Chemical etching of Sb2Se3 solar cells: surface chemistry and back contact behaviour

The effect of (NH4)2S and CS2 chemical etches on surface chemistry and contacting in Sb2Se3 solar cells was investigated via a combination of x-ray photoemission spectroscopy (XPS) and photovoltaic device analysis. Thin film solar cells were produced in superstrate configuration with an absorber layer deposited by close space sublimation. Devices of up to 5.7% efficiency were compared via current–voltage measurements (J–V) and temperature-dependent current–voltage (J–V–T) analysis. XPS analysis demonstrated that both etching processes were successful in removing Sb2O3 contamination, while there was no decrease in free elemental selenium content by either etch, in contrast to prior work. Using J–V–T analysis the removal of Sb2O3 at the back surface in etched samples was found to improve contacting by reducing the potential barrier at the back contact from 0.43 eV to 0.26 eV and lowering the series resistance. However, J–V data showed that due to the decrease in shunt resistance and short-circuit current as a result of etching, the devices show a lower efficiency following both etches, despite a lowering of the series resistance. Further optimisation of the etching process yielded an improved efficiency of 6.6%. This work elucidates the role of surface treatments in Sb2Se3 devices and resolves inconsistencies in previously published works.


Introduction
Antimony selenide (Sb 2 Se 3 ) is a rapidly developing inorganic material which shows great promise within the field of thin film photovoltaics (PV); it is a stable, binary V-VI compound with a direct band gap of 1.18 eV [1,2] -well suited for use in PV. Its absorption coefficient is >10 5 cm −1 and the constituent materials are cheap and abundant [1] with low toxicity [3]. Sb 2 Se 3 adopts a 1D nanoribbon structure [4] with strong, covalently bonded chains linked by weaker van der Waals interactions. Sb 2 Se 3 solar cells are presently an emerging technology with rising efficiencies; the record power conversion efficiency for Sb 2 Se 3 devices currently stands at 9.2% [5] and this has improved rapidly from only 2% in 2014 [3,6]. Impressive device efficiencies have been achieved with a range of device structures and both with and without a variety of interfacial layers. [7][8][9][10] Previous work has shown that device performance depends strongly on the orientation of the nanoribbons in the film and, thus far, a large fraction of work on this material has focused on optimising growth conditions to achieve optimal orientation [4][5][6][10][11][12]. As a result there are many aspects of material and device behaviour that are still poorly understood, such as front and back contacting. Sb 2 O 3 and free elemental selenium contaminants have been identified at the surface of Sb 2 Se 3 films [13,14], however the impact of this on device behaviour has not yet been properly investigated.
It is well established in other technologies that surface modification prior to back contacting can lead to performance improvement, such as via the nitric-phosphoric acid, [15][16][17][18] bromine methanol [19], methyl ammonium iodide [20] or hydrogen iodide [21] etching step in CdTe solar cells. These etching steps often improve device characteristics through the creation of a Te-rich layer prior to back contacting. It is likewise important to investigate whether an equivalent approach for Sb 2 Se 3 can yield similar benefits. Understanding the chemical composition of the free back surface is a key component in developing such a process. In 2017, Wang et al stated that the removal of free elemental selenium using a carbon disulphide (CS 2 ) chemical etchant was successful in increasing the fill factor (FF) of their Sb 2 Se 3 devices [13]. X-ray photoemission spectroscopy (XPS) analysis showed reduction in Se levels at the surface following the etch treatment. Similarly, Chen et al reported in 2017 that the use of an ammonium sulphide ((NH 4 ) 2 S) chemical etch was effective in removing both Sb 2 O 3 and Se contamination (also supported by XPS analysis) and that this led to an increase in FF [14]. The XPS interpretation, however, was not consistent across the two studies (despite coming from the same group), there being a large discrepancy in the assigned positions of the chemically shifted contaminant peaks (refer to section 4 for details). Additionally, no direct comparison between the etching processes was made as the two studies utilised different Sb 2 Se 3 device structures. The exact mechanism of the etching processes are unclear-there have been reports on the use of (NH 4 ) 2 S to dissolve Sb-chalcogenides [22] however, to the best of our knowledge, there are no reports on the mechanism of CS 2 as an etchant for Sb 2 Se 3 . In this work, therefore, we have performed rigorous XPS analysis to accurately determine the effect of the two etches on the Sb 2 Se 3 surface and directly link this to changes in the back contact behaviour and device electrical parameters. Importantly, all samples, cells and surfaces were kept consistent throughout the study, with the only variable being the chemical etching process. It is noted that surface passivation layers have been used in the past to improve the performance of Sb 2 Se 3 solar cells [10], however these were omitted from this study in order to clarify the role of the etches. This allows us to properly assess the impact and effectiveness of surface treatments for Sb 2 Se 3 back contacting for the first time. Sb 2 Se 3 samples were deposited by close space sublimation (CSS)-a highly tunable, scalable synthesis method [28]. Sb 2 Se 3 solar cells made by CSS have so far achieved efficiencies competitive with the record efficiencies achieved using vapour transport deposition (VTD) and rapid thermal evaporation (RTE) (as shown in table 1, the use of CSS and the very similar technique of VTD have aided the progression of Sb 2 Se 3 to higher efficiencies). Samples were left as an unetched control sample, or were treated with either a CS 2 or an (NH 4 ) 2 S chemical etch to modify the surface chemistry prior to back contacting [13,14,29] (figure 1). Film morphology and crystallinity were investigated by scanning electron microscopy (SEM) and x-ray diffraction (XRD), while the surface chemistry of the back surface was determined using XPS measurements. Devices were produced in the superstrate configuration with a device structure consisting of glass, F:SnO 2 , TiO 2 , Sb 2 Se 3 and Au. Device performance was tested under AM1.5 illumination in order to correlate with changes in surface chemistry. In addition, temperature-dependent current-voltage (J-V-T) analysis was carried out to measure the change in the back contact barrier height, f b [30,31], of the devices with chemical etching.

Methods
2.1. Film, surface and device preparation TiO 2 films were deposited by spin casting titanium isopropoxide onto fluorine-doped tin dioxide coated glass substrates (TEC10 FTO, NSG Group) prior to annealing in air, via an established two-step process [32]. Sb 2 Se 3 was then deposited via a multi-step CSS process [10] utilising a seed layer deposited at a pressure of 10 −2 mbar, followed by a thicker layer with a source temperature of 470°C for 15 min at a pressure of 13 mbar N 2 to give a total layer thickness of ∼2 μm. This process was developed to generate layers with a more compact grain structure. In addition to an unetched control sample, two surface treatments were used, these being a CS 2 etch and an (NH 4 ) 2 S etch. For the CS 2 etch, a 25 mm×25 mm sample was rinsed with 3 ml of CS 2 as carried out by Wang et al [13]. For the (NH 4 ) 2 S-etched sample, 3ml of 0.3 mmol dm −3 (NH 4 ) 2 S (aqueous), and subsequently 3 ml of H 2 O, was applied to the substrate using a spin coater at 2500 rpm, as described by Chen et al [14]. For completion of devices, 100 nm thick Au back contacts were then thermally evaporated through a shadow mask, creating N =16 cells per sample type, each having an active area of 0.1 cm 2 . Figure 1 shows a schematic diagram of the device structure.

Materials and device characterisation
SEM images were taken using a JEOL 7001F with an acceleration voltage of 5 kV. XRD was performed with a Rigaku SmartLab x-ray diffractometer in a parallel beam configuration. XPS measurements were performed inside an ultra-high vacuum chamber operating at a pressure less than 4×10 −9 mbar. Samples were secured to a sample plate using double sided carbon tape and a narrow strip of tantalum connecting the surface to the plate, to prevent charging. The surface composition was probed using core-level-XPS with a Mg Kα x-ray source (hν=1 253.6 eV) operating at 200 W and a hemispherical Scienta SES200 electron energy analyser comprised of a double channel plate and phosphor screen with a CCD camera. The resolution was determined from the full width at half maximum of a Ag 3d 5/2 peak of a Ag calibration sample to be 0.7 eV, allowing binding energy determination with a precision of ±0.1 eV. Device performance was measured under simulated AM1.5 illumination at 1000 W m −2 using a TS Space Systems solar simulator and a Keithley 2400 source-meter. The Au/Sb 2 Se 3 back-contact barrier height, f b , was determined from dark current-voltage measurements (J-V) measurements as a function of temperature (J-V-T) using a CTI-cryogenics cryostat in the range 250-350K, with J-V curves taken using a Keithley 2400 sourcemeter over the range −1 to 1 V. Using the method proposed by Bätzner et al [31], the series resistance, R s , is determined at forward bias above open-circuit voltage (V oc ) as a function of temperature i.e. the slope method. R s (T) may then be separated into an invariant, an Ohmic and an exponential component, the latter resulting from the passage of the carriers over the back contact via thermionic emission. R s is thereby expressed as follows: where C is a fitting parameter, R Ω0 is the Ohmic resistance, and k is the Boltzmann constant. As shown by Al Turkestani [30], at low temperature the exponential term is much greater than the first two terms Hence, for this experiment, the third term is deemed sufficient and the expression is simplified to:   301) and (002) planes respectively (using the Pbnm space-group labelling convention [11]). These planes are representative of nanoribbons lying at 37°, 44°, 46°and 0°from normal to the substrate for the (211), (221), (301) and (002) planes respectively. Growth of nanoribbons normal to the substrate surface would be expected to optimise both vertical conduction along the nanoribbon axes and also increase the fraction of van der Waalsbonded rather than covalent-bonded grain boundaries, both of which are expected to be beneficial to photovoltaic performance [2,4]. Even the ∼45°angle of the (221) and (301) planes from normal is not expected to impede carrier transport due to the size of the grains in this material. It is reasonable to assume that the majority of the grains, and therefore the nanoribbons, span from the bottom to the top of the film [2,11]. There is a small difference in the peak at 26.6°, however this is attributed to an additional Sb 2 Se 3 (021) orientation [2] and is therefore not indicative of any significant change at the surface.
Overall the SEM and XRD studies revealed that the etching processes introduced no significant morphological or crystallographic change that was detectable using these methods. Therefore, the surface sensitivity of XPS was required to detect the important chemical changes. For the Sb 4d region, three chemical species were identified: Sb 2 Se 3 , Sb 2 O 3 and metallic Sb [33][34][35][36]. The oxide component was determined to be Sb 2 O 3 and not Sb 2 O 5 or a mix of the two species due to a clear difference in chemical shift (∼1 eV) for the two species relative to Sb 2 Se 3 [34,35]. For the Se 3d region, two chemical species were identified: Sb 2 Se 3 and free elemental Se [13,34,37,38]. The binding energies of the components are displayed in table 2. As only relative quantities can be determined from peak areas in XPS analysis, all contamination levels are presented as percentages of the Se or Sb present at the surface in each of their chemical  environments (table S1 and figure 6). For example, the amount of free elemental selenium present is characterised as the amount of free elemental selenium as a percentage of total selenium detected in all its environments.

Surface chemistry
In figure 4, although the magnitude of the signal is understandably dwarfed by the Sb 2 Se 3 signal, it is noticeable that the Sb 2 O 3 peaks are reduced in size after etching with both CS 2 and (NH 4 ) 2 S, indicating removal of the oxide via etching. However, it is also noted that here the level of contamination is already very low in the unetched sample compared with the other unetched samples previously reported by Wang et al and Chen et al [13,14]. The CSS deposition technique used in this work uses an inert nitrogen atmosphere and samples were exposed to air for minimal time (<30 min) between deposition and XPS measurement. A low level of contamination is therefore consistent with our expectations.
The levels of Se and Sb 2 O 3 contamination are summarized in table S1 (available online at stacks.iop.org/ JPENERGY/1/045001/mmedia) and figure 6. It is clear that both etching processes reduce the level of Sb 2 O 3   contamination, with the (NH 4 ) 2 S etch being marginally more effective. However, while the CS 2 etch has little or no effect on the amount of free elemental selenium at the back contact surface, the (NH 4 ) 2 S etch increases the proportion of free elemental selenium at the surface. This indicates that the etches do not selectively remove selenium and in the case of the (NH 4 ) 2 S etch it in fact produces a more selenium rich surface. In this work, by ensuring consistency in the fitting procedure used to analyse the effect of both etches, we show beyond doubt that the primary result of the etch process is Sb 2 O 3 removal and not Se removal. This is contrary to previous work, in which inconsistencies in the XPS fitting procedure led to misidentification of Se removal as the primary result of the CS 2 etching process [13,14] (as discussed in section 4). The next step is then to determine what impact this has on the device contacting and performance.

Device performance and contact barrier height determination
A series of complete cells were produced with absorber layers and etching steps identical to those used for XPS analysis to allow direct comparison. Average and peak device performance parameters extracted from J-V analysis are shown in tables 3 and 4 for an unetched sample and samples etched by CS 2 and (NH 4 ) 2 S. Here it can be seen that chemical etching has a negative effect on both the average and peak performances of the device. The main differences occur in V oc and J sc , with the (NH 4 ) 2 S etch being especially harmful to the J sc . In the peak performing cells ( figure 7), both etches see a reduction in R s at high forward bias, with the reduction being greater in the (NH 4 ) 2 S-etched device. This correlates well with the removal of oxide seen in the XPS spectra. There is also a reduction in shunt resistance (R sh ) following both etches which suggests some damage to the film by the etches leading to and increase in shunting pathways. This is supported by the loss in J sc caused by both etches, in particular the (NH 4 ) 2 S etch which sees the greatest reduction in R sh .   J-V-T analysis was carried out on all CSS deposited Sb 2 Se 3 devices in order to determine the magnitude of f b . R s values were measured in the dark via the slope method for a range of temperatures (250-350K) for all devices prior to fitting them using equation (2). The fits can be found in figure 8 and the results of the fitting are shown in table 5. The results show that the barrier height is lowered by both etches, with a similar decrease for both the (NH 4 ) 2 S etch and the CS 2 etch (albeit slightly larger for the CS 2 ). This correlates well with the XPS results and the device results, indicating that a removal of the oxide layer is beneficial to the back contact performance [13,14].

Discussion
The impact of (NH 4 ) 2 S etch and CS 2 etches on Sb 2 Se 3 identified here is notably different from prior reports [13,14]. Wang et al [13] suggested CS 2 improved FF by dramatically reducing the proportion of free elemental selenium at the back contact. However, in this work we observe a reduction in Sb 2 O 3 and not elemental selenium to be the primary role of CS 2 etching. This discrepancy in the effect on selenium can be explained, in part, by the   much lower selenium contamination levels in our devices (figure 5) due to use of a CSS process rather than VTD (as used by Wang et al and Chen et al) and may be the reason why they see a greater influence on device performance [13,14]. Chen et al reported that the (NH 4 ) 2 S etch effectively removes both Sb 2 O 3 and free elemental selenium from the back contact [14]. However, the XPS peak fitting of the Se region by Chen et al was dramatically inconsistent with both Wang et al [13] and this work. In this work, the binding energy separation between elemental Se and Sb 2 Se 3 in the Se region is ∼1.3 eV, consistent with Wang et al, and other reports [13,34,37]. However, Chen et al [14] state this separation as ∼0.6 eV, combined with a very broad line shape for the alleged elemental Se component. Given the 0.1 eV precision of binding energy determination in XPS, this difference casts doubt over the validity of these assignments, and therefore it cannot be concluded from their work that the (NH 4 ) 2 S treatment is in fact effective in removing free elemental selenium. Indeed our results show there to be no removal of elemental selenium and the improvement they observed in device performance is most likely due to removal of Sb 2 O 3 , which would be consistent with the work carried out in this study.
Our results indicate that while these chemical etches have led to an improvement in device performance in previous studies they are in some cases harmful [13,14]. The removal of Sb 2 O 3 from the back contact decreases the barrier height and lowers the series resistance, however the decrease in J sc and R sh in J-V measurements can be explained by damage to the film upon etching. Whilst it is likely that the removal of a resistive Sb 2 O 3 layer would benefit device performance by lowering of the series resistance and increasing the FF, there are clear differences in the initial level of contamination on the surfaces of our samples compared to those of Wang et al and Chen et al [13,14]. Our samples show very low levels of Sb 2 O 3 and free elemental selenium contamination compared to those shown in the previous studies, presumably due to the different deposition method. For example, the RTE process uses no inert gas during deposition and also deposits at a much higher rate than CSS and at a different pressure (∼0.01 mbar rather than ∼13 mbar) [4]. Additionally, in the RTE process the Sb 2 Se 3 melts before evaporating, whereas in CSS the material directly sublimes onto the substrate which could lead to differences in contamination from the source material [4]. It seems likely therefore that the state of the asdeposited Sb 2 Se 3 surface plays a major role in the effectiveness of these chemical etches. We postulate that if the sample has a thick layer of Sb 2 O 3 contamination then the etches are effective in reducing the thickness of a resistive layer, thereby improving the fill factor of the device. However, if the Sb 2 O 3 layer is very thin, the reduction in R s is counteracted by a more significant decrease in R sh and J sc that lowers the overall performance of the cell. This is supported by the fact that EQE measurements show that the (NH 4 ) 2 S-etched sample has uniformly lower EQE across all wavelengths whilst showing no change in shape in the EQE spectrum (figure S1).
Building on these conclusions, a more dilute treatment of the (NH 4 ) 2 S etch was applied using the same process as previously described but with a concentration of 0.03 mmol dm −3 . This succeeded in improving the device performance to over 6.6%, as shown in figure 9 (PCE=6.63%, V oc =0.42V, J sc =32.5 mA cm −2 , FF=48.5), which further supports the conclusions of this work and shows a way forward in optimising back contact treatments for Sb 2 Se 3 solar cells.

Conclusion
This work shows with clarity the benefits and drawbacks of etching prior to application of the back contact for Sb 2 Se 3 solar cells and resolves inconsistencies in the literature. XPS analysis of the back surface shows a reduction of Sb 2 O 3 contamination from both etches. J-V-T analysis reveals a significant decrease in potential barrier at the back contact upon etching, from 0.43 to 0.29 eV with the (NH 4 ) 2 S etch, likely as a result of Sb 2 O 3 removal. J-V analysis reveals a decrease in series resistance after etching which can be attributed to the presence of a smaller amount of resistive Sb 2 O 3 . Additionally, there is an decrease in J sc and overall efficiency following both etches due to damage of the film. Upon diluting the (NH 4 ) 2 S etch, device efficiency was improved to 6.6%. Overall, the study clarifies the role of the etches in removing surface contamination, showing a more varied result when the initial deposition conditions are changed, and provides insight into the way forward in optimising Sb 2 Se 3 back contacts.