Secondary Crystalline Phases Influence on Optical Properties in Off-Stoichiometric Cu2S–ZnS–SnS2 Thin Films

Cu2ZnSnS4 (CZTS) is an economically and environmentally friendly alternative to other toxic and expensive materials used for photovoltaics, however, the variation in the composition during synthesis is often followed by the occurrence of the secondary binary and ternary crystalline phases. These phases produce changes in the optical absorption edge important in cell efficiency. We explore here the secondary phases that emerge in a combinatorial Cu2S–ZnS–SnS2 thin films library. Thin films with a composition gradient were prepared by simultaneous magnetron sputtering from three binary chalcogenide targets (Cu2S, SnS2 and ZnS). Then, the samples were crystallized by sulfurization annealing at 450 °C under argon flow. Their composition was measured by energy dispersive X-ray spectroscopy (EDX), whereas the structural and optical properties were investigated by grazing incidence X-ray diffraction (GIXRD), Raman spectroscopy and optical transmission measurements. As already known, we found that annealing in a sulfur environment is beneficial, increasing the crystallinity of the samples. Raman spectroscopy revealed the presence of CZTS in all the samples from the library. Secondary crystalline phases such as SnS2, ZnS and Cu–S are also formed in the samples depending on their proximity to the binary chalcogenide targets. The formation of ZnS or Cu–S strongly correlates with the Zn/Sn and Cu/Zn ratio of the total sample composition. The presence of these phases produces a variation in the bandgap between 1.41 eV and 1.68 eV. This study reveals that as we go further away from CZTS in the composition space, in the quasi-ternary Cu2S–ZnS–SnS2 diagram, secondary crystalline phases arise and increase in number, whereas the bandgap takes values outside the optimum range for photovoltaic applications.


Introduction
The world must meet the requirement for clean electricity (without carbon emissions) generation of 30 terawatts by 2050, associated with the expected increase in global energy demand [1]. Solar photovoltaic systems have great potential to address the challenge of future clean electricity supply on a large scale [2]. The quaternary semiconductor Cu 2 ZnSnS 4 (CZTS) has gained wide attention and has been intensively investigated as a new generation photovoltaic (PV) absorber material. CZTS is theoretically derived from the CuInS 2 (CIS) structure, where in the formula unit of

Materials and Methods
The library was synthesized on nine SiO 2 substrates that were cleaned by successive sonication in different liquids (acetone (Chemical Company, Iasi, Romania), ethanol (Chemical Company, Iasi, Romania), deionized water (prepared using a Thermo Scientific Smart2Putre UV water treatment system, Hatvan Hungary)), dried in a nitrogen flow (Linde, Bucharest Romania) and placed next to each other in the deposition equipment as in Figure 1a. Cu-Zn-Sn-S thin films were deposited by RF magnetron co-sputtering (Gencoa Ltd., Liverpool, UK) from Cu 2 S, ZnS and SnS 2 binary chalcogenide targets (99.99% purity, Mateck GmbH, Jülich" Germany, 2 inch in diameter). The magnetron sputtering system is a custom-built setup that consists of a cylindrical deposition chamber (Excel Instruments, Maharashtra, India) with hemispherical up and down caps (Excel Instruments, Maharashtra, India). Three magnetrons are equidistantly placed on the bottom hemisphere. The substrates are placed on a holder in the upper part of the chamber. The distance between the center of the P5 sample and each target was of 11 cm, whereas the rest of the samples were closer to at least one of the targets, resulting in a continuous variation of composition. The angle between the targets and the substrates was 45 degrees. After initially evacuating the chamber at 10 −6 Torr, Ar gas (Linde, Bucharest Romania) was introduced at a rate of 30 sccm and the pressure inside the chamber, during deposition, was maintained constant at 5 × 10 −3 Torr. A few minutes of pre-sputtering was performed in order to remove any unwanted contaminants from the target surfaces prior to the deposition process. The sputtering power was set at 80, 50 and 20 W for Cu 2 S, ZnS and SnS 2, respectively, leading to a sputtering rate of 0.5 Å/s for each material. The sputtering rates were optimized using an Inficon Q-bridge monitoring software (Bad Ragaz, Switzerland) connected to a quartz microcrystal (Inficon, Bad Ragaz, Switzerland). The deposition time was approximately 35 min, in order to obtain thin films with a thickness of 300 nm. The substrates were not heated during deposition. binary targets (circles) and glass substrates (squares) during the magnetron co-sputtering deposition of the Cu 2 S-ZnS-SnS 2 thin films; and (b) the average chemical composition calculated from EDX showed on the Cu 2 S-ZnS-SnS 2 quasi-ternary diagram. In addition, Cu 2 ZnSnS 4 (CZTS) is also showed. Sulfurization annealing was performed by placing the samples in a quartz tube, inserted in a tubular furnace, at 450 • C for 1 h. A continuous flow of 83 sccm Argon was used to transport the sulfur vapors obtained from the evaporation of an upstream sulfur powder. After annealing, the furnace was turned off and the samples were cooled to room temperature in Ar flow in order to avoid oxidation.
The determination of the elemental concentration in the films was carried out by means of energy dispersive X-ray (EDX) spectroscopy using a Zeiss EVO 50 XVP scanning electron microscope (Carl Zeiss, Oberkochen, Germany) equipped with a Bruker Quantax 200 detector (Bruker AXS Microanalysis GmbH, Berlin, Germany).
The investigation of the Cu 2 S-ZnS-SnS 2 thin films structure was performed by grazing incidence X-ray diffraction (GIXRD) at an incidence angle of 0.3 • with a Rigaku SmartLab diffractometer (Rigaku, Tokyo, Japan) provided with Cu K α radiation (λ = 1.54178 Å) and HyPix-3000 2D Hybrid Pixel Array Detector (Rigaku, Tokyo, Japan) (in 0 D mode). The identification of the crystalline phases was performed using the DIFFRAC.SUITE Software package (Bruker, Billerica, USA).
Raman spectra were recorded at room temperature, in the 200-450 cm −1 range, in backscattering configuration, with a LabRAM HR Evolution spectrometer (Horiba Jobin-Yvon, Palaiseau, France) equipped with a confocal microscope. A He-Ne laser (Horiba Jobin-Yvon, Palaiseau, France) operating at 633 nm was focused using an Olympus 100× objective (Olympus, Tokyo, Japan) on the surface of the samples. Accurate and automated calibration was performed on a standard Si wafer (provided by Horiba Jobin-Yvon, Palaiseau, France) by checking the Rayleigh and Raman signals. The laser excitation power was adjusted to avoid laser-induced heating in the thin films.

Chemical Composition
The average chemical compositions of the Cu 2 S-ZnS-SnS 2 thin films after sulfurization annealing, obtained from EDX, are shown in Table 1. These values (in the limit of EDX measurement errors) show that the sulfurization does not produce major changes of sulfur content in annealed samples as compared with the as-deposited samples (data not shown). Its role is to avoid sulfur evaporation during annealing, given that sulfur is highly volatile. Starting from the average Cu, Zn and Sn percentages, we calculated the corresponding average percentages of Cu 2 S, ZnS and SnS 2 in each sample and represented these values in a Cu 2 S-ZnS-SnS 2 quasi-ternary diagram (Figure 1b). In this diagram, it was easily seen that the average chemical compositions of the P5, P6, P7, P8 and P9 samples were the closest to the stoichiometric CZTS ratio (Cu = 25%, Zn = 12.5%, Sn = 12.5% and S = 50%), the ones of P2, P3 and P4 are at intermediate distance, whereas the one of P1 is further away. We note that the elemental composition of the films varies according to the binary target proximity ( Figure 1a). However, during magnetron co-sputtering deposition, some amount of sulfur from the targets no longer reached the glass substrate. Thus, if it was computed, the corresponding average percentage of sulfur (p ideal S ) necessary to bind with the entire quantity of Cu, Zn and Sn in the sample (to form Cu 2 S, ZnS and SnS 2 ), results that for most of the samples, p ideal S is higher than the average percentage of sulfur (p S ) found by EDX, except for P6, where we have an excess of S. These differences can be seen in the last row of Table 1 ( ) are also shown in Table 1.
In Figure 1b, we can see the labels of the elemental ratios Cu/(Zn+Sn) and Zn/Sn for each sample. If Cu/(Zn+Sn) is greater than 1, the material is considered Cu-rich, otherwise Cu-poor. The same applies for the Zn/Sn ratio, resulting in Zn-poor or Zn-rich compositions. These ratios together with the Cu/Zn ratio, are good indicators of the secondary phases formed after annealing. For instance, a high concentration of copper, when Cu/Zn > 1.5, as in the case of P3, P6, P8 and P9, leads to the formation of secondary Cu-S crystalline phases, whereas a high Zn/Sn ratio is a sign for the formation of ZnS crystalline phases, as seen in P9.

GIXRD Characterization
The GIXRD diagrams of the thin films in the as-deposited state ( Figure 2) show that they are already crystallized to some extent (the percent of initial crystallization p i , is shown in Table 2). The percent of crystalline phases increases in the samples after the sulfurization annealing (p a in Table 2), as calculated from the GIXRD diagrams in Figure 2. The percent of crystalline phases before and after annealing (p i or p a ) was calculated as (A C /A C+Am )×100, where A C is the total area of crystalline peaks (from all the crystalline phases) and A C+Am is the total area of the peaks (crystalline + amorphous). X-ray diffraction patterns of the nine Cu 2 S-ZnS-SnS 2 thin films in the as-deposited state (gray curves) and after sulfurization annealing at 450 • C for one hour (black curves). The assignment of phases and orientations was done using the International Centre for Diffraction Data (ICDD) database (see Table 3). h = hexagonal; t = tetragonal; r = rhombohedral. The Powder Diffraction File (PDF) for each crystalline phase is also indicated.
It can be observed that, between 14 • and 40 • in 2θ, there is a broad peak (centered at 24.57 • ), which is given, mainly, by the silicate glass substrate. When processing the diffraction curves, to calculate the percent of crystalline phases (p i and p a ), the signal from the substrate was subtracted.
We observed by analyzing Table 2 and Figure 1b, that the P1 sample (the richest in Sn) is the least crystallized in the as-deposited state, however, it has the highest crystallization rate after annealing.
The samples richest in Zn (P8, P9) are the most crystallized in the as-deposited state, and as a consequence they have the lowest crystallization rate after annealing. The same situation was found for the sample richest in Cu (P3). Table 2. The percent of crystalline phases in each sample (p i -as-deposited; p a -annealed) and the ratio (r = p a /p i ). In Table 3, the majority and the minority phases (italic) and the mean dimension of the crystallites, after the sulfurization annealing of the Cu 2 S-ZnS-SnS 2 samples, were listed. The percent of the amorphous phase (p amorphous ) was calculated as p amorphous = 100 − p a . For a given sample, the percent of the "X" crystalline phase was given by the ratio p a ×A X /A C , where A X is the total area of crystalline peaks of phase "X". Moreover, the position and the half width maximum of the most intense diffraction peaks are compared in Figure 3. Thus, we can observe that in all the samples, the peak position of the majority phase does not match with the Powder Diffraction File (PDF) 00-026-0575 of the CZTS phase. The phases mentioned in Table 3 are those with the smallest deviation from the peaks' positions found in the International Centre for Diffraction Data (ICDD) database. The most intense peaks (at~28,54 • , 47.5 • ,~56.3 • ) are very close to the CZTS positions, so it is very probable that CZTS exists as a minority phase (see Raman measurements). The CZTS phase could not be detected by GIXRD because the peaks are broad and a deconvolution was not possible. The majority crystalline phase is most abundantly formed in P5, the sample with the chemical composition closest to ideal CZTS. In contrast to P5, in P8 the majority phase is the least formed (only 45 %), even if its composition is quite close to that of CZTS ( Figure 1b). It has the highest percent of crystalline phases (p i ) in the as-deposited state and after annealing this percent remains the same. The average size of the crystallites increases after annealing and has the highest value among all the annealed samples (52 nm, see Table 3). The average size of the crystallites was approximated using the Scherrer equation, d = Kλ/βcos(θ), where K is the Scherrer constant, λ the wavelength, θ the angle and β the peak broadening. The instrumental line broadening and K α2 were subtracted a priori. Table 3. The majority and minority phases after the sulfurization annealing of the Cu 2 S-ZnS-SnS 2 samples. The PDF file number, the lattice system, and the space group were indicated for each crystalline phase. In addition, the sample chemical composition (Cc), the area of the peak at 2θ~28.540 • (A) and the average size of the crystallites of the majority phase (d) computed from this peak are given.   One should note that the probable presence of macrostrain changes the lattice parameters and shifts the recorded 2θ values. Thus, the assignment of one phase or another, also considering the small average crystallite sizes, is rather difficult, and the CTS phases can be easily confused with CZTS and vice versa.

Raman Spectroscopy
In the μ-Raman scattering technique, unlike GIXRD, the investigated volume of a sample is very small, allowing a more localized analysis of the thin films. Therefore, the residual phases, revealed by Raman spectra, are not always present in the GIXRD diagrams, because these quantities are below the detection capabilities of XRD. One should note that the probable presence of macrostrain changes the lattice parameters and shifts the recorded 2θ values. Thus, the assignment of one phase or another, also considering the small average crystallite sizes, is rather difficult, and the CTS phases can be easily confused with CZTS and vice versa.

Raman Spectroscopy
In the µ-Raman scattering technique, unlike GIXRD, the investigated volume of a sample is very small, allowing a more localized analysis of the thin films. Therefore, the residual phases, revealed by Raman spectra, are not always present in the GIXRD diagrams, because these quantities are below the detection capabilities of XRD.
The Raman peaks for the polycrystalline phases, which are difficult to be discriminated by XRD, are shown in Table 4. There is a strong similarity between the Raman spectra given by "stannite" and "kesterite" type structures and it is difficult to differentiate between them. For "kesterite", a shift of the most prominent peak (~337 cm −1 , given only by vibrations of the S atoms) to lower values (between 327 cm −1 and 331 cm −1 ) or the existence of a visible shoulder near this peak (at 331 cm −1 ), could indicate the presence of statistically disordered Zn and Cu cations in the "kesterite" structure with copper vacancies, or a stressed "kesterite" phase [18,19].
The Raman spectra measured on the Cu 2 S-ZnS-SnS 2 samples are shown in Figure 4. The Raman peaks, determined by the fitting of the spectra with Lorentzian curves, can be attributed, according to Table 4, as follows: From the Raman spectroscopy results, it can be observed that the CZTS phase is present in all the samples. The formation of the SnS 2 , ZnS and Cu-S binary secondary phases was observed in most samples, except for P4 and P7. The last two samples were Cu-poor/Zn-poor which suggested that a lower quantity of Cu and Zn prevented the formation of a binary secondary phases. P1 is also Cu-poor/Zn-poor, but the high concentration of Sn, due to the proximity to the SnS 2 target, led to the formation of SnS 2 secondary crystalline phase.
The Raman spectra revealed that for some samples the Raman peak from~336 cm −1 was very narrow (P3, P5, P6, P8, P9) while for the others it was broad. From Table 3, we saw that the width of this Raman peak was inversely proportional with the average size of the CZTS crystallites.

Cu2SnS3
Cubic, F-43m 28.470, (111) 47.350, (220) 56.180, (311) 488 [26,27] 267 303 355 ZnS [28] Cubic, F-43m From the Raman spectroscopy results, it can be observed that the CZTS phase is present in all the samples. The formation of the SnS2, ZnS and Cu-S binary secondary phases was observed in most samples, except for P4 and P7. The last two samples were Cu-poor/Zn-poor which suggested that a lower quantity of Cu and Zn prevented the formation of a binary secondary phases. P1 is also Cupoor/Zn-poor, but the high concentration of Sn, due to the proximity to the SnS2 target, led to the formation of SnS2 secondary crystalline phase.
The Raman spectra revealed that for some samples the Raman peak from ~336 cm -1 was very narrow (P3, P5, P6, P8, P9) while for the others it was broad. From Table 3, we saw that the width of this Raman peak was inversely proportional with the average size of the CZTS crystallites.
The full shape of the Raman spectrum can also indicate one compound or another. Compared with the CZTS spectra from the literature [23], P6, P8 and P9 were the closest ones.

Optical Properties
The optical characteristics of combinatorial Cu2S-ZnS-SnS2 films were evaluated in terms of optical transmission spectra. From the transmission data, we estimated the bandgap of the films. This estimation was done by extrapolating the linear part of the curve ( ℎ ) = (ℎ ) to the point ( ℎ ) = 0 (Tauc analysis). The results are presented in Figure 5.   The full shape of the Raman spectrum can also indicate one compound or another. Compared with the CZTS spectra from the literature [23], P6, P8 and P9 were the closest ones.

Optical Properties
The optical characteristics of combinatorial Cu 2 S-ZnS-SnS 2 films were evaluated in terms of optical transmission spectra. From the transmission data, we estimated the bandgap of the films. This estimation was done by extrapolating the linear part of the curve (αhν) 2 = f (hν) to the point (αhν) 2 = 0 (Tauc analysis). The results are presented in Figure 5.
The gradient of the bandgap as a function of distance from the CZTS chemical composition in the quasi ternary Cu 2 S-SnS 2 -ZnS 2 diagram is shown in Figure 6. We can observe that the samples closer in composition to CZTS have similar bandgaps, and as we go further away in the compositional space, the bandgap increases. The bandgap of the synthesized films varies in the interval of 1.41-1.68 eV. The samples P1, P2, P4 and P7 have bandgaps slightly outside the ideal range (1.4-1.5 eV) of CZTS films [29,30]. Surprisingly, these are the samples with the smallest mean crystallite size. This may be justified by the off-stoichiometry which might favor structure deformation and consequently, modifies the bandgap as proposed by Jeffe and Zunger for chalcopyrite semiconductors [31]. The presence of secondary phases affects the bandgaps of these films [32,33]. The presence of o-Cu 3 SnS 4 with a bandgap of 1.6 eV [20], h-SnS 2 which has a bandgap of 1.82 eV, and t-CTS with 1.35 eV [20] are responsible for the variations in the bandgap values. On the other hand, P5 shows a slightly higher bandgap, also obtained by Tanaka et al. [34] in single-phase CZTS films obtained by a sputtering-sulfurization method.
A sulfurization treatment as well as a suitable annealing time are necessary to make improvements in the structural properties and elemental composition [26,33]. As we go further away from CZTS in the composition space, we can observe that the bandgap takes values outside the optimum range for photovoltaic applications.

Conclusions
Combinatorial Cu 2 S-SnS 2 -ZnS 2 thin film samples, with a gradient of chemical composition, were synthesized by magnetron co-sputtering on silicate glass substrates using Cu 2 S, SnS 2 and ZnS binary targets. A ratio of Cu/Zn > 1.5 indicates the formation of secondary Cu-S crystalline phases in P3, P6, P8 and P9, whereas a ratio of Zn/Sn > 1 is a sign for the formation of Zn-S crystalline phases in P9. The XRD diffractograms indicate that annealing in a sulfurized environment increases the polycrystalline fraction of the thin films, but amorphous phases are still present. The sample richest in Sn (P1) is the most amorphous in the as-deposited state, however, it has the highest crystallization rate after annealing, whereas the samples richest in Cu and Zn (P3, P8 and P9) are the most crystallized in the as-deposited state, and they have the lowest crystallization rate after annealing. In all the samples, the majority crystalline phase, obtained by XRD, is a Cu-Sn-S phase that has a stannite type structure, which might be confused with a Cu-Zn-Sn-S phase if a moderate Zn 'doping' is present or a macrostrained CZTS can be considered taking into account the presence of Zn inferred from EDX spectroscopy. The CZTS phase could not be detected by XRD also because the peaks are broad and a deconvolution was not possible, but the Raman results show that the CZTS phase was present in each sample. The presence of binary secondary phases (SnS 2 , ZnS and Cu-S) has been observed in most of the samples, except for P4 and P7 (in P5 only 2%). The Raman peak from~336 cm −1 is very narrow for P3, P5, P6, P8, P9, while for the others it is broad, which means that its width is inversely proportional with the average crystallite size of the CZTS phase. The bandgap of all samples lies in the interval of 1.41-1.68 eV. The results show that the probability of having secondary crystalline phases increases as we travel farther away from CZTS and also the bandgap changes outside the optimum range for photovoltaic applications. Finally, this study offers useful insight into how chemical composition influences the structural and optical properties of Cu-Zn-Sn-S films, knowledge which can be used in the materials design of future solar cells.