Metal ratios and bulk disorder

The highest-efficiency devices in the literature all have in common a zinc-rich and copper-poor (Cu/[Zn + Sn]) ~ 0.80–0.85, Zn/Sn ~1.1–1.2) thin-film stoichiometry.^[Reference Katagiri, Jimbo, Tahara, Araki and Oishi^15] There have been several hypotheses for why this off-stoichiometry is required for high efficiency. First, the phase diagrams of CZTS and CZTSe indicate a narrow region of phase stability, where slight excesses of Cu, Zn, or Sn result in the co-precipitation of CZTS(e) and its secondary phases such as Cu₂S(e), ZnS(e), Cu₂SnS(e)₃, and others.^[Reference Chen, Walsh, Gong and Wei^14, Reference Vora, Blackburn, Repins, Beall, To, Pankow, Teeter, Young and Noufi^16, Reference Chen, Yang, Gong, Walsh and Wei^17] Many of these phases, including Cu₂S(e), SnS(e), and Cu₂SnS(e)₃, are conductive and can shunt the device when present in the film.^[Reference Vora, Blackburn, Repins, Beall, To, Pankow, Teeter, Young and Noufi^16] Approaching zinc-rich and copper-poor compositions favors the precipitation of wider-E _g ZnS(e), which is not expected to trap electrons or holes, at the expense of the more conductive phases, thereby preventing a shunt mechanism. Surprisingly, CZTSe devices appear highly tolerant to the presence of secondary ZnS(e) phases; Vora et al. showed that increasing the Zn/Sn ratio from 1.15 to 2.1 only decreased the efficiency of CZTSe devices from 8.1 to 6.4%.^[Reference Vora, Blackburn, Repins, Beall, To, Pankow, Teeter, Young and Noufi^16]

First-principles calculations reveal a complicated composition-dependent point defect landscape in CZTS and CZTSe (Fig. 1). Significant disorder of Cu and Zn atoms on the Cu/Zn sublattice is expected for all compositions due to the chemical similarity of these atoms.^[Reference Chen, Walsh, Gong and Wei^14, Reference Chen, Wang, Walsh, Gong and Wei^18] The Cu_Zn point defect would introduce an acceptor level, whereas the Zn_Cu defect would introduce a donor level; the [Cu_Zn + Zn_Cu] defect cluster is therefore a compensated defect pair. It is also found that the compensated defect pair [V_Cu + Zn_Cu] has relatively low formation energy and is therefore expected to form. Given that CZTS(e) is a p-type semiconductor, a net acceptor density is present amid the high degree of compensation, likely originating from a slight excess of either Cu_Zn or V_Cu. Experimental reports of the hole density in CZTS, CZTSe, and CZTSSe range from ~10¹⁵ to 10²⁰ cm⁻³, indicating a clear dependence of carrier concentration on sample preparation conditions and composition.^[Reference Nagaoka, Miyake, Taniyama, Kakimoto and Yoshino^19–Reference Gunawan, Virgus and Tai^21]

Figure 1.

(a) The calculated stable region of the phase diagram for CZTSe as a function of zinc and tin chemical potentials (fixed copper chemical potential). (b) The calculated formation energies of various point defects and defect complexes in CZTSe as a function of chemical potential (specified in (a), p-type conditions were assumed). (c) The ionization energies of various point defects in the band gap of CZTSe, where red lines represent acceptors, blue lines represent donors, and the initial and final charge states are indicated in parentheses. (d) Calculated band gap shifts induced by various defect clusters in CZTSe, where red and green lines indicate the valence and conduction bands, respectively. Reproduced with permission.^[Reference Chen, Walsh, Gong and Wei^14]

Experimental studies of the CZTS atomic structure have indeed confirmed the presence of significant atomic disorder in the Cu/Zn sublattice.^[Reference Schorr, Hoebler and Tovar^22–Reference Schorr and Tovar^24] Several studies have reported complete^[Reference Schorr, Hoebler and Tovar^22] or partial^[Reference Schorr and Tovar^24] disorder on the Cu/Zn sublattice by applying Rietvield analysis to neutron diffraction data. Another study directly observed regions of local Zn enrichment and Cu depletion via aberration-corrected scanning tunneling electron microscopy, with compositional fluctuations of the order of 1.5–5 nm reported.^[Reference Mendis, Shannon, Goodman, Major, Claridge, Halliday and Durose^23] Local Zn enrichment inherently implies a local segregation of donor defects, whereas local copper enrichment implies local segregation of acceptors. Thus, assuming these defects are ionized, the density of charged defects in the semiconductor fluctuates along with the compositional fluctuations. These charge fluctuations in turn cause electrostatic potential fluctuations of the conduction and valence bands, in a manner previously described by Shklovskii and Efros.^[Reference Shklovskii and Efros^25] Thus, compositional fluctuations of this type can cause band tailing in CZTSSe.^[Reference Gokmen, Gunawan, Todorov and Mitzi^26]

Low-temperature photoluminescence (PL) studies of CZTS materials and devices also support the presence of a large number of compensated donor–acceptor pairs in CZTS, of the order of 10¹⁹ cm⁻³,^[Reference Gokmen, Gunawan, Todorov and Mitzi^26, Reference Gershon, Shin, Bojarczuk, Gokmen, Lu and Guha^27] which is high enough to form electrostatic potential fluctuations.^[Reference Gershon, Shin, Bojarczuk, Gokmen, Lu and Guha^27, Reference Romero, Du, Teeter, Yan and Al-Jassim^28] Higher concentrations of donor–acceptor pairs, as measureable in PL and other studies, are correlated with reductions in V _oc and fill factor values in devices.^[Reference Gershon, Shin, Gokmen, Lu, Bojarczuk and Guha^29] Other room-temperature studies have detected the presence of significant band tailing in CZTSSe, likely caused by disorder, which is again correlated with reductions in the V _oc values of CZTSSe devices.^[Reference Gokmen, Gunawan, Todorov and Mitzi^26, Reference Miller, Warren, Gunawan, Gokmen, Mitzi and Cohen^30] These works examined both transient photocapacitance spectra^[Reference Miller, Warren, Gunawan, Gokmen, Mitzi and Cohen^30] and sub-band gap absorption as measured via device EQE (Fig. 2)^[Reference Gokmen, Gunawan, Todorov and Mitzi^26] and came to a similar conclusions that the presence of significant disorder in CZTSSe causes measureable band tailing, which in turn has a detrimental impact on V _oc. Recent work has even concluded that this band tailing may be severe enough to account for the entire observed V _oc deficit in CZTSSe.^[Reference Gokmen, Gunawan and Mitzi^31]

Figure 2.

Internal quantum efficiency measurements overlaid with PL spectra from (a) high-efficiency CIGSSe (E _g = 1.19 eV) and (b) CZTSSe (E _g = 1.13 eV). (c) schematic of band gap fluctuations and electrostatic potential fluctuations, both of which could give rise to band tailing in the absorber. Reproduced with permission.^[Reference Gokmen, Gunawan, Todorov and Mitzi^26]

The concentrations of various defect types appear to depend on sample stoichiometry. For example, Nagaoka et al. demonstrated control over the carrier concentration in single crystals of CZTS by changing the Cu:Zn ratio. A reduction of this ratio from 1.82 to 1.55 (i.e., a move toward copper-poor and zinc-rich conditions) led to a reduction in the carrier concentration by approximately one order of magnitude.^[Reference Nagaoka, Miyake, Taniyama, Kakimoto and Yoshino^19] This appears to indicate a reduced acceptor density under copper-poor conditions. A similar finding has been observed in CZTSSe thin films.^[Reference Gunawan, Virgus and Tai^21]

Furthermore, composition can also determine the likelihood of forming deep and detrimental defect clusters such as [Cu_Zn + Sn_Zn] and/or [2Cu_Zn + Sn_Zn].^[Reference Chen, Wang, Walsh, Gong and Wei^18] These are predicted to form more readily under zinc-poor conditions, which may help explain the better results obtained when the CZTSSe is prepared under zinc-rich conditions. Some studies also suggest that copper may preferentially occupy tin lattice sites as the copper content increases^[Reference Washio, Nozaki, Fukano, Motohiro, Jimbo and Katagiri^32]; the Cu_Sn point defect is expected to contribute to a detrimental deep level in CZTSSe. Thus, the requirement for copper-poor and zinc-rich conditions to achieve high-efficiency CZTSSe devices appears related to the suppression of both undesirable secondary phases and harmful point defects and disorder in CZTS. We note that significant disorder has been observed in CZTSSe for nearly all compositions and preparations methods reported to date. As discussed above, it is likely that the main cause of this disorder is the compensating [Cu_Zn + Zn_Cu] defect cluster, which is predicted to have low formation energy independent of the composition.^[Reference Chen, Walsh, Gong and Wei^14]

Sulfur and selenium

The sulfur-to-selenium ratio, or S/(S + Se), determines the band gap of CZTSSe, where the pure sulfide has a band gap of ~1.5 eV and the pure-selenide has a band gap of ~1.0 eV. The CZTSSe materials in the highest-efficiency devices reported to date have a selenium-rich composition and a band gap of ~1.13 eV.^[Reference Wang, Winkler, Gunawan, Gokmen, Todorov, Zhu and Mitzi^1] This band gap is lower than the 1.34 eV band gap predicted by Shockley–Queisser considerations to produce the highest-efficiency devices under AM 1.5 G illumination.^[Reference Shockley and Queisser^33, Reference Rau and Werner^34] However, the higher-sulfur CZTSSe compounds with more “optimal” band gaps typically display lower efficiencies and worse V _oc deficits. There are several likely explanations for the better performance observed from selenide materials.

First-principals calculations predict that both donor and acceptor defects are expected to produce shallower energy levels in the selenide materials, compared with the sulfides.^[Reference Chen, Walsh, Gong and Wei^14, Reference Chen, Wang, Walsh, Gong and Wei^18] Experimental admittance spectroscopy studies have shown that indeed the depth of the acceptor level depends on the sulfur-to-selenium content, with shallower levels observed for selenide materials [Fig. 3(a)].^[Reference Gunawan, Gokmen, Warren, Cohen, Todorov, Barkhouse, Bag, Tang, Shin and Mitzi^35] Another experimental study found that samples with higher sulfur content displayed not only deeper defects, but also an increase in bulk defect density, resulting in poorer carrier collection and higher rates of recombination.^[Reference Duan, Yang, Bob, Hsu, Lei and Yang^36] Increasing sulfur content has been experimentally correlated with more severe band tailing in the absorbers and larger V _oc deficits in devices [Fig. 3(b)].^[Reference Gokmen, Gunawan, Todorov and Mitzi^26] These phenomena are likely related in that deeper defects, and/or higher defect densities, are both expected to cause more severe band tailing and performance losses.

Figure 3.

(a) Increase in activation energy of the acceptor level and simultaneous decrease in the dielectric constant in CZTSSe as the band gap of the absorber (i.e., sulfur content) increases. Reprinted with permission.^[Reference Gunawan, Gokmen, Warren, Cohen, Todorov, Barkhouse, Bag, Tang, Shin and Mitzi^35] (b) Increase in the V _oc deficit as the band gap of the absorber (i.e., sulfur content) increases. Reproduced with permission.^[Reference Gokmen, Gunawan, Todorov and Mitzi^26]

Sulfide compounds are also expected to exhibit larger electron effective masses and smaller dielectric constants than the selenides.^[Reference Persson^37] The reduction in dielectric constant with increasing sulfur has also been observed experimentally [Fig. 3(a)].^[Reference Gunawan, Gokmen, Warren, Cohen, Todorov, Barkhouse, Bag, Tang, Shin and Mitzi^35] We note that a reduced dielectric constant would amplify all Coulombic (electrostatic) interactions because of reduced screening and hence may partially account for larger activation energies and more severe band tailing for sulfide compounds. The reduced screening of charged point defects in the material would also impact carrier mobility and recombination. This, coupled with a larger electron effective mass,^[Reference Persson^37] would result in poorer carrier transport through sulfide materials, as compared with selenides. A recent study examined minority carrier diffusion lengths (L _d) in CZTSSe materials with varying sulfur to selenium compositions and indeed found that this value decreased with increasing sulfur content (Fig. 4).^[Reference Gokmen, Gunawan and Mitzi^38]

Figure 4.

Minority carrier diffusion lengths in CZTSSe and CIGSSe materials of varying band gaps. The shaded region is a guide for the eye. Sample Z0 is a recent pure-selenide CZTSe sample, prepared by co-evaporation,^[Reference Lee, Gershon, Gunawan, Todorov, Shin, Virgus and Guha^2] whereas the rest of the samples were reported in Ref. Reference Gokmen, Gunawan and Mitzi38. Reprinted with permission.

Devices with high-sulfur CZTSSe have also been observed to have higher series resistance values than those with higher selenium content,^[Reference Gunawan, Gokmen and Mitzi^39–Reference Shin, Gunawan, Zhu, Bojarczuk, Chey and Guha^41] indicating that sulfur reduces bulk conductivity, mainly due to reduced carrier density^[Reference Gunawan, Virgus and Tai^21] in the absorber, and makes the back contact less Ohmic. Note that even though a larger band gap tends to increase the V _oc and thus the FF (see, e.g., empirical relationship of V _oc and FF in Ref. Reference Green42) the monotonic increase in series resistance at higher band gap eventually degrades the FF, resulting in a maximum FF value when the CZTSSe band gaps is ~1.15 eV. This band-gap-dependent FF profile leads to the observed empirical optimum efficiency at ~1.15 eV.^[Reference Wang, Winkler, Gunawan, Gokmen, Todorov, Zhu and Mitzi^1, Reference Todorov, Tang, Bag, Gunawan, Gokmen, Zhu and Mitzi^43]

Finally, both sulfur and selenium are volatile at temperatures commonly used to prepare CZTSSe absorbers, with sulfur being the more volatile of the two chalcogen species. In many chalcogenide materials, chalcogen deficiency results in deep states, which are harmful to performance. Based simply on the higher vapor pressure of sulfur compared with selenium, one might expect that sulfur deficiency would be more difficult to avoid than selenium deficiency. Additionally, first-principles calculations suggest that sulfur vacancies have lower formation energies than selenium vacancies.^[Reference Chen, Walsh, Gong and Wei^14] Thus, high-sulfur CZTSSe materials are expected to suffer more from the presence of deep recombination centers than are high-selenium materials.

Surface structure

Studies of the CZTSSe surface employing x-ray photoelectron spectroscopy (XPS) and other surface-sensitive techniques have found it to be highly depleted of copper, with respect to the bulk.^[Reference Haight, Shao, Wang and Mitzi^44, Reference Bär, Schubert, Marsen, Krause, Pookpanratana, Unold, Weinhardt, Heske and Schock^45] First-principles calculations find that charge-compensated point defects such as vacancies (V_Cu, V_Zn), antisites (Zn_Cu, Zn_Sn, Sn_Zn), and defect clusters (Cu_Zn + Cu_Sn, 2Zn_Cu + V_Sn) are required to stabilize the polar (112) surface, which is the preferred orientation of CZTSSe^[Reference Xu, Chen, Huang, Xiang, Gong and Wei^46] (Fig. 5). In copper-poor, zinc-rich samples, the favored charge-stabilizing defects are those that are free from copper. The most prevalent point defects are predicted to be 2V_Cu, 2Zn_Cu + 2V_Sn and Zn_Sn. These defects are also found not to contribute harmful states in the band gap. These calculations are consistent with what is experimentally observed of the CZTSSe surface, which indeed displays a copper-depleted character.

Figure 5.

Defect formation energies on the (112) (top) and (112) (bottom) as a function of chemical potential (i.e., Cu rich, poor and Zn rich, poor). Reprinted with permission from Peng et al..^[Reference Xu, Chen, Huang, Xiang, Gong and Wei^46] Copyright (2013) by the American Physical Society.

Copper-depletion of the CuInSe₂ (CIS) surface has also been widely reported and studied.^[Reference Liao and Rockett^47, Reference Schmid, Ruckh, Grunwald and Schock^48] In this system, it is also found that charged surface defects stabilize the (112) surface.^[Reference Jaffe and Zunger^49] It has been reported that the so-called “ordered vacancy compound” (OVC) that spontaneously forms on the CIS surface has a much lower copper concentration (CuIn₃Se₅), is n-type, has a slightly larger band gap (1.3 eV, compared with 1.04 eV), and is responsible for forming a p–n junction with the CIS layer below.^[Reference Schmid, Ruckh, Grunwald and Schock^48] Upon deposition of the buffer, the junction becomes CIS/OVC/CdS. Furthermore, Cd may diffuse into the OVC layer thereby further doping it n-type and improving the junction characteristics,^[Reference Liao and Rockett^50] as cadmium is a known donor defect in CIS.^[Reference Migliorato, Shay, Kasper and Wagner^51]

An open question is: to what extent does the copper-deficient surface observed in CZTSSe resemble that observed in CIS and CIGS? If Cd inter-diffuses with the CZTSSe surface during chemical bath deposition, which sites does it occupy and what would the junction look like? One study examined these questions using first-principles analysis and found that substitution of cadmium onto Cu and Zn sites is more favorable than on Sn sites,^[Reference Maeda, Nakamura and Wada^52] where Cd_Cu is a donor defect and Cd_Zn is neutral. More favorable still is the compensated defect complex (Cd_Cu + V_Cu). This defect pair is expected to dominate in equilibrium, with small amounts of Cd_Cu and Cd_Zn also present. If this were the case, the degree of n-type doping of the copper-poor CZTSSe surface would be limited compared with the CIS case; however, real experimental preparation conditions may not kinetically allow equilibrium to be reached. Non-equilibrium could therefore allow n-type doping of the surface region, in a manner similar to what is observed in CIS. These phenomena warrant further investigation to fully understand this interface and how it evolves during chemical bath deposition of the buffer.

Another notable aspect of the CZTSSe surface is that the CZTSSe bands are found to bend upwards, indicating a condition of accumulation.^[Reference Haight, Shao, Wang and Mitzi^44] Such upward band bending indicates that excess negative charge is trapped at or near the surface. This was determined using femtosecond pump (1.55 eV photon energy)/probe (26.35 eV) ultraviolet photoelectron spectroscopy (fs-UPS), where spectra are recorded both in equilibrium and flat-band (photoexcited) conditions and the spectra are compared with determine the direction of band bending (Fig. 6). Photoexcitation with an intense pulse of 800 nm light produces a dense electron–hole plasma that flattens the CZTSSe bands near the surface. Under photoexcitation, the bands shift to higher binding energy (red) indicating that under equilibrium the bands are bent up at the surface. An accumulated surface has been noted for CZTSSe samples with or without surface oxide layers removed, indicating a fundamental property of the CZTSSe surface. Haight et al. attribute the accumulation at the surface to the large copper deficiency in this region, where the copper vacancy is expected to act as a shallow acceptor in CZTSSe.

Figure 6.

Femtosecond UPS spectra of air-annealed CZTSSe under equilibrium (unpumped, black) and flat band (pumped, red) conditions. The arrows indicate the shift of the spectrum following photoexcitation. The bands bend upwards, indicating a condition of accumulation at the surface. Reproduced with permission.^[Reference Haight, Shao, Wang and Mitzi^44]

Air annealing of CZTSSe films prior to chemical bath deposition of CdS buffers has been shown to dramatically improve both absorber PL and device performance. Following air anneal, strong sodium and oxygen signals are detected by XPS,^[Reference Haight, Shao, Wang and Mitzi^44] indicating both the formation of a surface oxide and diffusion of Na to the surface. Even following NH₄OH etching to remove oxygen and Na at the surface, secondary ion mass spectrometry (SIMS) reveals that the sodium and oxygen signals are present throughout the CZTSSe film thickness. However, the most likely location for most of the sodium and oxygen species detected is at grain surfaces rather than the grain interior.^[Reference Niles, Al-Jassim and Ramanathan^53] The presence of sodium and oxygen at the grain surfaces in high-efficiency CIS, CIGS, and CZTSSe has been correlated with high device performance. One explanation for this observed relationship is that Na may aid in the oxidation of grain boundaries, i.e., the filling of chalcogen vacancies with oxygen, which could help passivate deep defects along these boundaries.^[Reference Haight, Shao, Wang and Mitzi^44, Reference Kronik, Cahen and Schock^54] Similar observations on the efficacy of air anneals in CIS-based devices are consistent with our understanding of diffusion of O and Na along grain boundaries and their effect on grain boundary passivation.^[Reference Haight, Shao, Wang and Mitzi^44, Reference Cahen and Noufi^55]

Alkali impurities

Much like in CIGS materials, sodium and other alkali species are found to be important for achieving high-efficiency CZTSSe devices. In particular, sodium is found to increase grain size,^[Reference Rockett^56, Reference Oo, Johnson, Bhatia, Lund, Nowell and Scarpulla^57] increase carrier concentration and/or mobility,^[Reference Prabhakar and Jampana^58–Reference Nagaoka, Miyake, Taniyama, Kakimoto, Nose, Scarpulla and Yoshino^62] passivate deep defects,^[Reference Kronik, Cahen and Schock^54, Reference Erslev, Lee, Shafarman and Cohen^63, Reference Gershon, Shin, Bojarczuk, Hopstaken, Mitzi and Guha^64] and improve the fill factors and open-circuit voltages of CZTSSe devices.^[Reference Erslev, Lee, Shafarman and Cohen^63, Reference Li, Kuciauskas, Young and Repins^65, Reference Nakada, Iga, Ohbo and Kunioka^66] It has been found that potassium behaves in a manner similar to sodium in CZTS.^[Reference Johnson, Baryshev, Thimsen, Manno, Zhang, Veryovkin, Leighton and Aydil^67]

Surface compositional studies of high-efficiency CZTSSe absorbers have found that sodium is present in large concentrations at surfaces and grain boundaries in CZTSSe.^[Reference Haight, Shao, Wang and Mitzi^44, Reference Niles, Al-Jassim and Ramanathan^53] First-principles studies suggest while grain boundaries in CIGS are fundamentally electrically benign,^[Reference Yan, Jiang, Noufi, Wei, Moutinho and Al-Jassim^68] unpassivated grain boundaries in CZTSe contribute deep states which are expected to act as non-radiative recombination centers.^[Reference Li, Mitzi and Shenoy^69] Kelvin-probe studies have found potential “spikes” at grain boundaries in CZTSSe, which have been proposed to behave as hole-repelling and electron-attracting regions, aiding the separation and transport of the minority electrons.^[Reference Li, Chawla and Clemens^70] However, these spikes may not exist in materials that do not contain sodium.^[Reference Li, Chawla and Clemens^70] These results seem to suggest that sodium segregates to grain boundaries, and once there may be important for suppressing non-radiative recombination. This was further supported in a PL study, which examined the total PL intensity in CZTS as a function of sodium, and as a function of temperature.^[Reference Gershon, Shin, Bojarczuk, Hopstaken, Mitzi and Guha^64] It was found that at all temperatures, samples with less sodium luminesced less than those with more sodium. Furthermore, as the temperature increased from ~4 K toward room temperature, the PL of samples with little or no sodium dropped at a much higher rate than samples with more sodium, indicating that when carriers had enough thermal energy to travel longer distances, they tended to reach non-radiative recombination centers, likely along grain boundaries [Fig. 7(f)]. Once sufficient sodium was present in the films, samples with vastly different grain sizes (and thus boundary densities) displayed identical radiative recombination profiles, indicating that the grain boundaries had become passive within the detection limits of the study.^[Reference Gershon, Shin, Bojarczuk, Hopstaken, Mitzi and Guha^64]

Figure 7.

SEM cross-sectional images of CZTS thin films where (a) 0 nm, (b) 1 nm, (c) 4.5 nm, and (d) 23 nm of NaF was evaporated onto the substrate prior to CZTS growth and annealing. (e) SIMS depth profile of zinc distribution in the samples in (a-d). (f) Integrated PL intensity of the samples in (a–d) as a function of temperature. Reproduced with permission.^[Reference Gershon, Shin, Bojarczuk, Hopstaken, Mitzi and Guha^64]

In addition to passivating grain boundaries, alkali species significantly improve CZTSSe microstructure by enhancing grain size^[Reference Rockett^56, Reference Oo, Johnson, Bhatia, Lund, Nowell and Scarpulla^57, Reference Gershon, Shin, Bojarczuk, Hopstaken, Mitzi and Guha^64, Reference Johnson, Baryshev, Thimsen, Manno, Zhang, Veryovkin, Leighton and Aydil^67, Reference Sutter-Fella, Stu"ckelberger, Hagendorfer, La Mattina, Kranz, Nishiwaki, Uhl, Romanyuk and Tiwari:^71] [Figs. 7(a)–7(d)]. It is believed that this microstructural improvement is related to the formation of a liquid sodium–chalcogen-based phase at the grain boundaries. Enhanced diffusion through a phase such as this is especially important in zinc-rich and copper-poor materials, where ZnS(e) must be expelled from growing CZTSSe grains and may drag on advancing grain boundaries, prevent grain growth, if not removed from these boundaries.^[Reference Gershon, Shin, Bojarczuk, Hopstaken, Mitzi and Guha^64, Reference Sutter-Fella, Stu"ckelberger, Hagendorfer, La Mattina, Kranz, Nishiwaki, Uhl, Romanyuk and Tiwari:^71] It has been shown that the distribution of zinc throughout CZTS thin films is heavily influenced by the concentration of sodium present in the absorber during the annealing stage [Fig. 7(e)]. A large average grain size in CZTS was correlated with a reduction in zinc content through the film bulk and simultaneously the accumulation of zinc sulfide toward the back interface near the Mo contact. The phase diagram of the Zn–Na–S system indicates a eutectic temperature of roughly 610 °C.^[Reference Tomashyk, Feychuk and Shcherbak^72] This may help explain why annealing at such high temperatures is required to achieve large grains in CZTSSe.^[Reference Shin, Zhu, Bojarczuk, Chey and Guha^73] Another study showed that even CZTS films prepared on alkali-free substrates could display these large grain size improvements when soda-lime glass substrates were placed in the same annealing ampoule, implying that alkali species may volatilize out of the glass during the anneal and impact the microstructure.^[Reference Johnson, Baryshev, Thimsen, Manno, Zhang, Veryovkin, Leighton and Aydil^67]

Sodium may also incorporate directly onto the CZTS lattice. Studies on single-crystals have found an increase in the unit-cell size upon incorporation of sodium as measured by x-ray diffraction,^[Reference Nagaoka, Miyake, Taniyama, Kakimoto, Nose, Scarpulla and Yoshino^62] indicating the incorporation of sodium into the structure. This finding was also correlated with an increase in both the carrier concentration and mobility of CZTS. The authors attributed this to a reduction in the acceptor activation energy and a reduction in point defect compensation in the CZTS structure. Enhanced fill factors and V _oc values of CZTSe devices have been attributed to improvements in carrier concentration and mobility seen upon sodium incorporation into CZTSe.^[Reference Li, Kuciauskas, Young and Repins^65]

Other substitutional elements in CZTS

Although it is possible to change the band gap in CZTSSe by varying the sulfur to selenium content, it appears that sulfur may have other detrimental impacts on device performance (see above). Other chemical variations to CZTSSe can also induce band gap changes. First-principles calculations find that Ge and Si may be alloyed into the CZTSSe structure, substituting onto the tin site, to increase the band gap [Fig. 8(a)].^[Reference Shu, Yang, Chen, Huang, Xiang, Gong and Wei^74] It was found that Ge is highly miscible into CZTSe with the Cu₂Zn(Ge,Sn)Se₄ band gap ranging from 1.0 eV for the pure-tin compound and 1.5 eV for the pure-Ge compound. Band gap tunability has been demonstrated experimentally in the Cu₂Zn(Ge,Sn)S₄ system, where Ford et al. prepared films with band gaps ranging from 1.5 eV (pure tin Cu₂ZnSnS₄) to 1.94 eV (pure Ge Cu₂ZnGeS₄), with various values in between [Fig. 8(c)].^[Reference Ford, Guo, Agrawal and Hillhouse^75] The magnitude of this shift is consistent with theoretical predictions.^[Reference Shu, Yang, Chen, Huang, Xiang, Gong and Wei^74] The same group later demonstrated efficiencies as high as 8.4% (without anti-reflection coating) using a partially Ge substituted kesterite material.^[Reference Guo, Ford, Yang, Hages, Hillhouse and Agrawal^76] In the same year, Bag et al. reported an efficiency of 9.1% by substituting 40% Ge into CZTSe.^[Reference Bag, Gunawan, Gokmen, Zhu and Mitzi^77] This group demonstrated effective Ge incorporation by monitoring the x-ray diffraction peak, which shifts to larger 2-theta with increasing Ge due to the smaller atomic size of Ge as compared with Sn. Silicon is less miscible in CZTSe, but offers a larger range of band gap tuning up to 2.4 eV for the Cu₂ZnSiSe₄ compound. In both cases (substitutional Ge or Si), the band gap change is expected to stem largely from a conduction band shift [Fig. 8(b)].

Figure 8.

(a) Calculated dependence of the band gap of CZTSe as a function of Ge (top) and Si (bottom) substitution for Sn. (b) Predicted conduction and valence band positions of CZSiSe and CZGeSe, compared to CZTSe. (a) and (b) are reprinted with permission from Shu et al..^[Reference Shu, Yang, Chen, Huang, Xiang, Gong and Wei^74] Copyright (2013) by the American Physical Society. (c) Experimental observation of band gap changes in CZTS as Ge is increasingly substituted for Sn. Reproduced with permission.^[Reference Ford, Guo, Agrawal and Hillhouse^75]

Furthermore, it was recently found that compounds in the I₂–II–IV–VI₄ family are generally stable when I = Cu, Ag, II = Zn, Cd, Hg, IV = Si, Ge, Sn, and VI = S, Se, and/or Te. Instability arises when II = Mg, Ca, Sr, Ba, when IV = Ti, Zr, Hf, or when IV = O due to competition from the related binary and ternary compounds.^[Reference Wang, Chen, Yang, Lang, Xiang, Gong, Walsh and Wei^78] The authors noted that phase separation is expected predominantly when the secondary phases had non-tetrahedral structures. This study opens up many possibilities to find a variant of CZTSSe that may have preferable optoelectronic properties and a lower propensity to self-compensate. Many of the possible substitutions mentioned above have never been reported experimentally, although they are promising materials to study in the future.

Prospectives

Over the past decade, our understanding of the factors impacting the material properties of CZTSSe, and consequently device performance, has improved dramatically. Through careful materials and device characterization, several key phenomena have been identified as likely culprits for the observed V _oc deficit in CZTSSe devices. Among these are the severity of the point defects and disorder within the CZTSSe bulk, which exceeds that observed in CIGS materials of the same band gap. It is believed that the higher degree of disorder is due to the chemical similarity of Cu and Zn atoms, and the low energetic barrier to the exchange of lattice positions amongst these two atoms. The disorder on the Cu/Zn sublattice may be responsible for the significant band tailing observed in the CZTSSe system, and therefore at least part of the V _oc deficit. It is possible that this disorder could be mitigated by an external stimulus, e.g., strain. Alternatively, recent theoretical calculations have indicated that the CZTSSe structure would be stable to various substitutions on the I, II, IV, and VI sites. By replacing either Cu, Zn, or both, a new I₂–II–IV–VI₄ compound may be identified that is less susceptible to the kind of disorder that causes detrimental band tailing and thus V _oc loss in the CZTSSe system.

Additionally, it has been proposed that disorder in CZTSSe may be reduced through low-temperature annealing for extended periods of time.^[Reference Scragg, Choubrac, Lafond, Ericson and Platzer-Björkman^79] A recent study documented improvements in an “ordering parameter” of CZTS materials when samples were annealed below ~250 °C, as measured via Raman spectroscopy. It was found that these low-temperature anneals reduced the full-width at half-maximum of the main CZTS Raman peak centered at ~338 cm⁻¹ (Fig. 9). The authors attributed the observed narrowing of the Raman peak to a reduction in the disorder of the bulk CZTS, i.e., fewer Cu–Zn antisites. The effect was found to be reversible, in that the Raman peak width of samples annealed above this temperature widened again, after initially decreasing during the low-temperature anneal. If it is indeed possible to reduce the disorder in CZTS using long low-temperature anneals without otherwise harming other material properties, this could offer a valuable means of reducing the V _oc deficit in CZTS devices. To date, there have been no published reports of device improvements using this technique.

Figure 9.

(top) Variation in “Q” (defined as an ordering parameter) for CZTS films annealed at different temperatures, for different durations. (bottom) Full-width at half maximum of the main Raman mode in CZTS samples annealed for 1 h at various temperatures. Shaded region indicates an “order/disorder” temperature. Reproduced with permission.^[Reference Scragg, Choubrac, Lafond, Ericson and Platzer-Björkman^79]

We have also seen experimental and theoretical evidence that moving away from sulfur and toward a pure-selenide material improves the overall bulk properties of the absorber. This observation has recently manifested in a record V _oc deficit of 578 mV in a pure-selenide CZTSe device.^[Reference Lee, Gershon, Gunawan, Todorov, Shin, Virgus and Guha^2] Although there seems to be a limit to the selenium content accessible via solution-processing due to solubility constraints, it is possible to prepare pure-selenide materials using vapor phase methods such as sputtering and co-evaporation. Unfortunately, the band gap of the pure CZTSe material (1.0 eV) is lower than the optimal value for high-efficiency single-absorber devices based on the Shockley–Queisser limit model.^[Reference Shockley and Queisser^33, Reference Rau and Werner^34] If a method is found to increase the band gap of CZTSe without compromising its optoelectronic properties, we could see performance improvements above the 12.6% efficiency benchmark that has already been demonstrated. Alternatively, if methods are found to increase the dielectric constant of the absorber or reduce the electron effective mass, i.e., induce the opposite effect to what is observed with increasing sulfur, then device performance is also expected to increase.

Recent work has offered important insights into the chemical and electronic properties of the CZTSSe surface. Many similarities have been found between the CZTSSe surface and that of CIS; however, the interface chemistry that occurs during junction formation is still not fully understood. Additional work is needed to elucidate what if any losses in performance are caused by an unoptimized interface between CZTSSe and its buffer layer. It has been shown that formation of a double-emitter comprised CdS/In₂S₃ improves the junction characteristics such that CZTSSe devices behave more like CIGS devices, in terms of temperature-dependent V _oc and Suns-V _oc. This indicates that the CZTSSe/buffer interface may not yet be fully optimized and that further efficiency improvements may be possible as we learn more about the chemical and electrical properties of the CZTSSe surface and the junctions it forms with buffer materials.

Some optimizations of the device layer thicknesses may also provide additional benefits. The CZTSSe layer in the highest-efficiency CZTSSe devices is typically 2–2.5 µm thick. This layer thickness is typically chosen because the hydrazine solution process used to prepare CZTSSe often results in voids toward the back of the absorber and thus multiple layers and a thick coating appear necessary to prevent detrimental device shunting through pinholes. Recent work has found that the minority carrier diffusion length in our high-efficiency CZTSSe layers (E _g ~ 1.15 eV) is closer to ~1.2 µm or less,^[Reference Gokmen, Gunawan and Mitzi^38] indicating that much of the CZTSSe thickness is inactive and/or possibly contributing to series resistance in the device. An optimized device might therefore employ films with thicknesses closer to the minority carrier diffusion length if methods can be found to suppress void formation toward the back contact and thereby avoid pinholes and shunt paths in the films.

Finally, high-efficiency CIGS devices typically contain a graded band gap structure, where the band gap increases slightly toward the front and back of the absorber layer to minimize recombination. Although this is generally acknowledged to be a promising route toward achieving improved device performance in CZTSSe,^[Reference Mitzi, Gunawan, Todorov and Barkhouse^4] no experimental studies demonstrating this effect have been reported. As discussed above, surface alloying with Ge, Si, and/or sulfur would allow a slight increase in the band gap at the CZTSSe surface.