# Chemical instability at chalcogenide surfaces impacts chalcopyrite devices well beyond the surface

Jul 20, 2020

### Point defect solubility limits and phase metastability

Throughout the manuscript, the word ‘metastability’ refers to a sample condition describing a stable state of a dynamical system other than the system’s state of least energy. Depending on the context, the term applies both to phase equilibria and electronic defect states.

The starting point for discussion is the In2Se3-Cu2Se pseudobinary system, as determined by Gödecke et al.8. Figure 1a illustrates the portion of phase diagram in the range 15–30 atomic % (at.%) Cu, showing the CIS homogeneity region of existence (α). The edges of the homogeneity region under equilibrium conditions are indicated by the purple solid lines and area. These edges were determined from the analysis of bulk material subject to step cooling, i.e., alternating slow cooling periods (2 °C min−1) and periods at constant temperature. On the contrary, water-quenching from the sphalerite δ phase at 850 °C revealed the existence of a metastable CIS phase (αm), shown by the dotted purple lines and patterned areas.

The reported existence of a metastable phase means that—from a thermodynamic point of view—there exists a region of relative stability (a relative minimum in the Gibbs free energy landscape) that may be reached via alternative sample histories, i.e., not necessarily by quenching.

The reported metastable CIS phase extends largely to the Cu-poor side by ca. 4 at.% Cu (red patterned area). Hence, maintaining single-phase CIS with Cu concentration as low as 18.5 at.% (Cu/In = 0.64) is possible. By comparison, highly efficient CIGS is grown as a Cu-poor material with Cu/(In + Ga) ratios as low as 0.8 (21.6 at.% Cu)17,20, i.e., potentially in a CIGS Cu-poor metastable homogeneity region. This should not be a surprise, because the deposition of CIGS thin films is a complex process performed out of thermodynamic equilibrium21.

The extension of the metastable CIS phase towards Cu-rich compositions is narrower, but still appreciable (cyan patterned area), and clearly visible in the zoomed portion of the diagram (Fig. 1b). It may then be possible for Cu supersaturation to occur in CIS phases grown as thin films under large Cu excess and after unforced cooling (Fig. 1c).

The breadth of the phase homogeneity field at growth temperature implies a certain solubility of point defects such as indium on copper antisites (InCu) or copper vacancies (VCu) (or both) on the Cu-poor side, as well as copper on indium antisites (CuIn), indium vacancies (VIn) or copper interstitial atoms (Cui) (or combinations of the three) on the Cu-rich side22. The solubility limit for these defects in the CIS lattice is even larger in the metastable phase.

The native doping density within this compositional window up to just above stoichiometry changes by nearly four orders of magnitude23. Both a drop of solubility limit from growth to room temperature (as proposed for sodium dopant24) and the possible occurrence of αm are likely to have an impact on CIS processing (i.e., film microstructure during growth and reactivity after cooling), and hence PV performance.

### Composition-dependent CIS micro/nanostructure

As revealed in the cross-sectional images (Fig. 1b), the microstructure of the films varies with increasing Cu content. Cu-poor CIS displays grains of maximum 1 µm size. Conversely, Cu-rich CIS has much larger grains, with at least double the lateral width and height up to 3 µm. The pronounced grain growth of Cu-rich films is generally attributed to the vapour–liquid–solid (VLS) mechanism exerted by Cu-Se fluxing phases during growth25. Crucially, stoichiometric CIS shows intermediate grain sizes, but the VLS mechanism cannot apply in this case, because no Cu-Se phases were formed during this ‘single-stage’ growth process.

Mainz et al.26 have studied in detail and in-situ the annihilation of planar defects in Cu-poor CIGS during annealing with increasing Cu content. Importantly, they observed that the annihilation rate increases rapidly, shortly before Cu2Se phase segregates at the CIGS surface, and attributed domain growth to the resulting stress relaxation27.

However, enhanced atomic diffusivity could also stem from copper supersaturation, especially in the presence of Cui expected from the larger defects pool at growth temperature. Cui defects could then migrate via the Frank–Turnbull dissociative mechanism, leading to lower migration barriers and faster atomic rearrangements28.

The fact that the planar defect annihilation onset coincides with the incipient formation of Cu2Se is intriguing. To reveal the effect of growth conditions on the nanoscopic structure of CIS and test the existence of the supersaturated metastable phase, atom probe tomography (APT) is performed and shown in Fig. 2. APT analysis of stoichiometric CIS reveals a non-uniform distribution of Cu within the specimen. Some regions of the sample far exceed the stoichiometric Cu concentration, up to above 30 at.% near what appears to be a grain boundary (GB), as evidenced by alkali metal decoration29, cf. region of interest 2. There is no clear-cut phase separation within the Cu-rich region to the left of the GB. The higher Cu content in the region is only weakly compensated by lower In and Se. This is compatible with the existence of metastable Cu supersaturation in single-stage CIS thin films, as reported by Gödecke et al.8 for bulk material. Having detected such Cu-rich regions in stoichiometric CIS also suggests that Cu supersaturation may start to occur at nominal compositions corresponding to the incipient formation of Cu2Se secondary phase (Cu-rich domains also occur in Cu-rich CIS, but are absent in Cu-poor CIS, cf. Supplementary Figs. 1 and 2).

One possibility (i) could be that the Cu-rich domains are accumulations of Cui defects within the lattice of the metastable phase at room temperature, formed by focused ion beam (FIB)-induced aggregation and chemical reduction to elemental copper. Another possibility (ii) could be that the domains result from segregation of the excess Cu2Se at extended defects during cool down (Fig. 1c). Recent high-resolution TEM evidence from Simsek Sanli et al.30 has revealed the presence of interspersed Cu2Se domains within CIGS grains, to which they attributed a topotactical grain growth effect during the three-stage process31. At this point, it is not possible to resolve with certainty which of the two scenarios occurs in reality.

Regardless of the actual mechanism of grain growth, the larger grains obtained under Cu-rich conditions display lower densities of detrimental planar defects26, which is the reason why highly efficient CIGS is grown by deliberately crossing the edge of the homogeneity region during the so-called three-stage process13.

Why then Cu-rich CIS consistently yields worse device performance than Cu-poor CIS, even after the removal of the excess Cu2Se phase14 and despite the proven superior transport properties, is yet to be understood. Here, the composition-dependent reactivity of CIS towards atmospheric oxygen and post-deposition etchants is investigated to solve this conundrum in light of the APT compositional results on stoichiometric CIS and the possible existence of the supersaturated metastable αm phase in CIS thin films.

### Thermochemistry and defect population metastability

The thermodynamic stability of CIGS against reactions at interfaces has been assessed in finished devices32,33,34 and on bare absorbers35,36. Kazmerski et al.37 revealed the formation of In2O3, SeO2 and CuxSe on the CIS film surface above 150 °C38, with Hauschild et al.36 and Lehmann et al.39 finding InOx to be the most prominent oxide in air-exposed Cu-poor films. Figure 3a shows the computed Gibbs free energy of bulk In2O3 formation, respectively, out of In2Se3 [Eq. (1)] and CIS [Eq. (2)], as well as the alternative formation of Cu2O out of CIS [Eq. (3)].

$${mathrm{In}}_2{mathrm{Se}}_{3({mathrm{s}})} + frac{3}{2}{mathrm{O}}_{2({mathrm{g}})} ,rightleftharpoons, {mathrm{In}}_2{mathrm{O}}_{3({mathrm{s}})} + frac{3}{{mathrm{n}}}{mathrm{Se}}_{{mathrm{n}}({mathrm{s}},{mathrm{l}},{mathrm{g}})}$$

(1)

$$2{mathrm{CuInSe}}_{2({mathrm{s}})} + frac{3}{2}{mathrm{O}}_{2({mathrm{g}})} ,rightleftharpoons, {mathrm{In}}_2{mathrm{O}}_{3({mathrm{s}})} + {mathrm{Cu}}_2{mathrm{Se}}_{({mathrm{s}})} + frac{3}{{mathrm{n}}}{mathrm{Se}}_{{mathrm{n}}({mathrm{s}},{mathrm{l}},{mathrm{g}})}$$

(2)

$$2{mathrm{CuInSe}}_{2({mathrm{s}})} + frac{1}{2}{mathrm{O}}_{2({mathrm{g}})} ,rightleftharpoons ,{mathrm{In}}_2{mathrm{Se}}_{3({mathrm{s}})} + {mathrm{Cu}}_2{mathrm{O}}_{({mathrm{s}})} + frac{1}{{mathrm{n}}}{mathrm{Se}}_{{mathrm{n}}({mathrm{s}},{mathrm{l}},{mathrm{g}})}$$

(3)

Conversion of In2Se3 to In2O3 and elemental selenium formally corresponds to oxidation of selenide anions and is very favourable in air (negative ΔG, solid red line). The positive slope of ΔG is consistent with the decreased degrees of freedom of the gas species, implying an even stronger thermodynamic driving force for conversion at room temperature.

Clearly, the Gibbs free energy of CIS formation is insufficient to stabilize the compound34 (dashed red line); hence, unprotected CIS is intrinsically unstable against oxygen (cf. Supplementary Fig. 4 for detailed point defect reaction energetics). The thermodynamic driving force is a necessary but insufficient condition for a reaction to happen (cf. Supplementary Note 1). Nevertheless, Eq. (2) is confirmed to occur at room temperature (Supplementary Fig. 3) and—thermodynamically speaking—cannot be reverted even under exposure to a static atmosphere of metastable Se2 molecules generated hypothetically by a cracker effusion cell (black dotted line). Stronger reagents, such as H2, are needed to reduce In2O340.

Equation (3) leading to Cu2O is only barely favourable, so the formation of In2O3 is thermodynamically more likely, in agreement with Hauschild et al.36 and Lehmann et al.39.

The presence of humidity was shown to accelerate further the oxidation of the related GaSe 2D compound41, so it is very likely that exposure of CIGS samples to (moist) air during processing affects the CIGS surface chemistry and the physics of the subsequent solar cell device.

To corroborate these considerations, a Cu-poor epitaxial CIS film, i.e., a proxy for CIS bulk (free from grain boundaries), was subject to a graded oxidation. The resulting library was characterized by scanning electron microscopy (SEM)/energy-dispersive X-ray spectroscopy (EDS) at low acceleration voltage and by photoluminescence (PL) along the macroscopic gradient, as shown in Fig. 3b. The regions closer to the O2 source result in a higher In/Cu ratio (full red dots) and O content (hollow dots), consistent with the outgrowth of an In-O phase. Concomitantly, the PL yield (black curves) is lower and a shoulder peak appears, as shown more clearly in Fig. 3c.

To gain mechanistic insights into the CIS surface reactions, it is useful to relate the thermochemistry at macroscopic level to the physics and chemistry of point defects in the material42. The shoulder peak at ca. 0.9 eV is ascribed to the transition from the conduction band minimum to the indium vacancy (VIn) in CIS43. It is hypothesized that the formation energy of VIn in CIS (ca. 2.5 eV/VIn44,45) resulting from the reaction between indium atoms and gas-phase oxygen is offset energetically by the large Gibbs free energy gain associated to Eq. (2) (ca. 2.9 eV/In atom).

The formation of VIn under kinetic control implies that the outgrowth of In-O does not occur entirely as per Eq. (2), but leads—at least partially—to a defected CIS lattice without the segregation of Cu-Se phases, as per Eq. (4).

$$2[{mathrm{Cu}}_{left( {1 – {{y}} – {{z}}} right)}({mathrm{In}}_{{mathrm{Cu}}})_{{y}}({mathrm{V}}_{{mathrm{Cu}}})_{{z}}]{mathrm{InSe}}_{2({mathrm{s}})} + frac{{3{{x}}}}{2}{mathrm{O}}_{2({mathrm{g}})} to\ {{x}mathrm{In}}_2{mathrm{O}}_{3({mathrm{s}})} + 2[{mathrm{Cu}}_{left( {1 – {{y}} – {{z}}} right)}({mathrm{In}}_{{mathrm{Cu}}})_{{y}}({mathrm{V}}_{{mathrm{Cu}}})_{{z}}]left[ {left( {{mathrm{In}}} right)_{1 – {{x}}}left( {{mathrm{V}}_{{mathrm{In}}}} right)_{{x}}} right]{mathrm{Se}}_{2({mathrm{s}})}$$

(4)

From the phase diagram viewpoint, Eq. (4) is ensured by the Cu solubility range (Fig. 1b), entailing an increase of the Cu/In ratio of the underlying Cu-poor CIS without Cu2Se ejection. Under thermodynamic equilibrium, the increased Cu/In eventually translates into a lower share of (InCu + VCu), but from a kinetic standpoint VIn defect formation is justified: given the relatively high energy barrier for In migration in CIS (ca. 1.1 eV46), once VIn is formed it is not easily dissolved.

It is then revealed that interface reactions can lead to a chalcopyrite material with metastable defect populations at temperatures as high as 550 °C.

### Composition-dependent CIS reactivity against O2

Equation (2) is now assessed at room temperature as a function of the Cu content in CIS. Figure 4a shows the X-ray photoemission spectroscopy (XPS) signature of Cu-poor, stoichiometric and Cu-rich CIS after air exposure of the single-phase films for 24 h (i.e., films previously subject to removal of the excess Cu2Se by surface chemical etching with potassium cyanide (KCN)).

The surface composition of air-exposed Cu-poor CIS is consistent with In-O outgrowth, at least partially compatible with Eq. (4), although Cu-Se segregation is not excluded considering that atomic redistribution at room temperature is slower than at 550 °C. Stoichiometric CIS shows a similar behaviour.

Conversely, the surface composition of air-exposed Cu-rich CIS reveals a higher Cu-Se concentration, which is compatible with larger Cu-Se outgrowth. A reaction pathway alternative to Eq. (4) should then be invoked to account for the XPS data in this case. If Cu-rich CIS is a Cu supersaturated αm solution, the Cu-Se outgrowth may proceed by consumption of Cu interstitial atoms and formation of VSe, as per Eq. (5), where indium oxide formation leads to In subtraction from a material already oversaturated with Cu.

This can exclusively be compensated by the formation of the Cu-Se secondary phase in Cu-rich material, unless the limit of Cu equilibrium solubility in CIS is exceeded.

$$2[{mathrm{Cu}}left( {{mathrm{Cu}}_{mathrm{i}}} right)_{{z}}]big[ {{mathrm{In}}_{(1 – {{z}})}{mathrm{Se}}_{left( {1 – {{z}}/2} right)}} big]_{({mathrm{s}})} + frac{{3{{x}}}}{2}{mathrm{O}}_{2({mathrm{g}})}\ to {{x}mathrm{In}}_2{mathrm{O}}_{3({mathrm{s}})} + {{y}mathrm{Cu}}_2{mathrm{Se}}_{({mathrm{s}})} \ ,,,,,, + 2big[ {{mathrm{Cu}}left( {{mathrm{Cu}}_{mathrm{i}}} right)_{({{z}} – {{y}})}} big]big[ {{mathrm{In}}_{(1 – {{z}} – {{x}})}left( {{mathrm{V}}_{{mathrm{In}}}} right)_{{x}}} big][{mathrm{Se}}_{left( {1 – {{z}}/2 – {{y}}/4} right)}({mathrm{V}}_{{mathrm{Se}}})_{{{y}}/4}]_{2({mathrm{s}})}$$

(5)

Figure 4b, c exemplify the extreme cases of interface reactions from the perspective of CIS point defects under Cu deficiency and Cu excess, as inferred from the experimental evidence. The formation of indium oxide provides most of the thermodynamic driving force for Eq. (5) to happen, but the Cu excess in Cu-rich CIS appears to increase the reaction rate compared with Cu-poor CIS, which can be represented by a lowering of the reaction energy barrier, as suggested in Fig. 4d.

It is proposed that the nucleation and growth of the Cu2Se phase in the Cu-rich film is eased either by the large pool of Cu interstitials present in the Cu supersaturated αm solution and/or by the pre-existence of Cu2Se domains acting as seeds.

It is then established that interface reactions can have profound consequences for the underlying defect chemistry of chalcopyrite even at room temperature without any liquid phase being involved. Furthermore, the defect reaction pathways can be markedly diverse if the substrate phase happens to occur near the edges of its existence region, which can have fundamental implications for further processing and ultimately for device performance.

### Composition-dependent CIS stability against cyanide

Cu-Se phases are very conductive and reduce severely the shunt resistance of copper-based chalcogenide solar cells18,25. Whether they occur as residues of absorber fabrication or following exposure to ambient air, they are effectively removed by aqueous cyanide (CN) solutions, presumably following Eq. (6) or Eq. (7).

$${mathrm{Cu}}_2{mathrm{Se}}_{({mathrm{s}})} + 7{mathrm{KCN}}_{({mathrm{aq}})} + frac{1}{2}{mathrm{O}}_{2({mathrm{g}})} + {mathrm{H}}_2{mathrm{O}}_{({mathrm{l}})}\ to 2{mathrm{K}}_2left[ {{mathrm{Cu}}left( {{mathrm{CN}}} right)_3} right]_{({mathrm{aq}})} + {mathrm{KSeCN}}_{({mathrm{aq}})} + 2{mathrm{KOH}}_{({mathrm{aq}})}$$

(6)

$${mathrm{Cu}}_2{mathrm{Se}}_{({mathrm{s}})} + 6{mathrm{KCN}}_{({mathrm{aq}})} to 2{mathrm{K}}_2left[ {{mathrm{Cu}}left( {{mathrm{CN}}} right)_3} right] + {mathrm{K}}_{mathrm{2}} {mathrm{Se}}_{({mathrm{aq}})}$$

(7)

KCN etching is also known to restore, at least partially, the original optoelectronic properties of aged absorber layers47,48, even after oxidation by strong agents such as hypochlorite19; hence, PV device fabrication baselines may include a cyanide etching step, enabling a diode-like electrical behaviour18,49. However, unlike widely accepted, cyanide is unable to chemically remove the In2O3 phase formed as a result of CIS oxidation, i.e., any removal of such a phase upon cyanide etching is consequential to the chemical dissolution of interspersed Cu-Se phases, as per Eq. (6) or Eq. (7) (cf. Supplementary Fig. 3).

Having established the composition-dependent reactivity of CIS towards ambient air, it is important to understand whether the classical KCN etching also alters the defect population near the interface50 differently, depending on the CIS bulk composition and etching conditions. To this end, unetched films (KCN-U) are compared with films subject to two etching conditions: a ‘weak’ (KCN-W) etch (5 wt.% KCN for 30 s duration) and a ‘bold’ (KCN-B) etch (10 wt.% KCN for 300 s duration) on selected samples of Cu-poor, stoichiometric and Cu-rich CIS. The sample naming in the following figures is attributed combining the CIS chemical composition (‘Poor’, ‘Stoichiometric’ or ‘Rich’) with the KCN etching condition (‘Unetched’, ‘Weak etch’ or ‘Bold etch’), e.g., yielding PW for Cu-poor with weak KCN etch (cf. Table 1 for full naming convention).

SEM/EDS analysis shows the morphological and compositional effects of KCN etching near the CIS surface and deeper in the films (Fig. 5a, b). Figure 5a is the corresponding matrix of top views. The small clusters identified at the surface of Cu-poor unetched and Cu-rich weakly etched CIS films are attributed to remnant Cu2Se phase, as they are clearly removed with further etching. EDS yields compositional information that is reasonably free from roughness convolution. At 7 kV acceleration voltage, the beam interacts approximately with the topmost 200 nm of film, whereas at 20 kV the interaction reaches at least 800 nm depth. It follows that Cu-rich CIS after bold etch shows a more pronounced surface Cu depletion than Cu-poor after bold etch, which is striking considering that the bulk composition is by far the opposite (consistent with the growth conditions and with APT analysis in Supplementary Fig. 1).

The Se content is described by the (Cu + In)/Se ratio. Cu-poor CIS films display values around unity in the bulk and slightly above unity at the surface, irrespective of the KCN etching conditions. Conversely, the Se content of Cu-rich CIS films depends heavily on the type of KCN etching because of the large amounts of Cu2Se in Cu-rich unetched and their partial removal by weak KCN etching. However, importantly, the value of (Cu + In)/Se at the surface of Cu-rich CIS after bold etch is above unity, indicative of surface Se deficiency, given that this film is surely free from Cu2Se. The higher (Cu + In)/Se ratio in the bulk of Cu-rich CIS after bold etch may seem puzzling; however, it is explained by the presence of the Cu-rich domains identified by APT, in line with the higher Cu/In ratio at 20 kV.

Similar conclusions are drawn by means of secondary ion mass spectrometry (SIMS) depth profiling (Fig. 5c). However, when comparing SIMS with EDS, it should be borne in mind that SIMS is affected by film roughness, due to signal averaging (cf. Fig. 5a), and matrix effects. For example, the convolution of topography and matrix effects influences heavily the Cu and Se SIMS yields (hence, the SIMS profiles) of Cu-rich CIS after weak etch, where remnant Cu2Se occurs.

Both EDS and SIMS show that stoichiometric CIS yields the smallest compositional change upon KCN etching.

Overall, it is hypothesized that Cu atoms are removed selectively from the lattice of CIS by KCN; the more so, the higher the deviation from stoichiometry. Furthermore, Se depletion is apparent in Cu-rich CIS, leading to the formation of both Se and Cu vacancies. The extent of removal by the cyanide complexant is thermodynamically more favourable for Cu-rich than for Cu-poor CIS, possibly due to the abundance of Cu interstitial atoms in the supersaturated metastable phase, in line with the argument described in Fig. 4d (cf. Supplementary Fig. 4).

### Optoelectronics of CIS metastability induced by cyanide

Having shown that KCN etching induces different chemical effects at the surface of CIS depending on the growth conditions, the next step is to investigate the effects of such surface alterations on the optoelectronic properties. To this end, photoelectrochemical (PEC), time-resolved surface photovoltage (TR-SPV), PL and thermal admittance (ADM) analyses were performed, giving complementary insights on the most relevant samples.

It is crucial to understand that the time constants discussed for PEC and SPV analyses are not carrier lifetimes, but time constants of electron transfer and of defect transformation/generation. A PEC current or photocurrent (JPh) is defined as the difference between the electrochemical current recorded under illumination and under dark conditions. Negative values, such as those observed here, correspond to photogenerated electron collection by the redox couple in solution from p-type CIS (Supplementary Fig. 5). The effect of bold KCN etching on the PEC transients is shown in Fig. 6a. Cu-poor CIS after bold etch (red) shows the typical photocurrent decrease with time after light perturbation until steady state. The behaviour follows the theory of ‘surface recombination’ at semiconductor/electrolyte junctions, treated extensively by Peter51. Here, it is stressed that the term ‘recombination’ used in this context does not refer to the typical fast decay discussed in PL or cathodoluminescence, due to electron-hole recombination, but to electron transfer to the solution and to the change of charge state of surface defects. The time constants are not carrier lifetimes, but the times needed for the electron transfer or for the (re-)charging of the interface defects. Recombination at the interface (or surface recombination, s.r.) is negligible upon light perturbation if the photogenerated carriers are collected fast enough by the redox couple (i.e., with a high kinetic constant for electron transfer, ktr). However, if the electron transfer is sufficiently slow (ktr lower than ks.r.), a bottleneck is created, i.e., the minority carriers accumulate near the interface over time, leading to recombination through leaky surface states and a decrease of photocurrent until steady state. This recombination pathway effected by surface states is most probably due to electrically conductive Cu-Se phases, as no quasi-Fermi level splitting can exist in metal-like solids. The inset compares the ks.r. of Cu-poor and Cu-rich CIS, both after weak etch (both of which display this typical transient) determined as a function of air-exposure time after etching at room temperature [cf. Eq. (8) in “Methods” section]. After weak etching, Cu-rich shows higher ks.r. than Cu-poor CIS and a higher rate of increase, compatible with the larger amount of Cu-Se at the surface (incomplete removal by weak KCN etching, Fig. 5a) and surface oxidation behaviour (XPS, Fig. 4).

The shape of the photocurrent transient in Cu-rich CIS after bold etch is very unusual, because JPh increases with time after the start of the light perturbation (note that the figure plots the measured current density that comprises dark current and photocurrent). Evidently, a different additional recombination or trapping pathway induced by the bold KCN etching must be invoked to account for this uncommon behaviour. The situation is explained by the presence of another defect(s) inside the absorber near the surface with different time constant (tentative band diagram in Fig. 6a), which saturates slowly. The observed time constant is not the carrier lifetime, but the time needed to change the charge state of the involved defects(s).

After bold etch, the stoichiometric CIS film shows a PEC transient under illumination that is a linear combination between those of Cu-poor and Cu-rich subject to the same bold etch. This confirms undoubtedly that the concentration of the point defect(s) formed upon bold KCN etch depends heavily on the Cu/In composition, hence the Cu and In chemical potentials in CIS, with Cu-rich CIS being the most susceptible to the defect(s) formation.

Intensity-modulated photocurrent spectroscopy (IMPS) was performed to gain quantitative information on the slow recombination pathway apparent in Cu-rich CIS after bold etch (Supplementary Fig. 6). In fact, after bold etch, both Cu-rich and Cu-poor films display recombination pathways with long time constants (2.2 µs and 2.5 µs, respectively). However, the extent of recombination for Cu-rich is ca. one order of magnitude higher than for the Cu-poor case, such that the pathway is perceived in the time domain of the Cu-rich film, but not in that of the Cu-poor one.

TR-SPV measured by Kelvin probe force microscopy (KPFM) is used to analyse the effect of bold etching on the photogenerated carrier dynamics of Cu-poor and Cu-rich films (Fig. 6b, c). The SPV, defined as the difference between the contact potential difference (CPD) under illumination and the CPD in the dark, is measured as a function of the modulation frequency of a light source. Regions II of the SPV (time domain) transients show an initial fast response reaching a maximum SPV at the beginning of the illumination sequence (starting at 10 MHz), followed by a slow decay and an abrupt drop of the SPV signal at the end of the sequence (finishing at 100 Hz). Upon initial illumination, both samples show a positive SPV, consistent with a p-type material behaviour. However, Cu-rich CIS after bold etch yields an SPV of 75 mV, which is approximately half the value obtained for the analogous Cu-poor film. A lower SPV is indicative of a smaller change of band bending at the surface, which can be related to a lower minority charge accumulation at the illuminated surface of CIS-RB. The average SPV spectra in the frequency domain (Fig. 6c) reveal decay times of τCIS-PB 6.2 ± 1.0 µs and τCIS-RB 0.75 ± 0.2 µs. This trend is consistent with a higher recombination rate for Cu-rich CIS after bold etch, in agreement with the relative ks.r. of Cu-rich and Cu-poor films subject to bold etch obtained from PEC.

During the illumination sequence, the average SPV of Cu-poor CIS after bold etch decays to ca. 50% of the initial value, indicating a full relaxation of the SPV (for sufficiently slow modulation above 10 kHz, the SPV is at the maximum and at 0 for half of the cycle, giving an average of 50%). In fact, when the light excitation is off after the modulation spectrum, the CPD value in dark conditions is fully recovered. However, the CPD of Cu-rich CIS after bold etch does not drop to the dark value after finishing the modulation spectrum, but a finite SPV value is maintained at about 20 % of the total SPV for at least ca. 26 min after switching the excitation light off (Region III). Thus, the average SPV decays only by 30% from the SPV maximum, indicating that the fast decay of the SPV does not lead to a full relaxation. The remnant SPV is attributed to the presence of charged trap states. This finding is consistent with the presence of additional recombination or trapping pathways in Cu-rich CIS after bold etch already identified by PEC.

The PL spectra (Fig. 6d) are acquired immediately after the etching (less than 3 min), to minimize surface degradation48. For Cu-poor CIS, the highest PL yield is displayed by the film subject to weak etch, showing a fourfold increase compared with the unetched, consistent with the known CIS surface renewal by cyanide etching with the removal of Cu2Se (Fig. 5a). However, bold etching appears to be detrimental for PL yield, suggesting the formation of detrimental defects.

For Cu-rich CIS films, bold etching leads to higher luminescence than weak etching. This is attributed to the fact that weak KCN etching is unable to remove the entire Cu-Se phases (cf. Fig. 5a) responsible for reduced photon absorption by CIS and subsequent emission due to light scattering, as it is most evident for the unetched film.

The energy of the PL maxima of Cu-poor and Cu-rich CIS films reflects the different bandgaps of the absorbers52 and the different extent of electrostatic potential fluctuations53.

The bar chart in Fig. 6d shows the relative activation energy (Ea) of the main capacitance step extracted from the slope of the Arrhenius plot of the ADM spectra of Schottky junctions formed with the bare absorbers (Supplementary Fig. 7). The activation energies provide information on the electrically active defects in the material. A value of 200 ± 20 meV is obtained if bold KCN etching is performed, regardless of the CIS composition and cyanide counterion (cf. Supplementary Fig. 7), twice as much as for Cu-poor after weak etching (100 ± 10 meV). No spectra could be recorded on unetched films and Cu-rich films after weak etching, due to the presence of the conductive Cu-Se phases at the surface.

A recent DFT study by Simsek Sanli et al.54 on the effect of CuIn antisites and Cu interstitials in CIS reveals the appearance of states at 270 meV above the valence band maximum and 210 meV below the conduction band minimum, respectively. However, it is not obvious why the population of these two defects would increase upon KCN etching. Instead, the 200 ± 20 meV defect could be related to the observed etching-induced depletion of Cu and Se from CIS. Besides, the CuIn and Cui have clear PL signatures (cf. Spindler et al.4) and do not correlate with the 200 ± 20 meV defect.

### Passivation of metastable defects improves PV efficiency

To shed light on the Cu and Se depletion, and link it to the trap defect(s) identified by PEC and ADM, the surface chemical reactivity of Cu-rich CIS after bold etch is investigated as a function of air-exposure time (Fig. 7a, b). It is revealed that oxygen has a remarkable influence on the shape of its anomalous PEC transient (blue curves shown under illumination) and magnitude of photocurrent (inset). The recombination pathway induced by bold KCN etching appears to be slowly passivated by oxygen incorporation, leading to a transition from the anomalous type of transient observed for Cu-rich (Fig. 6a, cyan) to the classical transient typical of Cu-poor films (Fig. 6a, red) and simultaneous increase of photocurrent density (JPh, inset). The photocurrent density reaches a maximum after 12 min of air exposure and then it decreases, due to interface deterioration caused by the outgrowing Cu-Se and In-O surface phases, as per Eq. (5).

Whatever the nature of the defect created by the bold KCN etching, oxygen appears to neutralize it, reducing the near-surface recombination pathway. It is plausible that oxygen passivates selenium vacancy donor defects, as proposed by Kronik et al.55. The Se vacancies formed during etching would effectively compensate or even type-invert the near-surface region of CIS, causing efficient capture of the photogenerated electrons, as observed within the PEC transient (ktr lower than kb.t. ~ ks.r.). During subsequent air exposure, the vacancies would capture oxygen, forming compensating acceptors that remove the electron trap and restore p-type doping.

To provide a hint on the nature of such a defect and its effect on the performance of solar cells, the composition-dependent performance of CIS solar cells is now investigated with respect to the chemical reactivity towards surface treatments. Solar cells comprising Cu-poor and Cu-rich absorbers subject to either weak or bold etching were fabricated following a standard baseline including CdS layer deposition by chemical bath (CBD). The electrical characteristics of the devices under simulated sunlight are shown in Fig. 7c.

The different etching does not affect appreciably the performance of Cu-poor-based devices in terms of electrical characteristics (red curves in Fig. 7c). Conversely, no valuable electrical data could be extracted from devices based on Cu-rich unetched and weakly etched absorbers, due to the residual Cu2Se and resulting resistor-like behaviour (cf. RW dashed cyan curve). Only with bold KCN etch, it is possible to effect a diode curve out of an absorber with Cu-rich CIS composition (RB solid cyan curve).

However, it is clear that the device based on Cu-rich CIS subject to bold KCN etching is affected by a significant deficit of open circuit voltage (VOC) compared with devices based on Cu-poor absorbers, regardless of the KCN etching type. In light of the PEC and ADM Schottky analysis, bold KCN etching could then be responsible for the formation of defects that cause recombination at the interface between the surface of Cu-rich CIS after bold etch and the buffer layer, hence the enduring VOC deficit of Cu-rich compared with Cu-poor CIS-based devices56. A series of experiments was performed to test this hypothesis.

If the origin of the VOC deficit is linked to the presence of Se vacancies, then post-deposition treatments (PDTs) performed on the absorber surface under Se-rich environments, should improve the VOC. Three types of such PDTs were performed: Se-KF (selenium and potassium fluoride), Se-only and Se-In57,58 (cf. Table 1 and ‘Methods’ section). All three treatments effectively improve the VOC (cf. JV curves in Fig. 7c and ref. 57), strengthening the hypothesis that the detrimental defect introduced by bold KCN etching is indeed related to Se depletion (the changes observed after the PDTs are linked to the Se incorporation and are not merely due to the associated heating).

To link the solar cell performance to the electronic defects identified by PEC and ADM on the bare absorbers, the PV devices were also characterized by ADM. The activation energies extracted from the Arrhenius plots are shown by the bar chart in Fig. 7c.

ADM reveals that the 200 ± 20 meV defect of Cu-poor and stoichiometric CIS films after bold etching disappears after the absorbers are subject to device completion (Supplementary Fig. 7), consistent with the fact that the etching type has negligible effects on the performance of Cu-poor-based devices. The activation energies become 100 ± 10 and 130 ± 10 meV, similar to those of devices based on weakly etched absorbers with Cu-poor and stoichiometric composition, which are close to the common A2 (100 meV) and A3 (130 meV) acceptor defects found in PL43,59. The removal of the 200 ± 20 meV defect in the devices based on Cu-poor and stoichiometric absorbers is consistent with the passivation of Se vacancies via sulfur incorporation during the baseline deposition of the CdS buffer layer by CBD.

Conversely, along with the significant VOC deficit, devices based on Cu-rich absorbers subject to bold etch also retain the 200 ± 20 meV defect, unlike cells based on Cu-poor CIS. It is important to note that both this defect and the VOC deficit are observed in all devices based on Cu-rich absorbers subject to bold etch regardless of the Cu/In ratio (above 1) or Se flux during absorber growth57. Crucially, Se-based PDTs succeed in passivating the 200 ± 20 meV defect, uncovering the 130 ± 10 meV A3 acceptor60 and improving VOC. Therefore, a link is established between the VOC deficit, the 200 ± 20 meV defect and the bold KCN etching on Cu-rich CIS absorbers.

Moreover, devices comprising Cu-rich CIS absorbers subject to bold KCN etching and Zn(O,S) layers deposited by CBD with eight times the concentration of sulfur source compared with CdS CBD do not display the 200 ± 20 meV defect and also show a significant VOC improvement compared with standard CdS-based devices comprising Cu-rich films subject to bold KCN etching (Table 2). It is then clear that the longstanding 200 ± 20 meV defect is responsible for the VOC deficit of devices based on Cu-rich CIS subject to bold etch due to the preferential formation of Se vacancies induced by bold KCN etching on the metastable Cu-rich CIS phase. If Cu supersaturation of this phase is in the form of interstitial Cu atoms (cf. APT in Fig. 2), the coordination of Cu and Se by the cyanide ions in solution would be thermodynamically more favourable compared with Cu-poor CIS.

The hypothesis is consistent with the near-surface composition revealed by EDS and with positron annihilation studies by Uedono et al.61. Cu and Se vacancies can cluster to form Cu-Se divacancy defect complexes. These complexes can account for larger losses in open circuit voltage than mere Se vacancies. Indeed, the energy level of the divacancy complex occurs deeper in the gap than the Se vacancy alone62. It follows that the improved performance upon the various PDTs is in accordance with the replenishment of Se vacancies with Se, even if the Cu vacancy population remains unchanged. Based on these considerations, Fig. 7d shows a conclusive schematic illustration of the proposed metastable defect population in CIS subject to bold KCN etching and of the passivating effects of CdS CBD and Se-PDTs.

The impact of the findings on solar cell performance is summarized in Table 2, where the device parameters of the various samples are reported.

### Metastability is caused by amphoteric (VSeVCu) divacancies

The theoretical model published by Lany and Zunger in 200662 provides a final support to the hypothesis that the preferential Se and Cu depletion caused by cyanide etching is responsible for the appearance of the defect identified by ADM with an activation energy of 200 ± 20 meV.

Figure 8 illustrates a simplified model consistent with the experimental evidence gathered by PEC and TR-SPV (cf. Fig. 6), describing the sequence of events occurring from light perturbation to relaxation. Given that the 200 meV defect is inaccessible by PL due to current detector limitations, the combined information from all the experimental techniques employed in this study and first-principle analysis sheds light on the chemical nature of this yet unidentified defect, i.e., complementing the recent review on CIGS electronic defects4 (cf. Supplementary Note 2 for a discussion on the energetic likelihood for the divacancy defect formation).