Electronic and compositional properties of the rear‐side of stoichiometric CuInSe2 absorbers

In‐depth understanding and subsequent optimization of the contact layers in thin film solar cells are of high importance in order to reduce the amount of nonradiative recombination and thereby improve device performance. In this work, the buried MoSe2/CuInSe2 interface of stoichiometric absorbers is investigated with scanning tunneling spectroscopy and Kelvin probe force microscopy combined with compositional measurements acquired via photo‐electron spectroscopy after a mechanical lift‐off process. We find that the local density of states, as measured with scanning tunneling spectroscopy, is similar to the front‐side of ultra‐high vacuum annealed CISe absorbers. The grain boundaries exhibit a weak upward band bending, opposite to Cu‐poor CuGaSe2, and we measure an increased Cu accumulation at the rear CISe surface compared to the bulk composition and a non‐zero concentration of Cu on the Mo‐side. Grazing incidence X‐ray diffraction measurements corroborate that a small amount of a CuxSe secondary phase is present at the MoSe2/CuInSe2 interface in contrast to reports on Cu‐poor material. Our findings shed new light into the complex interface formation of CuInSe2‐based thin film solar cells grown under Cu‐rich conditions.

was the absence of the Ga back-contact grading. 10 The minority carrier lifetime improved after exfoliation of the absorber layer from the back-contact, which showed that recombination at the back is a major issue in nongraded CISe absorbers. Meanwhile, some laboratories showed that a Ga back-grading could be introduced in CIGSe in a way that the bandgap was still close to the pure CISe, which resolves the disadvantage for tandem applications, 11,12 albeit increasing the complexity of the process. In agreement with the previous discussion, the PCE increased, and meanwhile, above 19 -% efficient solar cells were reported with a bandgap close to 1.0 eV. 12 There are already quite a few important reports that focused on the buried MoSe 2 /CI(G)S(e) interface, which mostly relied on the fact that the absorber could be peeled off mechanically from the back-contact. 10,13,14 This made it possible to use surface sensitive techniques such as X-ray and UV photo-electron spectroscopy (XPS/UPS) to study compositional and electronic variations at that interface. It was shown that there is a Ga accumulation at the back-contact 15 and the CuInS 2 (CIS) back-surface was Cu-depleted despite Cu-rich growth conditions. 13 Transmission electron microscopy measurements (TEM) showed that there was some in-diffusion of Cu into the Mo 16 and atom probe tomography (APT) measurements showed an increased Cu and Ga concentration in the first 50 nm of the CIGSe absorber (measured from the back-surface). 17 Most of these measurements were performed on Cu-poor CIGSe (Cu/In<1). For the case of CuGaSe 2 , grown under Cu-excess, it was shown via TEM measurements that Cu x Se precipitations could also exist in the bulk of the absorber layer. 18 Already at that time, it was speculated that this may be the reason for the low efficiencies of Cu-rich absorber layers. However, until today, the general consensus in the community is that Cu-rich absorbers exhibit a secondary phase free bulk with Cu x Se on top, which can be etched away via a potassium cyanide etching step. 19 Much less is known about the lateral variations of the electrical properties on the nanometer scale of the back-surface. A detailed combined electron beam induced current (EBIC) and cathodoluminescence (CL) study 6 showed that, for a number of absorbers with different compositions, the grain boundary properties varied and there was no direct correlation between the CL signal and the collection efficiency, as measured via EBIC. For low symmetry grain boundaries, CL measurements showed a reduced signal, which pointed towards an increased number of defects at the grain boundaries. Furthermore, there are some Kelvin probe force microscopy (KPFM) measurements available 20 on CuGaSe 2 , which showed that most of the grain boundaries exhibited downward band bending in the order of approximately 50 meV compared to the grain surfaces. Furthermore, Ga remnants were found on the Mo-side. So far, no scanning probe microscopy measurements are available for CISe.
It is well known in the CI(G)Se community that the Cu-content is very important. 21,22 High performance CIGSe devices are grown under Cu-poor conditions (Cu/In<1), which improves V OC and quasi-Fermi level splitting compared to Cu/In>1. 22 For Ga-free CISe absorbers, the difference in quasi-Fermi level splitting between Cu-poor and Cu-rich absorbers is small. 23,24 However, in agreement with the Ga-containing absorbers, the PCE is higher for Cu-poor than for Cu-rich devices.
In the case of Cu-rich absorbers, the excess Cu condensates as Cu x Se on top of the CISe absorbers. A subsequent potassium cyanide (KCN) etching 13,25 removes the Cu x Se (or Cu x S in the case of CIS), thereby creating a stoichiometric absorber with less Cu-vacancies in the bulk. The KCN etching, however, also has some adverse effects on the composition and defects in the near surface region. 26,27 The investigation of the rear-surface thereby offers a very nice opportunity to investigate stoichiometric absorbers (grown Cu-rich) without KCN etching.
In this contribution, stoichiometric CISe absorbers (grown Cu-rich and subsequently KCN etched) were mechanically peeled off from their substrate and systematically scrutinized by means of scanning probe microscopy (SPM) techniques such as scanning tunneling microscopy (STM) and Spectroscopy (STS) and KPFM. The local density of states was evaluated by STS and compared to measurements performed at the top surface [27][28][29][30] whereas variations in workfunction of the films were probed by KPFM in order to compare the results to already published data on CuGaSe 2 . 20 Additionally, XPS and energy dispersive X-ray analysis (EDX) were used to explore the surface and bulk compositions and the impact of the peeling process on quasi-Fermi level splitting was investigated with photoluminescence (PL) imaging. Finally, X-ray diffraction measurements were used to investigate the occurrence of secondary phases in the CISe absorber layers.

EXPERIMENTAL DETAILS
SPM measurements were performed in a variable temperature ultra-high vacuum system with a base pressure in the 10   In order to examine the back-side of the absorbers, they were peeled off from their original substrates following the procedure schematized in Figure 1. to 1600 nm. The system was calibrated to absolute photon numbers in order to estimate the PL quantum efficiency and the quasi-Fermi level splitting of the absorber before and after the peeling process. which we observed in all of the images we analyzed. Furthermore, some deep holes (black contrast) were observed, which were likely to be caused by pinholes in the absorber layer prior or during the peeling process. In Figure 2B, the corresponding STM image of the Mo-side is presented, where no indications of grain boundaries were found. The surface was covered with small precipitates, which had an extension of approximately 70 nm.

The MoSe 2 /CISe interface
In order to corroborate that the absorber peeled at the MoSe 2 /CISe interface, XPS measurements were carried out on both samples (back-side and Mo-side) and the scans are presented in Figure 2C.
On the back-side, we did observe all the peaks of the CISe matrix, that is, Cu, In, and Se. Contrary, the Mo-side did only show peaks that are related to Mo and Se in the survey scan. At a later stage of the manuscript, we will discuss high resolution scans in more detail.
In contrast to the survey scan presented in Figure 2C, we found a very small amount of Cu on the Mo side but no traces of In. The XPS measurements corroborated that the absorber cleaved exactly at the MoSe 2 /CISe interface in accordance with previous reports 6,13,20 due to the formation of a MoSe 2 layer during synthesis, which is known to be a van der Waals solid where exfoliation is feasible.
The higher oxygen signal at the Mo-side was likely to be caused by the longer storage time of this sample compared to the back-side as discussed in the experimental part.
We also found an Ag peak in the XPS spectra, which we related to glue that was present at the edges of the exfoliated absorber due to the small size of these samples (see Figure 2F). Complementary EDX measurements showed that the Ag concentration in the bulk is lower than 1 at%, which is within the experimental error of the machine, considering the close distance of the In and Ag L-lines ΔE rad F refers to the maximum achievable V OC in the absence of nonradiative recombination. In our case, the bandgap of the absorber is 1.0 eV, which translated to ΔE rad F =749 meV. 37,38 The remaining constants in Equation (1) are the temperature T at which the sample was measured and the Boltzmann constant k B . From the measurement right after KCN etching, we deduced a ΔE F =471 meV (measured by illuminating the front side of the CISe), which is a typical value for Cu-rich CuInSe 2 absorbers. 23,24 After the gluing process and the subsequent liftoff, the non-cleaved part of the absorber did not change significantly and ΔE F = 465 meV was very close to the fresh case (still measuring the front-side of the CISe absorber). The result of the back-surface, which is presented in Figure 2F shows a somewhat reduced ΔE F of only 426 meV. This showed that the gluing and exfoliation procedure had a negative impact on the opto-electronic properties of the absorber layer. However, it is important to note here that the absorber was still intact and ΔE F = 426 meV was only 39 meV lower, which ruled out extensive metallic in-diffusion of Ag that would have increased nonradiative recombination massively. We also note that the roughness of the exposed back-contact was extremely low, in contrast to the front-surface, which may also have altered the reflection coefficient of the impinging laser light and thereby the reabsorption probability of the emitted photons, which also has an impact on PLQY. 39 From this analysis, we concluded that the peeling process of the back-surface was successful and that this procedure did the maps are relatively featureless, which is in contrast to previously reported top-view measurements on potassium cyanide etched or air exposed CISe absorber layers. 27,28,30 Especially, no distinct grain boundary contrast at U = 0 V could be observed, which indicated that the density of states at the grain boundaries was similar to the grain surfaces. The measurements showed some similarity to absorbers after UHV annealing, where a passivation of the defect states at the Fermi-level (U = 0 V) was observed. 27,29,30 Because the CITS maps at all specific voltages were featureless (no lateral variations), we extracted an average dI∕dU curve from the scanned region, shown in Figure 3A. The curve is shown in Figure 3E as a solid line together with a measurement on the Mo-side (see Figure 2B) and one that was measured after heat induced passivation on the front surface of a CISe absorber synthesized with the same process in the same physical vapor deposition system. 27 The first thing to note is that the UHV annealed sample and the measurement from the back-surface were very similar. The parts at positive voltages were almost identical whereas the curves at higher negative applied voltages were different. We relate this difference to a different densities of states of the tips, which were certainly different in the two measurements. Important however, was the excellent agreement of the valence band and conduction band onsets. The Fermi-level, which is located at U = 0 V was very close to mid-gap, in agreement with the UHV annealed case. 27,29 However, we also note that the valence band and conduction band edges were not sharp and there was substantial tailing into the bandgap region. A possible reason for this will be discussed later in the manuscript.
The measurements on the Mo-side, which mainly consisted of MoSe 2 were different and we did see a finite conductance at E F . Furthermore, the Fermi-level was closer to the conduction band, which points towards n-type doping in agreement with UPS measurements. 14 The finite conductance at E F was an indication that the Mo-side had a lot of defects and explained why we did not observe a well-defined semiconducting gap as we would expect for MoSe 2 .
The absence of a clear grain boundary contrast in STS on the back-side is somewhat surprising because it was shown via a combined electron beam induced current (EBIC) and cathodoluminescence (CL) study that carrier collection efficiency and the CL yield were different at distinct grain boundaries. Our STS measurements do not support this interpretation. However, we need to keep in mind that the information depth was very different for EBIC/CL (several hundred nanometers) and STS (<1 nm). Furthermore, the oxygen content may have been very different. In the present study, the cleaving was carried out in a glovebox preventing air exposure. This is usually not done for EBIC/CL measurements, which are much less surface sensitive. Another question to be discussed is the sensitivity limits of the STS measurements to detect changes in work function. This can conveniently be done using the fact that the tunneling conductance can be approximated by the following formula: 40 This equation assumes that the density of states of the tip T is constant for all applied voltages and that the voltage dependence of the tunneling coefficient T is small and can be neglected. The density of states of the sample is denoted with S . The tunneling coefficient Assuming that the density of states of the sample S is not changing if we change the work function of the sample, we can simulate the effect of a different work function via  In our case, the experimental dI∕dU curve was not symmetric, which led to a larger change at positive voltages than at negative ones.
Overall, we see that the curve at positive voltages is influenced mostly by the changes in work function. However, it is important to note that we do need a rather large variation in the work function in order to see an effect in the STS spectra. From the graph, we concluded that at least, 100 meV difference is needed to see a clear change in the experimental dI∕dU-curves.

KELVIN PROBE FORCE MICROSCOPY MEASUREMENTS
We therefore carried out additional KPFM measurements, which is a technique that is more sensitive to changes in the work function than STS. 33 The results are presented in Figure 5 where a topography image ( Figure 5A) and a CPD map ( Figure 5B) are depicted. The CPD map is related to the work function via Equation (5).
Consequently, the brighter regions in the CPD map corresponded to regions with higher work functions, assuming a constant work function of the tip. In the topography image, we again observed the grain boundaries as slightly lower regions, in agreement with the STM measurements. Furthermore, we also observed very small dots on the rear-surface. The contact potential difference map showed, in contrast to STS a slightly higher contrast at the grain boundaries. A line profile along such a grain boundary is shown in Figure 5C. The However, one important aspect still needed to be discussed in more detail. A careful inspection of the density of the small dots that cover the complete back-surface showed that, at the grain boundaries the density seemed to be slightly lower (see Figures 2A and 5A). We therefore needed to analyze if the changes of the dot density at the grain boundaries were responsible for the observed work function changes. At least the dots seemed to have an impact on the work function because we also measured variations in CPD on the grain surfaces. The line profiles in Figure 5C showed that the variations were approximately 20-30 mV in the area outside of the shaded region. In Figure 5C, we highlighted all the prominent minima in the topography with green dashed lines whereas the maxima are highlighted with black dotted lines. For both situations (maxima or minima) in the topography, we also measured CPD variations in both directions, that is, no direct correlation between the maxima and minima in topography and CPD. We concluded that the dots are not the main reason for the systematic changes in the work function that we observed at the grain boundaries. However, we realized from this analysis that the dots do have an impact on CPD albeit not always in the same direction.
In the following, we would like to discuss the EDX and XPS results in more detail in order to link them to the SPM measurements. Table 1   XPS measurements are much more surface sensitive (≈ 6-8 nm) than EDX measurements, which are more bulk sensitive (information depth ≈ 0.1-1 μm at 5-20 kV acceleration voltage). 41 Interestingly, the Cu concentration, as deduced via XPS was much higher close the interface with a Cu/In ratio of 1.7. This was a clear indication that excess Cu was present at the back-contact and further strengthens the EDX results. At the Mo-side we measured a small amount of Cu and no In, which corroborated that there was no CISe remainders on the Mo-side. The Mo/(Se+O) ratio is 0.6, which points towards a MoSe 2 layer, partially oxidized due to the rather long storage time of this substrate in the glovebox before introduction into the XPS system. The oxygen concentration was more than doubled compared to the back-side absorber, which was introduced directly after the exfoliation. However, residual oxidation of the substrate prior to the growth process cannot be excluded and may also have contributed to the higher oxygen content.

ELEMENTAL COMPOSITION AT THE REAR-SURFACE
We will now discuss the high resolution scans of the individual elemental transitions acquired via XPS. In Figure 6A for CuInSe 2 . 42 The Auger parameter, which is only sensitive to the chemical environment and not to changes in binding energy as a result of different band bending, was 1849.7 eV, which was also in good agreement with CISe. 42 The Cu concentration at the Mo-side was only 1 % (see Table 1). In Figure 6D, the Cu 2p spectra were normalized to the peak maxima in order to improve the visibility. Within the error of the measurement, we could not observe a significant shift in the binding energy and the Auger lines were too weak to estimate an Auger parameter. The In 3d lines are presented in Figure 6B. The binding energy was in accordance with the reported values, 42 and we could not observe Indium at the Mo-side. This showed that the detected Cu that we measured on the Mo side is not bound to CISe. The Se 3d lines are depicted in Figure 6C. A clear shift in the binding energy was visible when comparing the two substrates. On the backside the Se 3d 5/2 was located at 54.2 eV whereas on the Mo-side the value Note. EDX and XPS measurements were acquired on large areas (compared to the grain size) in order to get representative averages. EDX values in brackets were measured after KCN etching of the back side. All samples were etched on the front side prior to the peel off process. All measurements performed on the back side were carried out in one session to limit the error bar. The measurements on the front side were done with another SEM machine and different systematic errors may influence the absolute numbers. The measured values of the binding energies were in agreement with reported values. 42,43 The elemental composition of the MoSe 2 (Table 1) was not equal to 2 and the large amount of oxygen was likely to influence the MoSe 2 concentration we measured. For both surfaces, the Na concentration was on a similar level with approximately 0.5%.
The XPS and the EDX measurements strongly suggested that there was excess Cu at the back contact of the absorber layer. In order to corroborate the existence of a Cu x Se secondary phase, grazing incidence X-Ray diffraction was used. The results are presented in Figure 6E. It needs to be emphasized that this absorber layer was KCN etched before the peeling process in order to make sure that all the Cu x Se detected in the diffractogram originated from precipitates in the bulk or from the back contact. We observed a peak at 25.6 • and additional shoulders around the (312) CISe peak at 52.39 • , which were also reported to originate from Cu x Se. 19,44 Consequently, the XRD data confirmed that precipitations exist in Cu-rich absorbers, despite the fact that the front surface was etched with KCN prior to the peel off process. In Figure 6F Figure 6G is not indicative of an accumulation of material but rather due to a higher secondary electron yield at the grain boundaries, i.e. a different electron affinity or workfunction.
Currently, we assume that the etching rate at the grain boundaries was enhanced, suggesting that there was a larger amount of Cu x Se present compared to the absorber back-surface. We attribute the small grains in Figure 6F to Cu x Se precipitates, which were removed during the KCN etching process. EDX measurements performed on the same sample showed a reduction of the Cu/In ratio from 1.4 to 1.3 due to the removal of a Cu x Se secondary phase.

DISCUSSION AND CONCLUSIONS
The following list summarizes the key findings of the preceding sections.
• • The STS data were similar to the results obtained on UHV annealed CISe absorbers. However, there was still substantial tailing as we did see a gradual decrease of the dI∕dU curves into the bandgap region and the absence of sharp conduction and valence bands.
• The Fermi-level was mid-gap, similar to the absorbers surfaces after UHV annealing. 27,29 • For the Mo-side E F was closer to the conduction band and we did measure a finite conductance at E F , which pointed towards a high number of defects in the MoSe 2 or very high doping levels.
The most intriguing finding is certainly the high concentration of are also mandatory to reach the current record power conversion efficiencies.