Chalcogenide Perovskite BaZrS 3 : Thin Film Growth by Sputtering and Rapid Thermal Processing

,


■ INTRODUCTION
The advent of efficient thin film solar cells based on wide band gap (ca.1.6−1.8eV) halide perovskite absorbers has opened a new path toward inexpensive tandem solar cells for large-scale production, in which commercial silicon cells are augmented with perovskite top cells. 1 Even though lab-scale silicon− perovskite tandem cells have reached efficiencies of 28%, 2 there is still a widespread concern that intrinsic chemical instability in halide perovskites might result in degradation of device performance during long-term operation. 3,4−9 Thus, a lot of attention is being paid to strategies to stabilize halide perovskite materials (see e.g.Wang et al. 9 ).Another approach is to employ alternative (non-halide) perovskites that may possess greater intrinsic chemical stability.To this end, chalcogenide perovskites, based on sulfur or selenium anions, have recently attracted interest.In particular, the chalcogenide perovskite BaZrS 3 possesses a band gap in the range suitable for tandem cells.BaZrS 3 crystallizes in an orthorhombic "distorted perovskite" structure similar to that of CaTiO 3 10 and, remarkably, is sufficiently stable to withstand heating in air up to 500−600 °C. 11In terms of detailed solarcell-relevant properties, the available information is still limited.From calculations, the band structure is very similar to the archetypal perovskite methylammonium lead iodide. 12,13ith a direct band gap of 1.75−1.8eV, 14,15 it is potentially a suitable partner for a bottom cell based on silicon or copper indium gallium selenide (CIGS). 1,16It appears that there are few or no deep defects among the likely intrinsic ones, 17 which, coupled with the screening effect of a relatively high dielectric constant, 18 would be conducive to long carrier lifetimes.Both p-and n-type conductivity have been reported, 14,19 possibly suggestive of a dependence on the stoichiometry.Finally, the earth-abundant constituents, and their relative lack of toxicity, are enticing advantages for sustainable photovoltaic technologies.
With regard to synthesis of BaZrS 3 , the majority of published data so far has concerned bulk material (i.e., powders).By heating a mixture of BaS and ZrS 2 in a sealed evacuated tube for several weeks, Hahn and Mutschke synthesized BaZrS 3 for the first time in 1957. 20It was also obtained by sulfurization of BaZrO 3 with CS 2 21 or by heating a mixture of BaCO 3 and ZrO 2 in a stream of H 2 S. 22 All these works used heating temperatures above 1000 °C and long reaction times.In 2000, Wang et al. managed to synthesize BaZrS 3 at temperatures as low as 450 °C by introducing BaCl 2 and excess sulfur to BaS and ZrS 2 , although the highest yield was achieved at 600 °C. 23n 2001, they also showed that BaZrS 3 already formed after a heating time of 10 min. 24In 2017, Niu et al. introduced iodine as a catalyst to the solid-state reaction between BaS and elemental Zr. 12 Wei et al. reported the first example of BaZrS 3 thin films in 2019 by deposition of BaZrO 3 thin films using pulsed-laser deposition followed by sulfurization at temperatures around 1000 °C. 14Recently, works on single crystal growth were also reported. 25,26n this work, we present the first report of thin-film BaZrS 3 synthesis by sputtering.Sputtering is a widely used industrial technique for thin film production on large areas.Besides reporting on the film formation process, we investigate the temperature dependence of crystallization and the reaction sequence of material formation.The material quality is comparable to that obtained by solid-state synthesis using much longer reaction, although the temperatures required for crystallization are rather high.The prospects and challenges for further development of this highly stable perovskite as an absorber layer in tandem cells are discussed.

■ EXPERIMENTAL SECTION
Films of BaZrS 3 were prepared by reactive cosputtering of Ba−Zr−S "precursor" films at ambient temperature, followed by a thermal treatment to induce crystallization.The sputtering was performed in a Von Ardenne CS600 system, with two magnetrons facing the substrate at an angle of 45°at a distance of 160 mm.A Huẗtinger PFG 3000 DC with Advanced Energy Sparc-le 20 pulsing unit supplied pulsed DC with a frequency of 20 kHz to the Zr target (purity: 99.9%; diameter: 102 mm; thickness: 6 mm), while an Advanced Energy CESAR 1312 RF generator powered the BaS target (purity: 99.9%; diameter: 102 mm; thickness: 3 mm bonded to a 3 mm backing plate) with a frequency of 13.56 MHz.They were simultaneously operated at a constant power of 160 W (BaS) and 320 W (Zr).The base pressure was around 8 × 10 −4 Pa.H 2 S with a purity of 99.5% and Ar were simultaneously supplied at a flow rate of 7 and 18 sccm, respectively.The sputtering time was 3 h.For deposition, 300 μm thick monocrystalline silicon wafers with a diameter of 50 mm were used as substrates.Rotation of the substrate was switched on only in some cases; with rotation off, useful composition gradients are created (see later).After sputtering, the samples were unloaded into a nitrogen-filled glovebag and then vacuum-sealed and stored until they were annealed.The sputtering process was tuned to obtain the composition ratio [Ba]/([Ba] + [Zr]) in the range 45−55 at.%, with the ratio [S]/([S] + [Ba] + [Zr]) close to 60 at.% (i.e., with stoichiometric sulfur content and with a range of metal compositions spanning the stoichiometric point).The thickness of the deposited films was in the range of a few hundreds of nanometers, depending on the location on the sample.
Four different methods and two separate systems were used to investigate distinct thermal treatments conditions.They are summarized in Table 1.A custom-made tube furnace was used for low temperature processes (up to 650 °C) to investigate the effect of adding sulfur to the thermal step (methods 1−3).In this system, sputtered samples (and sulfur powder, if desired, to create a S-rich atmosphere) are placed in a graphite box with a tiny opening.This is loaded via a load lock that is pumped to remove traces of air before the box is inserted into the preheated, Ar-filled furnace for the annealing process.For higher temperature processes without added sulfur (method 4), the samples were placed onto a Si holder plate, without being covered, and annealed in a rapid thermal processor (RTP, MPTC RTP-600S) in a N 2 atmosphere.The lamps were heated at a rate of 73 °C/s.The set temperature, measured by a pyrometer on the rear side of the Si holder, was varied from 700 to 1000 °C, while the annealing time was 1 min for all samples.The cooling process took place in situ by simply turning off the heating lamps.
Elemental analysis was performed by energy dispersive spectroscopy (EDS) at 10 keV in a Zeiss LEO 1550 with an EDAX EDS system.
Depth-resolved compositional information was obtained by means of ion beam analysis (IBA) using the 5-MV NEC-5SDH-2 tandem accelerator at Uppsala University. 27Rutherford backscattering spectrometry (RBS) and coincidence time-of-flight/energy elastic recoil detection analysis (ToF-E ERDA) were performed with 2 MeV helium and 36 MeV I 8+ probing beams, respectively.For the RBS, the detector position was fixed at 170°scattering angle, and two different spectra were measured for two different sample normal incident angles (5°and 45°with respect to the beam), facilitating a more accurate depth analysis of the chemical composition homogeneity of the thin film.For the ToF-E ERDA, the ToF-E telescope tube (with a segmented gas ionization chamber for energy detection 28 ) was fixed at 45°scattering angle, and the sample incident angle was placed at 67.5°with respect to the beam.The recorded RBS and ToF-E spectra were analyzed self-consistently following an iterative approach, in which the obtained information from one technique (e.g., the total amount of light species deduced from ToF-E ERDA analysis) was used as boundary condition to fit the other IBA spectra (e.g., RBS).Further details regarding the detection system, as well the data analysis routines for the IBA used in this work, can be found elsewhere. 29razing incidence X-ray diffraction (GI-XRD) was performed in a SIEMENS D5000 system with parallel beam geometry at 45 keV with Cu Kα radiation (λ = 1.5406Å) at 1°incident angle.A Renishaw inVia system was used for ambient temperature photoluminescence (PL) at an excitation wavelength 532 nm, with a spot size of 5 μm.The raw data were corrected to compensate for the sensitivity of the instrument to different photon energies.
Cross-section samples of the thin film for transmission electron microscopy (TEM) analysis were prepared with a focused ion beam and scanning electron microscope (FIB-SEM, FEI Strata DB235).The samples were extracted from the compositional different regions of the substrate at 100°, 120°, and 180°.The samples were attached to a Cu lift-out grid and thinned to electron transparency with a final polishing step using a 5 kV Ga-ion beam.
The TEM analysis was performed on a probe-corrected FEI Titan Themis operated at 200 kV and equipped with the SuperX system for EDS analysis.Spectral images were acquired in scanning (STEM) mode and quantified with the software Esprit 1.9 by Bruker using theoretical k-values provided by the software.

■ RESULTS
For compound semiconductors such as BaZrS 3 , control of composition is very important.In our experiments, we typically performed cosputtering without sample rotation.The results are Ba−Zr−S films with a certain spread of cation compositions across their area, reflecting the relative sputter fluxes received at different locations on the substrate from the two targets.Such samples are termed "compositionally graded" (in contrast, if substrate rotation was used, the samples are termed "uniform").This approach enables study of compositionally resolved properties and avoids the need to co-optimize both composition and other process variables simultaneously.

ACS Applied Energy Materials
In this particular case, strong variations in adhesion of the BaZrS 3 were observed as a function of chemical composition, resulting in cracks and pinholes with varying degrees of severity in different parts of the film.This is attributed to stresses arising during sputtering.Because of this, a high-resolution analysis of film properties as a function of composition is not presented in this report; rather, representative areas are analyzed to highlight the main observations.The composition variation of as-sputtered compositionally graded films on their circular wafer substrates was characterized by EDS as a function of angular position around the wafer, at a fixed radius of 23 mm (i.e., close to the edge).As illustrated in Figure 1a, the BaS and Zr targets faced the samples at angles of 60°and 180°(projected on the horizontal plane), respectively, which correspond to the angular range on the wafer in which the full spread of elemental compositions is observable.Stoichiometric metal content occurs at around 130°(±10°depending on the precursor); the films become increasingly Zr rich with respect to BaZrS 3 for higher angles and increasingly Ba rich for lower angles.The constant S content as a function of angle confirms that the S incorporation is not determined by the availability of S, but rather by the reaction between the metals and S. Aside from the deposited elements and the Si substrate, a certain oxygen content was observed in all cases.Composition ratios as a function of angular position in a typical sputtered precursor are shown in Figure 1b.As the film thickness continuously decreases down to zero when approaching 60°, EDS measurements for angles lower than 100°are unreliable and therefore not displayed.We attribute this nonuniform thickness to resputtering of the film by negative S ions created at the Ba target surface, although this issue has not been thoroughly resolved and is outside the scope of this paper.
To evaluate the accuracy of the EDS measurements, as well as to obtain compositional depth profiles, ion beam analysis techniques (i.e., RBS and ToF-E ERDA) were performed on a uniform precursor.A typical RBS spectrum recorded for the incident angle of 5°is shown in Figure 2a (black solid line)  together with the fit (red solid lineother colors for the elements) using the SIMNRA code. 30Note that the fit presented in Figure 2a already contains information from the other IBA technique (e.g., H and O concentration from the ToF-E ERDAinvisible in the RBS spectrum) as a result of the iterative data analysis procedure (see Moro et al. for further details 29 ).The quality of the RBS fit shows that there are no strong gradient effects in the bulk composition of the film and that no other heavy elements are present in the sample.The S content based on RBS was overall somewhat S-rich compared to the expected value of 60 at.%, which is calculated from the oxidation states of Ba and Zr being II and IV, respectively.The ToF-E ERDA measurement in Figure 2b and the derived depth profile shown in the inset indicate an O concentration of ca.20 at.% at the surface.Only traces of H (<1 at.%) were detected, although the accuracy of the present setup for such a light element can be rather limited (see the discussion below).Assuming a bulk density of 4.22 g/cm 3 for BaZrS 3 , the total film depth-profiled thickness is estimated to be ca.370 nm.The O is concentrated at the surface and tails smoothly into the bulk of the film, while the S content has the opposite trend.This suggests that most of the O content arises from oxidation upon air exposure of the precursor due to the replacement of S. A certain O content (ca. 1 at.%) appears throughout the film.No other impurities were found in the present detection limit of both combined techniques (≥0.1 at.%).The overall fit accuracy of the BaZrS 3 film for the RBS spectrum is ≥0.5%.The statistical uncertainties for the concentration of the main constituents from both RBS and ToF-E ERDA analysis are estimated to be around 1% on average, mainly due to counting statistics (although Zr and Ba peaks slightly overlap in the RBS spectrum, they are better resolved in the ToF-ERDA).Nevertheless, systematic uncertainties in a stand-alone ToF-E ERDA measurements can be estimated to be around 10−20% for light elements (e.g., H), stemming mainly from deficiencies on the efficiency correction for the ToF system and from losses induced by heavy ions (detailed discussion by Arvizu et al. 31 ).The systematic uncertainties from RBS are estimated to be ≤1%.
The overall chemical composition from IBA is in close agreement with the EDS data from the same sample (see Table 2); the [Ba]/([Ba] + [Zr]) ratio was in fact identical within errors.The main difference was that EDS underestimated the S content by a few at.%.It can be concluded that EDS is a good method for metal content and only slightly underestimates S and O content.Because of its simplicity, spatially resolved EDS is used for elemental characterization in the rest of the study, with the above findings in mind.
For as-sputtered precursors, no peaks were detected in XRD measurements (not shown), from which we conclude that the precursors are essentially amorphous.This is in line with expectations for sputtering at ambient temperatures.To induce crystallization of the films, thermal treatments were applied.The initial goals were to establish (a) whether additional sulfur vapor is required to avoid decomposition and release of S(g) from the film during thermal processing (as is sometimes needed for chalcogenides 32 ) and (b) to gauge a suitable temperature range for formation of crystalline material.To investigate stability in the presence/absence of S vapor, three nominally identical compositionally graded precursors were annealed in a custom-made tube furnace.Two runs were made with the precursors inside a closed graphite box featuring only a tiny orifice: in one case without added S powder (method 1) and one case with added S (method 2) to create a S-rich atmosphere.In the third run, a precursor was heated on an open holder with a flow of Ar passing over it (method 3).All three runs were made for 20 min at 650 °C.The resultant S and O contents for the samples (from EDS) are shown in Figures 3a and 3b, respectively.
The fractional S content for films processed without added S dropped uniformly from the initial value of ca.61 at.% in the precursor to around 53−56 at.%, i.e., apparently substoichiometric with regard to BaZrS 3 .The aforementioned underestimation of S content by EDS, coupled with the fact that some of the Zr and Ba must have been included in oxide phases, means that it is not possible to conclude whether this corresponds to removal of "excess" S only or whether the S content in the main phase actually does drop below stoichiometry for BaZrS 3 .However, there is clearly a loss of S, and it is notable that this loss was identical (apart from for the most Ba-rich region of the film) for the case with the sample on the open holder with Ar flow (method 3) and the case of the sample inside a closed graphite box without added S (method 1).If this loss was corresponding to decomposition of the bulk phase, a greater extent of loss should result in the case of the open box (method 3) due to the continuous removal of gas products from the sample surface, whereas the closed box of methods 1 and 2 should suppress losses due to formation of an equilibrium between the solid and gas phases. 32One explanation for the observed behavior is that the sputtered films contain minor amounts of an unstable, Srich phase, which is responsible for a small, constant S loss under both conditions.The sample processed with added S (method 2) retained its original S content in the Ba-rich region (angles below 140°) and lost S in the Zr-rich region, but not as much as the samples annealed without S (methods 1 and 3).At the same time, this sample (method 2) also oxidized heavily (Figure 3b), with the fraction of O ranging from 15 to 50 at.% (being highest in the Ba-rich region of the film).Meanwhile, the samples annealed without any additional S (methods 1 and 3) oxidized only very slightly compared to the precursor.As previously discussed in the IBA section, oxidation seems to occur upon unloading of the samples to air after the thermal process.From the above comparisons, it appears that the main phase in the films does not undergo decomposition in the absence of S (methods 1 and 3) but that S-rich phases are formed (or retained) in the presence of excess S and are easily oxidized (method 2).Thus, further thermal processes were made without added S.
The XRD pattern of a sample processed in the tube furnace without S at 650 °C (method 1) is shown in Figure 4a (lowest scan).The relatively low and broad peaks indicate limited crystallization at this temperature, which was the upper limit for the custom-made tube furnace.Therefore, the influence of 3.4 4.0 0.0 higher temperatures on crystallization was studied by using an RTP system (method 4).The temperature was varied from 700 to 1000 °C, while a process time of 1 min was used in all cases.EDS was performed after annealing and the results are plotted in Figure 1b.For the samples processed at 700−900 °C, a similar S loss is seen after annealing, with the fraction of S falling to 52−55 at.%.A larger loss is found for the 1000 °C anneal, at which point it would appear that the main phase finally undergoes bulk decomposition, as also supported later by XRD and PL analysis.Again, the O content slightly increases after annealing and does not appear to be temperature-dependent, although there are run-to-run variations.In particular, O content was lowest for 900 °C, which was found to be the optimal annealing temperature among those tested (see later).High O and Ba content at low angles might suggest the presence of some Ba oxide or sulfate in the film in that area.
To observe the effect of process temperature on formation of crystalline phases, XRD was performed.Figure 4a shows the patterns for precursors annealed at progressively higher temperatures.For all temperatures in the range 650−900 °C, the patterns correspond to BaZrS 3 , 21 without obvious intermediate phases, although the minor reflexes only start to appear around 800 °C.The peak width (FWHM), also shown in Figure 4a, continuously decreases as the anneal temperature is raised, which indicates an improvement in crystal size and/or quality for the higher temperatures.At 1000 °C, although the peak width of the BaZrS 3 phase remains the same, numerous other peaks arise, corresponding to various oxide and sulfide phases, indicating extensive degradation of the film.On this basis, 900 °C was chosen as a suitable temperature for further annealing.Figure 4b shows detailed patterns for two samples both annealed at 900 °C.The lower pattern corresponds to a uniform sample, with composition close to stoichiometry in both metal and sulfur content.In this case, only BaZrS 3 -related peaks are found.The upper pattern corresponds to a compositionally graded sample measured in a Zr-rich region.Several additional peaks are notable, marked by lowercase letters.Matches could not be found among the binary or ternary sulfides in the Ba−Zr−S system, but they could correspond to oxides: the peaks labeled b (at 30.2°, 42.9°, and 50.4°) are a good match to the main diffraction peaks of a tetragonal ZrO 2 phase. 33The peaks marked a could not be assigned.In the Ba-rich region, no additional peaks were found.
For each sample and annealing temperature, PL was measured on several spots.Because of the inhomogeneity of the deposited film, PL was very sensitive to the measured  ACS Applied Energy Materials position.To understand the effect of the annealing temperature, the most intense PL signal for each tested temperature was selected.These were mostly found for angles in the range 100°−135°, i.e., close to the region of theoretical stoichiometry (centered at around 120°, see Figure 1).After smoothing and normalizing, the most intense PL curve for each temperature is plotted in Figure 5a along with the curve published by Niu et al. 12 for comparison.The height of the main peak and its FWHM are plotted against temperature in Figure 5b.
First of all, it can be seen how the PL intensity at low photon energies reduces with increasing temperature up to around 900 °C, while the PL peak becomes more intense and narrower and converges to 1.84 eV for a 900 °C anneal.At this point, it agrees very well in shape and position with what has been reported in the literature for bulk powders. 12The long tail at lower energies could arise from a high density of defects close to the band edges or a secondary phase with low band gap.Increasing the annealing temperature even further does not seem to improve PL but rather reverses the trend.Further samples prepared with longer and shorter anneal times at 900 °C did not show any improvements.
Conventional and scanning transmission electron microscopy and selected area electron diffraction (SAED) analyses were performed on different positions of a compositionally graded sample annealed at 900 °C. Figure 6 shows those measured at 120°, i.e., where the elemental composition is closest to stoichiometry.Both TEM and SAED confirm the polycrystalline nature of the fabricated film by the grain contrast and the appearance of diffraction spots, respectively.STEM-EDS spectral imaging was performed on the samples extracted at the substrate positions of 100°, 120°, and 180°to obtain information about chemical homogeneity within the film.For the 120°and 180°sample, no obvious inhomogeneity could be detected, and the chemical composition in the film appeared uniform (results shown in the Supporting Information).For the 100°sample (Ba-rich), chemical inhomogeneity was detected, and the result is shown in Figure 7. Zr and O are often present in high concentrations where Ba and S are almost absent, as the contrast in the combined maps in Figure 7b,c indicates.According to the chemical identity, we suggest the presence of small ZrO 2 particles in these areas.

■ DISCUSSION
Despite some adhesion problems, the sputter−anneal process presented here generated relatively phase pure films of the chalcogenide perovskite BaZrS 3 .Based on the width of the diffraction peaks, and the energy and distribution of the photoluminescence response, [20][21][22]24 material made in a rapid   thermal process had comparable crystalline quality to that from bulk synthesis methods, which instead required reaction times on the order of hours.Minor oxide phases of ZrO 2 (detected by XRD and STEM-EDS) and possible Ba−O (implied by compositional measurements) are present, especially for Zrrich or Ba-rich material, while the PL response was also highest for near-stoichiometric compositions. Becaue the PL response could be influenced by the presence of secondary phases (including oxides) as well as the variable morphology of the films, a more detailed analysis of PL as a function of the stoichiometry (i.e., defect type) of BaZrS 3 is not viable at the present time.
While film quality is expected to improve with process optimization, one challenging aspect of the synthesis from a practical standpoint is that achieving reasonable crystalline quality for BaZrS 3 required annealing temperatures in the range 800−900 °C.This is considerably more than used for other inorganic thin film solar absorbers, e.g., Cu(In,Ga)Se 2 and Cu 2 ZnSnS 4 (usually prepared at around 450−550 °C) or halide perovskites (usually prepared below 100 °C).Even at 900 °C, the resulting grain size was still comparatively small, ca. 100 nm (see Figure 6a).This is evidence of a large energetic barrier for nucleation or grain growth/diffusion in the growth process employed here.Interestingly, such a feature could be closely connected to the apparently exceptional chemical stability of chalcogenide perovskites toward, for example, oxidation or decomposition (S-loss) at high temperature. 11This case is opposite to that of halide perovskites, for which a low barrier for phase formation and diffusion allows film growth at low temperatures but leaves the material susceptible to various decomposition processes even at ambient conditions.While their stability is an extremely important advantage, if chalcogenide perovskite thin films are to be employed in tandem cell (or other) applications, it would be desirable to reduce the annealing temperature to facilitate integration of the annealing step into the device fabrication process.In particular, standard substrates (e.g., soda lime glass) and other surrounding materials such as contact layers must be compatible.It remains to be seen whether the high growth temperature is an inherent characteristic or whether it can be reduced by using alternative formation routes.As an aside, one convenient outcome of the high stability of BaZrS 3 is that additional S vapor was not required to stabilize the material during high-temperature processing, which enables the use of conventional equipment without special adaptations.
A further complication in the sputter-recrystallization process employed here was the persistent presence of O in the films.Although several different anneal methods were used, the O content was similar.Despite attempts to minimize the duration of air exposure prior to measurement, precursors always showed a certain O content, at least 3−4 at.%, although this did not rise even after several weeks of ambient storage.Further O incorporation only occurred after the samples were annealed, bringing the total to ca. 10−12 at.%.
The reaction seems connected to the presence of excess S in the samples; the precursors were slightly S-rich (about 4 at.% above BaZrS 3 stoichiometry as measured by ion beam analysis), while oxidation was especially obvious when we tested the addition of excess S vapor to the thermal step.In the following, we propose a phase formation mechanism that can account for the various observations.In BaZrS 3 , the stoichiometric S content is 60 at.%; thus, the precursor composition of 64 at.% S implies the presence of additional S-rich phases.There are no S-rich ternary phases in the Ba−Zr− S system, but S-rich binary phases do exist, namely BaS 2 and ZrS 3 . 21,23The overall S content, if all the Ba and Zr atoms were in these phases, would be ca.71 at.%.A comparison of these numbers reveals that our precursors' modest S excess may cause about 30−40% of the Ba and Zr atoms to reside in S-rich phases instead of a BaZrS 3 -like phase.Because all binary sulfides of Ba and Zr rather easily hydrolyze in air, we propose that these S-rich phases are the origin of the S-loss and oxidation phenomena seen in our film growth process.
A reaction schematic for the phase distribution through the stages of thin film growth is illustrated in Figure 8 and described as follows: (1) During sputtering, a slight S excess leads to inclusion of S-rich phases in the film.(2) Upon unloading from the sputter chamber, near-surface S-rich phases oxidize, leading to S loss and O gain in the near-surface region.
(3) During the thermal process, BaZrS 3 begins to crystallize from the precursor, and some of the S-rich material is reduced by evaporation of S(g), resulting in S loss (without oxidation).At the same time, film stresses cause cracks and pinholes in the films.(4) Upon unloading from the thermal processor, residual S-rich material is exposed due to cracking in the film and oxidizes.The final sample is primarily BaZrS 3 with minor inclusions of oxide phases.
Secondary phases (both oxides and binary sulfides) could be detrimental for applications of BaZrS 3 films; for example, they could be responsible for the tailing to lower energies in PL (see Figure 5a), which would reduce the open circuit voltage in a solar cell.However, improved growth methods, in particular focusing on fine-tuning the S content in the precursor, so as to avoid exceeding the stoichiometric value for BaZrS 3 , are expected to lead to improvement.

■ CONCLUSIONS
The chalcogenide perovskite BaZrS 3 is hoped to present the excellent optoelectronic properties of halide perovskites such as MAPbI 3 but in an extremely chemically robust, nontoxic compound.In this work, we report the first details of a sputterrecrystallization approach for thin film deposition of BaZrS 3 that yields nearly phase pure, polycrystalline perovskite films.Despite some adhesion problems that remain to be addressed, the crystalline quality and photoluminescence response of the perovskite films were comparable to that achieved from bulk synthesis methods, which is extremely encouraging.A limitation with the present method, which is thought to relate to the high stability of the compound, is that temperatures in the range 800−900 °C were required for good crystallization, with obvious consequences for solar cell integration.Oxygen incorporation in the films appears to occur due to the presence of air-sensitive, S-rich secondary phases formed during film deposition.Future work should focus on strategies to reduce process temperatures and achieve better control of S content to improve film quality further.Detailed investigation of optoelectronic and transport characteristics of relevance to application in solar cell devices would be therby enabled.

Figure 1 .
Figure 1.(a) Sketch of a compositionally graded sample indicating relative positions of the sputtering targets during deposition.The dotted arc depicts the range where EDS was performed.(b) Film composition expressed as ratios of elemental concentrations, measured by EDS before and after annealing at different temperatures.The x-axis represents the angular position on the substrate as described in (a).

Figure 2 .
Figure 2. (a) Experimental RBS spectrum (black solid line) for 2 MeV helium and incident angle of 5°, plotted together with a fit (red solid line) using boundary conditions from ToF-E ERDA analysis (b), as a result of the iterative analysis approach (see text for details).Other color lines represent signals from the other detected elements.(b) Experimental coincidence ToF-E ERDA spectrum for the sample as in (a), recorded using 36 MeV I 8+ as probing beam.The inset represents an atomic depth profile deduced from the ToF-E spectrum (see text for details).

Figure 3 .
Figure 3. (a) S and (b) O content in as-sputtered precursor and after annealing at 650 °C for 20 min, with and without additional S in a closed graphite box, and without S in an open box with Ar flow.

Figure 4 .
Figure 4. (a, left) XRD patterns for compositionally graded samples annealed at increasing temperature from 650 to 1000 °C.(a, right) Full width at half-maximum (FWHM) of the (121) peak as a function of anneal temperature.(b) Two samples prepared with 900 °C anneal.The lower scan is a uniform stoichiometric sample; the upper is a compositionally graded sample measured in the Zr-rich region (at 180°).Lines and indices come from reference pattern for BaZrS 3 of Clearfield et al.21

Figure 5 .
Figure 5. (a) Normalized PL from samples annealed at different temperatures.The PL curve taken from Niu et al. 12 is also shown with squares for comparison.(b) Maximum intensity and FWHM of the curves shown in (a).

Figure 6 .
Figure 6.(a) TEM bright-field image showing crystalline grain contrast.The measurement was performed at 120°on a compositionally graded sample annealed at 900 °C.(b) SAED pattern showing many diffraction spots of crystals in arbitrary orientation, as expected for a polycrystalline thin film.

Figure 7 .
Figure 7. (a) STEM bright field image of the area where the EDS map was recorded.The measurement was performed at 100°on a compositionally graded sample annealed at 900 °C.(b) Combination of the EDS map of Ba (green) and O (red).(c) Combination of the EDS map of Zr (blue) and S (pink).Comparing both maps reveals that O and Zr are highly concentrated in regions where Ba and S are deficient.

Figure 8 .
Figure 8. Proposed phase formation through the stages of thin film growth.

Table 1 .
Illustration of the Different Thermal Treatment Methods Used in This Work

Table 2 .
Comparison of the Theoretical Stoichiometric Composition of BaZrS 3 with the Experimental Values Calculated from EDS and IBA Measurements on a Uniform Precursor, Expressed as Ratios of Elemental Concentrations