Sulfide Chalcopyrite Solar Cells––Are They the Same as Selenides with a Wider Bandgap?

3.1 Shallow Defects

Experimental information on the defects in Cu(In,Ga)Se₂ and Cu(In,Ga)S₂ comes to a large part from low-temperature photoluminescence (PL) measurements on material grown under Cu-excess, i.e., the chalcopyrite is stoichiometric.^{[ 35 ]} In general, spectroscopy of shallow defects is not possible in Cu-poor selenide chalcopyrite: Cu-poor Cu(In,Ga)Se₂ is compensated, mostly by the combined occurrence of In_Cu donor defects and V_Cu acceptor defects.^{[ 29, 36, 37 ]} This compensation leads to fluctuating electrostatic potentials, which broaden and red-shift the low-temperature PL emissions.^{[ 38-41 ]} These broad emissions render spectroscopy of shallow defects impossible. Whether sulfide chalcopyrites show the same broadening when grown Cu-poor, is not finally clarified at the moment, since no PL measurements on material with a Cu/In or Cu/Ga ratio <0.9 have been reported. The reason might be the narrow existence region of Cu-poor CuInS₂, as discussed in Section 4.3. The observation of tail states in Cu-poor sulfides (see Section 4.4) supports the assumption that Cu-poor sulfides are also compensated. In contrast to Cu-poor, the low-temperature PL spectra of CuInSe₂ ^{[ 41 ]} and CuGaSe₂ ^{[ 40 ]} grown under Cu-excess are dominated by two donor–acceptor pair transitions, which start from the same shallow donor (In_Cu antisite or Cu_i interstitial). They involve the two main acceptors: the shallower one (V_Cu), which leads to the higher energy PL emission (DA1), dominating near stoichiometric compositions and the somewhat deeper one (Cu_In) becoming dominant as the Cu-content increases, leading to a somewhat lower energy PL emission (DA2). The attribution of electronic defects to a certain defect chemistry has been based on compositional trends and mostly on comparison with state-of-the-art calculations.^{[ 35 ]}

A similar behavior of the low-temperature PL is found in Cu-rich sulfides CuInS₂ ^{[ 42, 43 ]} and CuGaS₂.^{[ 44-46 ]} The spectra observed in CuInS₂ follow exactly the same pattern as observed in selenide chalcopyrites: they are dominated by two donor–acceptor pair transitions, which involve the same shallow donor and two different acceptors, where again the shallower (involved in DA1) dominates for near stoichiometric compositions and the somewhat deeper one (involved in DA2) increases with higher Cu-content.^{[ 42 ]} Composition-dependent experiments^{[ 47 ]} and first-principle calculations^{[ 19, 48 ]} identify V_Cu as the most shallow acceptor, followed by Cu_In. Thus, it appears that the two dominating acceptors in sulfide and in selenide chalcopyrite are the same. Cu_i has been found as a shallow donor.^{[ 48 ]} The same study also presents In_Cu as a shallow donor, whereas other calculations find In_Cu to be a deep donor.^{[ 19, 49 ]} The shallow donor has also been attributed to the sulfur vacancy,^{[ 50 ]} which could be supported by the fact that CuInS₂ grown or annealed under a low S atmosphere has been observed to be n-type.^{[ 51, 52 ]} A second shallow donor that was reported in earlier experiments could be identified as a phonon replica^{[ 42 ]} Two shallow donor–acceptor pair transitions are also observed in CuGaS₂, which have been identified as due to one shallow donor and two shallow acceptors, however, the composition dependence does not follow the same pattern: The (higher energy) DA1 is observed up to a Cu/Ga ratio of 2.0 and the ratio between the two peaks does not show a characteristic composition dependence.^{[ 46 ]} For both In compounds (sulfide^{[ 42 ]} and selenide^{[ 41 ]}), the DA2 transition becomes dominant already at a very low Cu excess of a few percent, whereas in CuGaSe₂ the DA2 becomes dominant only beyond a Cu/Ga ratio of 1.1^{[ 40 ]} and in CuGaS₂ no clear trend with composition between the intensities of the two DA transitions has been observed. All compositions were determined by energy dispersive X-Ray spectroscopy (EDX). Considering the error in EDX, it remains unclear whether the difference in trend between CuInSe₂/CuInS₂ and CuGaSe₂ is significant or not.

A third shallow acceptor was found in polycrystalline CuInSe₂ grown by coevaporation on glass, covered with Mo.^{[ 53 ]} It was identified as the In vacancy and seems to be favored by the high Se pressure present in the coevaporation process, which can be attributed to the lower formation energy of the In vacancy at high Se pressure.^{[ 37 ]} A similar defect is found in CuInS₂, however not with an increase in Se pressure but with an increase in Na content. Figure  3 shows low-temperature PL spectra of CuInS₂ films. The red and the dark blue line depict the typical spectra^{[ 42 ]} for Cu-rich (red) and for near stoichiometric (dark blue) CuInS₂: both show an excitonic emission near 1.51 eV (EX). The Cu-rich material is dominated by the DA2 transition at 1.39 eV and the near-stoichiometric one by the DA1 transition around 1.43 eV. The light blue spectrum is measured on a sample that was grown in the same process as the near-stoichiometric one, but the substrate, in this case, was covered by a thin NaF film. The spectrum also shows the excitonic emission and the DA1 emission, but in addition another DA emission at 1.37 eV, which we label DA3. Its phonon replicas are clearly visible towards lower energies with the same 39 meV distance as the phonon replicas of DA1 and DA2. Furthermore, the flux distribution can be nicely fitted by a Huang-Rhys distribution^{[ 54, 55 ]} (see Figure S1, Supporting Information). The DA nature is confirmed by a slight blue shift of 3.8 meV per decade with increasing excitation intensity, together with a power-law exponent of the intensity that starts out at 1 and becomes ½ at higher excitations (see Figure S2, Supporting Information). This change in the power-law behavior is exactly what is expected of a DA transition that is energetically between another one.^{[ 56 ]} From the free-to-bound transition observed at 80 K (Figure S3, Supporting Information), the third acceptor can be placed at (170 ± 5) meV above the valence band.

In CuInSe_2, it was found that the appearance of the third acceptor is linked to sufficient Se pressure during growth.^{[ 53 ]} In CuInS_2, it seems that Na enhances the third acceptor. This raises the question if the Se-dependent behavior in CuInSe₂ is actually due to the enhanced transport of Na in the presence of Se.^{[ 57 ]} Further investigations are needed to clarify which growth conditions strengthen the A3 acceptor in CuInSe₂ and CuInS₂.

A third DA transition has also been observed in some CuGaSe₂ films, grown under low Se supply and without Na, however, its nature is unclear.^{[ 35 ]} CuGaS₂ films show a third DA transition, with an increasing intensity with an increasing Cu/Ga ratio.^{[ 46 ]} Thus, it appears that all Cu(In,Ga)(S,Se)₂ chalcopyrites show a third shallow acceptor, although it is not clear whether all “A3” acceptors have the same defect chemistry.

In summary, the shallow defects in CuInSe₂, CuGaSe₂, CuInS_2, and CuGaS₂ are very similar. All four materials feature three shallow acceptors and one shallow donor. The observed defects and their low-temperature PL transitions are summarized in Figure  4 and Table  2 . It is interesting to note that the shallow donor has an energy of around 10 meV in both selenides (CuInSe₂ and CuGaSe₂), whereas it is found at around 30 meV in both sulfides (CuInS₂ and CuGaS₂). The acceptors become deeper when replacing Se with S. In the selenides, the three shallow acceptors become deeper when replacing In with Ga, in agreement with the expectations from the hydrogen-like defect model.^{[ 58 ]} There is no clear trend with Ga addition in the sulfides: the two most shallow acceptors are deeper in CuInS₂ than in CuGaS₂, whereas the third acceptor is deeper in CuGaS₂ than in CuInS₂. Not only are the doping defects similar in selenide and sulfide chalcopyrites, also the observed doping levels are similar: in Cu-poor Cu(In,Ga)S₂ the net doping has been observed in the range of several 10¹⁵–10¹⁶ cm⁻³, whereas Cu-rich Cu(In,Ga)S₂ shows a net doping in the low 10¹⁷ cm⁻³ range.^{[ 19, 59, 60 ]}

3.2 Deep Defects

The shallow defects, discussed in Section 3.1, are responsible for the doping behavior of Cu(In,Ga)(S,Se)₂ absorbers. For the performance of these materials in solar cells, deep defects play a decisive role since they act as recombination centers and lead to nonradiative recombination, which decreases the open-circuit voltage of the solar cell.^{[ 61-63 ]} In general, wide bandgap semiconductors are more likely to show deep defects. Many deep defects tend to remain at the same absolute energy, see, e.g.,.^{[ 35, 64, 65 ]} Thus, defects that are shallow in narrow bandgap semiconductors can become deep in wide-gap semiconductors. An example is the Ga_Cu antisite defect in the selenide chalcopyrites, which is deep in CuGaSe₂ and becomes shallow with the addition of In below a Ga/(Ga+In) ratio of about 0.5.^{[ 35 ]} This defect is responsible for the PL transition in Cu(In,Ga)Se₂ at 1.1 eV.^{[ 35, 66 ]} Additionally, one deep defect with an energy around 0.7–0.8 eV has been observed in almost all chalcopyrites: in CuInSe₂ and CuGaSe₂,^{[ 35 ]} in Cu(In,Ga)Se₂ ^{[ 67 ]} and in CuInS_2. ^{[ 50, 68 ]} In Cu(In,Ga)S₂ and CuGaS₂ it could be present, but hidden by a very strong emission around 1.05 eV. In Cu(In,Ga)Se_2, the 0.8 eV defect appears to be more prominent in Cu-rich material.^{[ 35 ]} In CuInS_2, it was observed that higher growth temperatures help to reduce this deep defect and at the same time increase the quasi-Fermi level splitting.^{[ 68 ]} In CuInS_2, the most prominent deep defect emission is seen at 1.05 eV in the PL spectra.^{[ 68 ]} A similar emission in Cu(In,Ga)S₂ has been detected in Cu-rich films, but is reduced with decreasing Cu content.^{[ 19 ]} This defect has been related to the open-circuit voltage (V _OC) loss.^{[ 19 ]} An additional emission of around 1.35 eV in Cu(In,Ga)S₂ is also reduced with lower Cu content of the films.^{[ 19 ]} Three additional deep defect transitions are observed in CuGaS₂ at energies of 1.8, 2.1, and 2.2 eV.^{[ 46 ]} The attribution of electronic defects to the defect chemistry is currently much less clear in sulfides than in selenides. An overview of the deep defect transitions observed by PL in the various chalcopyrites is given in Figure  5 . It is important to note that these are the PL emission energies, not the defect energies. And most likely these deep defect transitions are influenced by phonon interactions and do not represent the zero-phonon lines.^{[ 35 ]} To identify defect energies and defect chemistries, more detailed PL investigations and also defect calculations are necessary. But the observed PL transitions give an indication of the presence of deep defects. It appears, that similar deep defects are present in all Cu(In,Ga)(S,Se)₂ chalcopyrites. The 0.8 eV transition has not been observed in Cu(In,Ga)S₂ and CuGaS₂, however, it could just be hidden under the very strong emission of around 1.05 eV. As expected, it is observed that with a wider bandgap, more deep defects appear. These defects are likely also present in the narrower bandgap compounds, but there they are shallow defects, just like the Ga_Cu antisite defect in Cu(In,Ga)Se₂.^{[ 35 ]}

4.1 The Cu-Rich Surface

The same observation has been made for the selenide chalcopyrites:^{[ 34 ]} lower efficiencies, in particular, lower open-circuit voltages are obtained with Cu-rich absorbers, which is due to interface recombination.^{[ 71 ]} It was only recently shown what causes this interface recombination in the selenide chalcopyrites: it is due to a defect that appears around 200 meV activation energy in admittance spectra,^{[ 73 ]} which is caused by the combined effect of oxidation and the necessary etching step in Cu-rich material.^{[ 74 ]} This effect can be traced back to the behavior of Cu(In,Ga)Se₂ when exposed to atmospheric oxygen (see Figure  6 ): when Cu(In,Ga)Se₂ is oxidized, the first to form is the group III oxide (see Ref. 74 and references therein as well as Ref. 75 for the Ga compound). This means that group III atoms are removed from the chalcopyrite phase, making the chalcopyrite more Cu-rich or less Cu-poor. This is no problem in Cu-poor material since it will just move somewhat closer to stoichiometry. However, the stoichiometric chalcopyrite that results from a Cu-rich growth will necessarily cross the boundary in the phase diagram and will form Cu₂Se. To form this phase, Se atoms will be removed from the chalcopyrite phase, creating Se vacancies. This defect or a related defect complex is the cause of the 200 meV admittance signal.^{[ 73 ]} This model is corroborated by the observation that annealing in Se-containing atmosphere removes or reduces this defect and improves the V _OC.^{[ 73 ]} Furthermore, the V _OC loss can be completely recovered by a treatment with In and Se.^{[ 76, 77 ]} The defect was shown to reside in a thin layer (a few tens of nm thick) just below the absorber surface.^{[ 78 ]} Modeling shows that such a layer results in the typical signatures of interface recombination: activation energy of the reverse saturation current lower than the bandgap and a V _OC considerably smaller than the quasi-Fermi level splitting.^{[ 78 ]}

The question arises if a similar mechanism causes the interface recombination in Cu-rich sulfide chalcopyrites. So far, it has not been possible to detect a corresponding defect signature in admittance spectroscopy on Cu-rich CuInS₂ or Cu(In,Ga)S₂ devices, because the spectra are dominated by the high doping level in Cu-rich absorbers.^{[ 19, 59 ]} However, it was shown that the interface V _OC loss, i.e., the difference between V _OC (times unit charge) and quasi-Fermi level splitting, in Cu-rich CuInS₂ solar cells is reduced by S treatment^{[ 79 ]} or by using buffers which are prepared from solutions with a high sulfur concentration.^{[ 80 ]} This observation, together with the similarity in the phase diagrams, suggests that a similar mechanism is active in Cu-rich sulfide chalcopyrites: defects related to the S vacancy are likely caused by oxidation and etching and cause the loss in V _OC due to near-interface recombination.

To summarize this comparison: Cu-rich growth leads to interface recombination in both sulfide and selenide chalcopyrite solar cells, which can be attributed to a defective layer just below the surface containing defects related to chalcogen vacancies.

4.2 Bulk Quality

Although the interface of Cu-poor CuInS₂ solar cells is superior, its bulk has been found to be of lower quality. The electronic quality of the bulk is assessed by the quasi-Fermi level splitting (qFls), determined from absolute PL measurements of the absorber, before finishing the solar cell.^{[ 63 ]} It should be mentioned that selenide chalcopyrites degrade quickly in air, in particular under laser illumination.^{[ 81-83 ]} Therefore, absorbers are covered with a thin CdS layer, which prevents degradation and maintains the qFls of the fresh absorber.^{[ 82, 83 ]} On the contrary, the qFls measured in sulfide chalcopyrites are stable over days, allowing to measure the qFls of sulfide absorbers without the CdS buffer.^{[ 84 ]} This has been observed for Cu-rich and Cu-poor absorbers alike.^{[ 84 ]} Since the qFls depend on the bandgap, it is better to compare the qFls loss, which is defined as the difference between the ideal Shockley–Queisser open-circuit voltage (times the unit charge) and the measured qFls: $q{V}_{\text{OC}}^{\text{SQ}}-\Delta E_{\mathrm{F}}$. The bandgap to determine $V_{\text{OC}}^{\text{SQ}}$ should be taken as the PV bandgap, determined from the inflection point of the external quantum efficiency (EQE) spectrum of the short-circuit current^{[ 85 ]} or the absorptance spectrum.^{[ 86 ]} Figure  7 gives an overview of how the qFls loss depends on the Cu/In or Cu/(In+Ga) atomic ratio. The qFls loss values are based on qFls values determined from a fit to generalized Planck's law, keeping the temperature in the fit to the measurement temperature.^{[ 63 ]} For the CuInS₂ samples $V_{\text{OC}}^{\text{SQ}}$ is determined from the PV bandgap, corresponding to the inflection point of the EQE spectrum. Since we do not have EQE (or even solar cells) of all samples, we use the typical values for the bandgap (see also Figure 9): 1.506 eV for all Cu-rich films, 1.475 eV for 1-stage Cu-poor films, and 1.490 eV for multi-stage Cu-poor films. For the CIGSe samples, already reported in Ref. 19 we used the PL maximum for the bandgap, this may underestimate the qFls loss somewhat (on the order of 10–20 meV).^{[ 63 ]} For the new data (Cu-poor Cu(In,Ga)S₂), the qFls loss was directly determined from the radiative efficiency (Y _PL) according to –kT lnY _PL.^{[ 63, 86 ]} Concerning the CuInS₂ samples, we can see that at least for Cu-rich material, higher growth temperatures are beneficial.^{[ 68 ]} In the literature, an inprovement with higher growth temperature has also been observed in Cu-poor absorbers.^{[ 18 ]} Furthermore, the qFls loss in Cu-poor CuInS₂ absorbers is observed to be higher than in Cu-rich absorbers.^{[ 68, 87 ]} This is seen on soda-lime glass^{[ 68 ]} (full symbols in Figure 7) with a relatively high content of Na, as well as on a low-Na high-temperature glass (open symbols in Figure 7).^{[ 87 ]} Though, it should be noted, that in both studies the Cu-poor films were grown in a single-stage process (circles in Figure 7) and had rather small grains.^{[ 68 ]} Recent experiments with Cu-poor absorbers grown by a three-stage process (green stars in Figure 7) indicate a considerable reduction of the qFls loss in Cu-poor CuInS₂, but still, the Cu-rich absorbers perform better by about 40 meV. It appears that a multistage process is not beneficial for Cu-rich CuInS₂ but improves the qFls in Cu-poor material considerably. Still, even when comparing the best processes for Cu-rich and Cu-poor CuInS₂ and the same growth temperatures, Cu-poor CuInS₂ is of lower quality than Cu-rich material from the viewpoint of bulk material.

The situation is different when considering Cu(In,Ga)S₂ with Ga added (blue symbols in Figure 7). In that case, the Cu-poor multistage films show lower qFls loss compared to any Cu-rich Cu(In,Ga)S₂ ^{[ 19 ]} and compared to any Cu-poor CuInS₂. The comparison between Cu-rich and Cu-poor Cu(In,Ga)S₂ (with Ga) is similar to what is observed in Cu(In,Ga)Se₂: qFls loss is higher in Cu-rich Cu(In,Ga)Se₂ than in Cu-poor Cu(In,Ga)Se₂ when comparing the same bandgaps,^{[ 83 ]} and the difference increases with increasing bandgap.^{[ 35 ]} This observation can be attributed to a deep defect. One could suggest that it is the defect responsible for the 0.8 eV emission in Cu(In,Ga)Se₂ ^{[ 35 ]} and possibly the defect responsible for the emission of around 1.05 eV in Cu(In,Ga)S₂ ^{[ 19 ]} that are both more prominent in Cu-rich material and become deeper, i.e., closer to midgap, when the bandgap increases. Defects around midgap are the most effective for nonradiative recombination. Alternatively, the lower quality of Cu-rich absorbers could be caused by the necessary etching step which damages the surface, as discussed earlier.

4.3 Existence Region and Segregation

The Cu-rich/Cu-poor comparison leads to different results in CuInS₂ (better Cu-rich absorbers) than in Cu(In,Ga)S₂ (better Cu-poor absorbers). Thus, the question arises why it is more difficult to prepare high-quality Cu-poor CuInS₂ (without Ga), whereas high-quality films with an overall Ga/(Ga+In) atomic ratio of around 15% are possible.^{[ 19 ]} The recently determined pseudo ternary phase diagram of the Cu₂S–In₂S₃–Ga₂S₃ system^{[ 30 ]} (Figure  8 ) might give an answer. In contrast to CuInSe₂, where a wide existence region of the chalcopyrite phase is observed for Cu-poor material,^{[ 24, 88-90 ]} the existence region of Cu-poor CuInS₂ is very small: Cu deficiency leads almost immediately to the formation of a secondary phase with a cubic crystal structure.^{[ 30 ]} This is in agreement with early reports that CuInS₂ could not be prepared single phase:^{[ 33 ]} Cu_2-xS was always present in Cu-rich growth (but this phase stays at the surface and can be etched away) and In-rich phases were present in Cu-poor growth. However, the phase diagram in Figure 8 shows that the existence region on the Cu-poor side widens with increasing Ga content in Cu(In,Ga)S₂. Thus, we propose the Ga-induced widening of the Cu-poor Cu(In,Ga)S₂ existence region as an explanation for the different trends in bulk quality in CuInS₂ and Cu(In,Ga)S₂, comparing Cu-rich and Cu-poor material, and for the recent progress with Cu-poor material. As discussed in the previous section: for sulfides as for selenides, Cu-poor material is preferable for solar cells, because it avoids oxidation and etching-related surface damage. But in sulfide chalcopyrite, it is necessary to add Ga to allow Cu-poor compositions.

The phase diagram in Figure 8 implies another problem that concerns the growth of homogeneous Cu(In,Ga)S₂: the black circles indicate the target compositions of the prepared material and the arrows indicate the phases that were obtained: a near stoichiometric chalcopyrite phase and a very Cu-poor cubic or trigonal phase. What is important to note is that the Ga content of the two phases is different: the Cu-poor phase is In richer than the near-stoichiometric phase. This difference seems to increase with increasing overall Ga content. These observations indicate that a growth process with a composition in the two-phase region will not only lead to segregation into phases with different Cu content but also with different Ga contents. This problem is particularly severe in a three-stage process, which starts from (In,Ga)₂S₃ and has to move through the two-phase region. Interestingly, the phase segregation does not occur in a homogeneous distribution of small grains but as a layered structure.^{[ 91 ]} Segregation into Ga-rich and In-rich phases is, however, not limited to three-stage coevaporation processes, it was also observed in sequential processes that start with a metallic precursor, even in Cu-rich films.^{[ 92 ]} The segregation leads to layers inside the absorber with a low Ga content and thus a low bandgap. Since the absorption edge depends on the lowest bandgap, this segregation is detrimental for high open-circuit voltages and for high bandgaps.^{[ 91 ]} It appears, that overall lower Ga contents, leading to an absorption edge up to about 1.7 eV, and higher temperatures allow a better control of the Ga profile throughout the absorber depth.^{[ 18, 19, 21 ]}

4.4 Tail States

A distinct difference between Cu-rich and Cu-poor selenide chalcopyrites is the occurrence of electrostatic potential fluctuations and the correlated tail states in Cu-poor material, which is due to doping compensation. These tail states have been detected by low-temperature PL studies,^{[ 35, 93 ]} by room temperature PL measurement, from which the absorption coefficient can be extracted,^{[ 34 ]} and from the steepness of the EQE spectrum.^{[ 88 ]} Cu poor Cu(In,Ga)Se₂ has stronger potential fluctuations, more tail states, and a more gradual increase of absorption and EQE at the absorption edge. In fact, the steepness of the absorption edge in EQE spectra has been used as an indicator to distinguish between Cu-rich and Cu-poor material.^{[ 94 ]} The same difference has been observed in CuInS₂: Figure  9 shows the EQE spectra of the CuInS₂ solar cells, which are discussed in Ref. 68. The onset of the Cu-poor cell is more gradual than that of the Cu-rich cells, even with lower growth temperature, indicating the same effect of tailing in sulfides as in selenides. It is important to note that the gradual onset, in this case, is not due to a bandgap gradient. These are pure CuInS₂ films, without Ga and without bandgap gradient. The EQE spectra reported in Ref. 19 seem to contradict that statement, since the steepness of the onset does not vary between Cu-rich and Cu-poor cells. It has to be noted, however, that the most Cu-poor sample in Ref. 19 has still a higher Cu content than the Cu-poor sample in Figure 9, which has a Cu/In atomic ratio of 0.9. Also in selenides, the more gradual increase in EQE has been observed typically for samples with a Cu/III atomic ratio around or below 0.9.^{[ 88, 94 ]} A broad asymmetric emission in low-temperature PL is also indicative of the appearance of potential fluctuations and the related tail states.^{[ 35 ]} This signature appears in CuInSe₂ at a Cu/In atomic ratio of 0.91 and below.^{[ 41 ]} It is interesting to note that in CuGaSe₂ the signature of potential fluctuations appears as soon as the material becomes Cu-poor.^{[ 40 ]} In summary: both Cu-poor sulfides and Cu-poor selenides show tail states due to electrostatic potential fluctuations and a more gradual increase of EQE spectrum, compared to Cu-rich material.

5.1 Interface Recombination

As mentioned earlier, the early Cu(In,Ga)S₂ solar cells, which were based on absorbers grown under Cu excess, were dominated by interface recombination, which reduces the open-circuit voltage.^{[ 80, 95, 96 ]} Even the best Cu-rich Cu(In,Ga)S₂ device is limited by interface recombination.^{[ 72 ]} In this context, it is always the absorber–buffer interface that is concerned. Two main measurement methods are used to determine if interface recombination is the main loss mechanism: 1) extrapolation of the temperature dependence of V _OC to determine the activation energy of the reverse saturation current. If the activation energy is smaller than the bandgap of the absorber, it can be assumed that interface recombination is the main loss channel.^{[ 95 ]} 2) a significant difference between qFls and V _OC also indicates severe interface recombination.^{[ 78 ]} The difference is, therefore, labeled the interface V _OC loss.

Interface recombination is caused by unfavorable band alignment, Fermi level pinning,^{[ 95 ]} or a near-interface defective layer.^{[ 78 ]} The band alignment is unfavorable if a cliff configuration exists, i.e., if the conduction band edge in the buffer or contact layer is below the one in the absorber. CdS is the buffer material used most often in Cu(In,Ga)(S,Se)₂ solar cells.^{[ 97, 98 ]} However, it has been shown to form a cliff already with CuInS₂ without Ga.^{[ 99, 100 ]} Similar to Cu(In,Ga)Se₂,^{[ 101 ]} it can be assumed that the addition of Ga shifts the conduction band in the absorber up, making the cliff even larger. The situation is illustrated in Figure  11 . In fact, a rather large cliff has been detected at the Cu(In,Ga)S₂/CdS interface.^{[ 102 ]} This can be mitigated by using buffers with a higher conduction band edge energy.^{[ 19, 72 ]} ZnOS has been shown to form a spike with CuInS₂ ^{[ 103 ]} and was also found to mitigate interface recombination in Cu-poor Cu(In,Ga)S₂.^{[ 19 ]} Other buffers that form a favorable interface with Cu-poor Cu(In,Ga)S₂ have been shown to be ZnMgO^{[ 70 ]} or ZnSnO.^{[ 20 ]}

However, as discussed earlier, solar cells based on Cu-rich absorbers are always dominated by interface recombination, regardless of the buffer used.^{[ 19, 60, 72 ]} Fermi-level pinning can be excluded based on photoemission spectroscopy, where interface-induced band bending is always observed.^{[ 102, 103 ]} The reason for the dominating interface recombination in Cu-rich Cu(In,Ga)S₂ has been discussed earlier: it is due to a defective layer near the surface of the absorber, with defects related to chalcogen vacancies generated by the oxidation behavior and the necessary etching step of the stoichiometric chalcopyrite obtained in Cu-rich growth. This cause of interface recombination can only be avoided by using Cu-poor absorbers. For Cu-poor absorbers containing Ga, CdS buffers may form a cliff; in this case, it is necessary to use an alternative buffer.^{[ 19, 20, 70, 104 ]} All our best solar cells are based on Cu-poor absorber and show an activation energy of the saturation current equal to the bandgap^{[ 19 ]} and an interface V _OC loss around 20 meV or lower, if alternative buffers are used with a higher conduction band edge.^{[ 19, 20 ]} It should be noted, though, that the Cu(In,Ga)S₂ solar cell with a reported bandgap of 1.65 eV and using a CdS buffer is not dominated by interface recombination.^{[ 21 ]}

5.2 Bulk Recombination

If the solar cell is dominated by bulk recombination, the open-circuit voltage will be similar to the quasi-Fermi-level splitting, measured by PL.^{[ 78, 105 ]} Thus, Figure 7 can give insights into the bulk quality of sulfide absorbers. It should be noted though, that the back contact is generally present in PL measurements, hence, back surface recombination will be reflected by a reduced qFls as well.^{[ 106, 107 ]} The best selenide absorbers show qFls losses below 130 meV,^{[ 108 ]} whereas it can be seen from Figure 7 that even our best sulfide absorbers exhibit qFls losses more than twice as high. This goes along with the observation of a much shorter minority carrier lifetime in sulfides than in selenides: Cu-poor Cu(In,Ga)S₂ shows lifetimes of a few ns,^{[ 19, 60 ]} whereas selenide chalcopyrites show lifetimes about two orders of magnitude higher.^{[ 81, 109 ]} A factor of 100 in the lifetime without a change in doping would decrease the V _OC loss or quasi-Fermi-level splitting loss by 120 meV,^{[ 63 ]} which is very close to the observed difference, as just discussed. This observation clearly indicates that the quality of sulfide solar absorbers needs to be improved. Several paths for improvement can be extracted from Figure 7. Multistage processes improve the quality of Cu-poor absorbers (comparing stars with circles in Figure 7). This is similar to the observations at selenide chalcopyrites: the best cells are prepared by a three-stage process^{[ 110, 111 ]} or by a modified three-stage process.^{[ 112, 113 ]} When comparing different growth temperatures in Figure 7 (high: red, medium: green and blue, low: grey) it appears that higher temperatures are beneficial for sulfide absorbers, as reported earlier.^{[ 18, 68 ]} Finally, comparing open symbols (glass with lower Na content) and closed symbols (soda-lime glass) of the same color indicates that for Cu-poor absorbers a higher Na substrate is beneficial. On the Cu-rich side, the influence of Na appears negligible, as has been observed in the past.^{[ 114 ]} Investigations of the influence of additional Na on Cu-poor Cu(In,Ga)S₂ are underway. Preliminary results (see also Table  3 ) show a small positive influence of a NaF post-deposition treatment (PDT) in single-stage Cu-poor CuInS₂, but a negative influence on the (better) absorbers from three-stage processes. A NaF precursor film, deposited on the Mo back-contact before the absorber growth, has a positive effect on CuInS₂, but a negative one on Cu(In,Ga)S₂. Codeposition of NaF during the second or third stage of the three-stage process has a clear positive effect with an increase in qFls of 80 meV. A qFls loss as low as 220 meV has been achieved by this method. In this case, qFls loss is determined directly from the radiative efficiency (Y _PL) according to –kT lnY _PL.^{[ 63, 86 ]} These experiments were done on soda-lime glass.

In Figure 7 we can also compare CuInS₂ and Cu(In,Ga)S₂. One important difference between CuInS₂ and Cu(In,Ga)S₂ absorbers is the presence of a Ga gradient in the latter,^{[ 19 ]} in particular when grown by a three-stage process, same as in Cu(In,Ga)Se₂.^{[ 115 ]} The Ga gradient in Cu(In,Ga)Se₂ absorbers has been shown to suppress backside recombination at Mo back contact.^{[ 107, 116 ]} When comparing CuInS₂ and Cu(In,Ga)S₂, grown under the same conditions or single and multistage Cu(In,Ga)S₂ (stars and diamonds) in Figure 7, there seems to be little influence of the Ga gradient in Cu-rich absorbers. In Cu-poor absorbers, the qFls loss in Ga-containing absorbers is clearly lower. A positive influence of a suitable Ga gradient in Cu-poor Cu(In,Ga)S₂ has been reported earlier.^{[ 18 ]} In selenide chalcopyrites, the exact Ga profile has been extensively optimized.^{[ 117, 118 ]} As discussed in Section 4.3, the control of the Ga gradient might be more difficult in sulfides, because of its phase diagram (Figure 8).^{[ 91 ]} However, the positive effect of Ga addition, obvious in Figure 7, could also be due to the larger existence region of Cu-poor Cu(In,Ga)S₂ with higher Ga, as discussed earlier (Figure 8). With the currently available data, it is clear that the addition of Ga improves the absorber quality, but the exact reason cannot be discerned, yet.