, particularly due to the presence of deep defects. 6,7 Moreover, solar cells realized with Cu-rich ([Cu]/[In] at % >1) absorbers suffer from large open-circuit voltage (VOC) deficit compared to corresponding qFLs. Severe interface recombinations at the absorber/buffer (i.e. CuInS2/CdS) interface are the prominent cause for this deficit. 8,9 Interface recombination has been identified as a limiting factor in many thin film solar cells: perovskites, 10 all chalcopyrites grown under Cu-excess, 11 CdTe. 12 In that case the open circuit voltage (VOC) of the solar cell is lower than the qFLs of the absorber. Dominating interface recombination is caused by a cliff-type band offsets i.e. conduction band minimum (CBM) of absorber is higher than CBM of buffer or/and by a high density of defects at or near the interface. 13 In Cu-rich CuInS2 solar cells both factors can play a role: an unfavorable cliff conduction band offset between CuInS2/CdS and a large number of near surface defects in the absorber. 8,[14][15][16] The use of an appropriate buffer layer circumvents the problem of unfavorable conduction band offset at the absorber buffer interface. 17,18 However, even with a suitable band alignment, the Cu-rich sulfide devices are still dominated by interface recombination. 19 Therefore, a suitable technique to passivate the surface defects is needed.
Recent photoluminescence (PL) studies on CuInS2 and CuInSe2 demonstrate that the defect chemistry in both the systems is similar, establishing a close resemblance between the two systems. 1,20 Furthermore, it has been demonstrated that both selenides and sulfide chalcopyrite solar cells are dominated by interface recombination when Cu-rich absorbers are used. 9 In a recent study the dominating interface recombination was traced back to a defect present near the surface, which is related to a Se deficit. 21 This defect is caused by etching the Cu2-xSe secondary phase, which is always present in chalcopyrite grown under Cu-excess. 22 This defect is responsible for the VOC loss in Cu-rich solar cells. A similar defect is expected in the CuInS2 compound. This defect affects the device VOC in a similar way as its counterpart CuInSe2. Thus, the present work aims to find a treatment that can passivate this S deficit related defect.
In preliminary experiments, two devices were fabricated with a CdS buffer layer using low and high thiourea (CH4N2S) concentrations (i.e. the source of sulfur S 2ions in the chemical bath solution), where the low concentration is our standard CdS recipe. The higher thiourea concentration led to a device with higher VOC. Since the CdS buffer layer is known to have an unfavorable band alignment with CuInS2, 14,16 an additional device with Zn(O,S) buffer layer was fabricated for comparison. Details of the process for both buffer layer depositions can be found in the supplementary information. It is worth mentioning that the concentration of thiourea in the Zn(O,S) buffer layer recipe (0.4 M), is eight times more concentrated than the standard CdS buffer layer recipe (see supplementary information).
The current density-voltage (J-V) characteristics of the devices with different buffer layers and thiourea concentrations are shown in Figure S1 in the supplementary information, which show a clear improvement in device performance, especially the VOC, with the higher thiourea concentration in the chemical bath. This improvement suggests the effect of sulfur concentration on the VOC of the devices. It is therefore hypothesized here that a dedicated sulfur treatment for CuInS2 might be beneficial to reduce the interface recombination and improve device VOC.
This study reports a post-deposition sulfur-treatment (S-PDT) for Cu-rich CuInS2 absorbers.
For PDT first the secondary phase Cu2-XS are etched from Cu-rich absorber using 10% KCN for 5 minutes followed by the S-PDT, i.e. the immersion of the absorbers in either ammonium sulfide (AS) or sodium sulfide (NaS) or thiourea (TU). Some of the absorbers were again etched with 5% KCN solution for 30 seconds. Finally, the absorbers are covered with buffer (Zn(O,S)) and window (aluminum doped zinc oxide i.e. AZO). Figure 1 depicts the entire procedure. These solutions were chosen because they were used for surface passivation treatments on selenide absorbers in the past, [23][24][25] they are used in solution processing of solar cells as a part of buffer solutions, 17,26 and because all of them contain sulfur species. The treatment aims at passivating surface defects related to the sulfur vacancy at or near the interface. We will demonstrate that S-PDT improves the device VOC and FF, and reduces the interface recombination, as confirmed by temperature-dependent current density-voltage (JVT) analysis.

Quasi-Fermi level splitting measurements
QFLs is measured by absolute calibrated PL. 27,28 Figure 2(a) depicts the transformed PL spectra transformed using Planck's generalized law and the fit to extract the qFLs, measured under 5-sun illumination. 27 A bar chart of qFLs values for the untreated and S-PDT absorbers with and without a buffer layer is presented in Figure 2(b). The first observation on untreated absorbers is that the buffer reduces qFLs, i.e. increases non-radiative recombination. This has been observed for all types of buffers that were tested in our lab [CdS, Zn(O,S), ZnMgO, not shown here]. Since, contact is necessary to make the absorber into a solar cell, it is imperative to study and improve the qFLs in absorbers covered with a buffer. Obviously, the buffer on sulfide absorbers increases recombination. This observation is in contrast to selenide absorbers, where the buffer layer was observed to passivate the surface. 28  If we first compare the bare absorbers without buffers, no change in recombination activity is observed after AS-PDT, whereas, NaS-PDT and TU-PDT reduce the qFLs, i.e. increase nonradiative recombination in bare absorbers. The reduction in qFLs in case of NaS-PDT is significantly more than TU-PDT. This can be a result of mechanical degradation of the absorber as during the treatment partial flaking of absorber from the Molybdenum surface was observed. However, by comparing the qFLs of absorbers with and without buffer, it becomes obvious that the NaS-PDT and TU-PDT prevent the degradation due to the buffer within measurement error. And the highest qFLs with buffer is obtained for the TU treated absorber.
The difference in recombination activity could be due to an improved interface or due to improved grain boundaries. To investigate if the S-treatment has an influence on the recombination activity of grain-boundaries, the best treatment (TU-PDT) was explored by cathodoluminescence. However, no difference was observed between the cathodoluminescence of the untreated and the TU-PDT absorber ( Figure S3). Thus, we conclude that the main effect of the treatment is not a grain boundary passivation, but a passivation at or near the buffer/absorber interface. We investigate this further by electrical characterization of complete devices. shape' which results in a particularly low fill factor (FF). The presence of defects near the surface are the origin of this 'S shaped' J-V curve, as will be discussed in the next section.

Current-voltage characteristics
Compared to the untreated device, none of the S-PDT devices exhibit the 'S shape' in the J-V curves. Consequently, these devices exhibit higher FF and efficiency compared to untreated devices. The S-PDT devices also exhibit slightly improved short-circuit current density (Jsc) except for NaS-PDT device, which is also the one that was mechanically damaged. To better understand the short-circuit currents, we study external quantum efficiency (EQE) spectra for the S-PDT and the untreated devices [ Figure 2(d)]. All devices show a lower response in the long wavelengths region. It can be assumed that the space charge region (SCR) width is rather narrow, due to the high doping > 1e17 cm -3 observed in Cu-rich CuInS2 devices. 30 Obviously the diffusion length in these devices is not long enough to compensate for the narrow SCR. As a result there is an incomplete collection of the photons in the long-wavelength region. The lowest long wavelength response is observed in the NaS-treated and potentially mechanically damaged device. In contrast, AS-PDT and TU-PDT leads to a slight improvement in the longwavelength region of the EQE spectra, suggesting improved diffusion length or space charge region width after the treatment. Additionally, optical effects may also play a role, as seen by the shifts in the peak wavelengths of the interference maxima. This is most evident in the EQE spectrum of the TU-PDT device, which is most distinctive among all spectra. The interference pattern in this curve is shifted to lower wavelengths [see Figure 2 Among all the S-PDT devices, only the TU-PDT device shows an additional improvement in the VOC, consequently exhibiting the highest PCE of 8.5%. The VOC of the NaS-treated device is even lower than the one of the untreated device. In the diode model the VOC can be influenced by two main effects: the shunt resistance (Rsh) and the dark saturation current density (Jo) as follows: 31 00 *ln sc where q is the elementary charge, kBT/q is the thermal voltage, A is the diode factor and Jsc is the photo current density which is assumed voltage-independent. The devices presented here have rather low shunt resistances (see table 1), yet, the impact of Rsh on VOC is almost negligible (see discussion in SI). Thus, the differences in VOC are due to differences in nonradiative recombination. A comparison of Jo is not possible, because the fit of the J-V characteristics to the 1-diode model is problematic, as the J-V curve does not show ideal diodic curve (more details in SI). However, the reduction in non-radiative recombination is supported by comparing the qFLs values in table 1 and Figure 2(b): the NaS-treated sample has the lowest qFLs, and the TU treated one the highest, although not significantly higher.
Still, it can be concluded from the combined observation of the trends in VOC and qFLs: the non-radiative recombination in the TU-treated device in reduced.

Metastable behavior in the electrical measurements
In the previous section, an 'S shape' was observed in the J-V curve of the untreated device [ Figure 2(c)]. An 'S shape' in the J-V curve has been observed before in chalcopyrite solar cells, particularly at lower temperatures. [32][33][34][35] The presence of this 'S' shape is characteristic of a carrier transport barrier in the device and leads to attenuation of FF and VOC. 34,[36][37][38][39][40] In literature the presence of a highly defective layer (p + ) at the absorber surface is invoked as an explanation: a thin layer near the surface of the absorber which has a higher net-doping than the bulk. 35,41 The formation of this p + layer can be explained by the existence of the double vacancy defect (VCu and VSe) in CuInSe2. 42 A recent study on Cu-rich CuInSe2 by Elanzeery et al. supports the model of a p + layer related to Se vacancies. 21 We believe a similar defect (involving VS) might also be present in CuInS2 system resulting in 'S shaped' J-V curves.  parameters of the best device improve with LS except for VOC, which is reduced. The reduction in VOC is due to degradation of the device over time, something that is commonly observed in all the devices presented in this study. This degradation can be partially recovered with light soaking but not completely.  In order to remove the S shape in J-V curve of the untreated or the etched device [ Figure 4 and Figure S6], a considerable duration of light soaking is required, this implies the engagement of 'slow' defects. To explore the nature of the metastable defects in the devices, the time-evolution of the SCR width of the reference device and the best S-PDT device ( Figure 5(b) and (c)) was analyzed. The metastable behavior after the second KCN etch step can be seen in the SI ( Figure S7). The SCR width transient is measured by the inverse capacitance using the relation: where C(t) is the measured transient capacitance as follows: first, the sample is kept under illumination with a certain intensity for 300 seconds starting from t = -300 seconds. The 'w(t)' is the transient space charge region width of the device, ɛ is the relative dielectric permittivity of absorber taken equal to 10 as commonly used in literature, 51,52 and ɛo is the dielectric permittivity of free space. It must be noted that the measured SCR width includes contribution from both absorber and buffer. However, this fact can be ignored here as we only discuss slow metastable changes of the capacitance. Throughout this illumination period, a reverse external voltage bias is applied to make sure the device always stays in short-circuit conditions (detailed information in SI), as the internal resistance of the inductance, capacitance and resistance (LCR) meter together with the short-circuit current puts the device in forward bias state. During the illumination period, the traps are occupied with photogenerated carriers, a nearly constant capacitance towards the end of the period ensures a saturation state. The illumination intensity and the applied voltage is then set to zero at t = 0 sec and the capacitance transient is measured for at least 300 seconds more. This allows the device capacitance to reach a constant value (after de-trapping of carriers) indicating the device is in a new certain quasi-steady state.  Figure S7 of the supplementary information. The device capacitance is higher and thus the effective SCR width lower under illumination due to the additional contribution of light-generated charge carriers. We are mostly interested in the SCR width change between the illuminated state and the dark state at t = 0 sec. The device capacitance transient was measured with lowest illumination intensity first followed by higher illumination intensity. After each measurement, the device is in a quasi-steady state, which is different from the previous steady state. Bringing the device back to the completely relaxed state is extremely time-consuming. Therefore, all the curves have been shifted to the same SCR width value at t = 0 seconds to allow a better comparison, the unshifted curves can be found in SI. For the untreated device, at t = 0 seconds, the SCR width increases abruptly (VCu+VS) divacancy defect complex. 53 We hypothesize a similar mechanism also applies to Cu-rich CuInS2 absorbers.
In summary: the response of the untreated device is dominated by slow defects, this response increases with higher illumination intensity, whereas the treated device shows much less response of slow defects accompanied with a free carrier response. Thus, both J-V and capacitance transient measurements show the effectiveness of S-PDT, especially TU-PDT, in the passivation of near surface defects. In addition, it also shows that these slow defects have characteristics, which are usually associated with metastable defects.

Interface recombination analysis
The low VOC compared to the bandgap has been attributed to interface recombination in Curich chalcopyrite solar cell. 9,21,54,55 In addition, the large deficit between the quasi-Fermi level splitting and the VOC (see table 1) is a result of recombination at or near the interface. 55 where Rsh is ignored and Joo is a weakly temperature-dependent pre-factor and Ea is the activation energy of the dominant recombination process. From equation 4, a linear temperature-dependent VOC extrapolation to 0 K yields the Ea of the dominant recombination process, assuming the diode factor A and the Jsc to be constant (at least at moderately high temperatures i.e. 220-300 K).  Na2S solution contains Na + ions which we hypothesize that they can aggregate at grain boundaries or even go on to Cu sites in the absorber grains themselves since sodium is very mobile. 61 Both of these possibilities could have led to passivation of recombination centers and perhaps higher VOC in the device. Unfortunately though, this solution also led to delamination of the absorber from the Mo substrate during our experiments, and this accounts for the slightly degraded device performance compared to AS. The case for TU is more complicated. Previously it was believed that TU releases sulfur in basic conditions via HSto form S 2-. 62,63 However, a new study suggests that TU only releases sulfur when it is directly complexed to a metal cation. 64  To summarize, the different effects of the various PDTs can be due to different S-species in solution, which are expected to show different absorption behavior and reaction mechanism behavior. Further investigation of the PDTs is required to understand the reaction mechanism better. However, we would like to stress, that a simple treatment with an S containing solution does improve the interface of the solar cells.

Conclusion
Interface recombination is one of the main factors for the low efficiency of Cu-rich CuInS2.
Using a simple sulfur solution immersion technique we were able to show partial passivation of these defects by using thiourea as a sulfur source. We probed the optoelectrical properties of Cu-rich CuInS2 before and after the treatment. After buffer layer capping, the qFLs

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author. It worth mentioning that in CuInS2 devices with CdS buffer layer the 'S shape' is not observed. This is because the additional barrier (other than p + layer) due to a positive conduction band offset is not present in these device. 16,18 However, we do notice an 'S shape'

Influence of thiourea concentration in chemical bath and buffer layer deposition
in some of our Cu(In,Ga)S2 devices prepared with CdS buffer layer. This is a result of an increase in front Ga grading (i.e. increase in Ga atomic concentration towards the surface), which leads to an increase the conduction band towards the surface. Thus providing an additional barrier to minority carriers together with the p + layer, similar to a positive conduction band offset at the absorber buffer interface.

Calibrated photoluminescence measurement and quasi-Fermi level splitting determination
The calibrated photoluminescence measurements to extract the qFLs have been performed under an equivalent illumination of five suns to ease the spectra acquisitions, to allow for faster and more reliable measurement, in spite of the quite low radiative efficiency of these absorbers. Then, those values have been corrected for one sun illumination, as listed in table 1 in the main part of the manuscript. The correction is based under the assumption that the optical diode factor k is unity, which is defined as: k PL I   (5) with and being respectively photoluminescence intensity and excitation density.
Under this assumption, the external radiation efficiency (ERE) is constant as well, as it is defined as the ratio between the integrated PL photon flux density and the incident photon flux density. Because of the low luminescence efficiency, we could not determine k for these samples.
The qFLs is related to the generation under illumination (G) and recombination in thermal equilibrium (R0) by the following relationship 67 Experimentally we often find k>1, 55 i.e. the ERE is higher at higher excitation intensity. In this case the qFLs at 1 sun would be even smaller, since equation (8) overestimates the qFLs at 1 sun. But the trends we discuss between samples would remain the same. Thus, we consider a worst case scenario when it comes to determining the additional Voc loss due to interface recombination.

Experimental details of cathodoluminescence:
The scanning electron micrographs and cathodoluminescence (CL) hyperspectral images were recorded using a Zeiss Merlin scanning electron microscope equipped with a DELMIC CL system at 10 keV beam energy and at beam currents of 500-700 pA. Figure S3 shows the CL images obtained on the cross-section of untreated and TU-PDT sample. Both samples show rather low CL intensity. No difference in grain boundary activity was observed.

Figure S3. SEM images (a, c) and panchromatic CL images (b,d) acquired on cross-section specimens from
CuInS2 solar cells with (a,b) and without TU treatment (c,d).   where the Rsh is low, Voc/Rsh becomes non-negligible with respect to Jsc. The low Rsh provides an alternative path for the light generated current reducing the Voc of the solar cell. Even though in our devices the Rsh is low still the impact of it on Voc will be just a few mVs. From  (see table S2). This follows the qFLs values after buffer, which remains unchanged after both TU-PDT and NaS-PDT, whereas, decreases for AS-PDT (see Figure 2). Among the three treatments studied, solely TU-PDT results in an improvement in both: a higher Rsh and a lower Jo. Thus, improvement in both Rsh and Jo leads to the device with highest FF, Voc and efficiency.
We would like to point out that all these devices showed performance degradation with time.
This degradation can be partially recovered with light soaking but not completely.

Transient capacitance measurements
The procedure to measure the capacitance transients is as follows: first, the sample is kept under illumination with certain intensity for 300 seconds starting from t = -300 seconds. Since the LCR meter has an internal resistance of about 100 ohm, this resistance under illumination puts the device in a certain forward biased state. Hence to keep the device under short-circuit conditions a reverse bias voltage is applied to compensate for the photo-voltage from t = -300 seconds to 0 seconds i.e. for the whole illumination period. This was done by measuring the DC voltage generated across the device due illumination, using the LCR meter. Thereafter, a voltage exactly opposite to this measured voltage is applied when the device is under illumination to keep the device under short-circuit condition. During the entire measurement procedure the voltage is monitored to make sure the device is always under short-circuit conditions. After this first step of 300 seconds, the illumination intensity is then set to zero at t=0 sec and the capacitance transient is measured for at least 300 seconds more. Note, no bias was applied on the sample during this second step i.e. for t > 0 seconds. Figure S7 shows the evolution of space charge region width with time in three samples: untreated, TU-PDT and TU-PDT followed by KCN etching.

Surface analysis by X-ray photoelectron spectroscopy
To check for the chemical impact of S-PDT on the absorber surface, X-ray photoelectron spectroscopy (XPS) was performed on the samples. For this, four pieces of 10% KCN etched Cu-rich CuInS2 absorbers from the same absorber deposition run were used. First piece was left untreated (ref), the second was treated AS-PDT (sample '1'), third one with TU-PDT (sample '2') and the last one with NaS-PDT, which delaminated and was not analyzed. All the three remaining pieces were then transferred with water layer (comes from rinsing the absorber with DI water after KCN and the S-PDT) on top into the glove box to avoid air exposure. From the glove box they were then transferred via a N2 filled cell into the XPS chamber for analysis. The entire procedure was designed to ensure the minimum air exposure.
XPS experiments were carried out using a Kratos Axis Ultra DLD instrument equipped with a monochromatic Al K source (1486.6 eV) working at 150W. The base pressure during the analyses was better than 5.10 -9 mbar. The narrow scans, for elemental quantification and chemical states investigations, were recorded with an energy resolution of 0.6 eV. The samples were sputtered with monoatomic Ar + ions of low energy (500 V) to limit the preferential sputtering effects, for 180 s to remove the surface contaminants and for 1080 s to access the deeper composition. The data were processed with the CasaXPS software (v2. 3.22) and the curve fitting obtained with 70% Gaussian -30% Lorentzian lineshapes. Figure S8 (a) shows the S 2p bulk spectra of the above-mentioned three samples. For TU-PDT absorber, proper fitting of 'S' spectrum required fitting with two doublets. The two peaks results from the S 2p3/2 and S 2p1/2 spin-orbit split: the first doublet correspond to CuInS2, which is present in spectra of all samples, and an additional doublet with peaks at 164.2-165.1 eV to account for bump at higher binding energy that could be signature of C-S-C. [69][70][71] Simultaneously TU-PDT sample also has significant amount of N present on the surface which is absent in the other two samples. Analysis of N 1s bulk spectra shows peak at 399.3 eV, which corresponds to C-NH2. 72