AgBiI 4 as a Lead-Free Solar Absorber with Potential Application in Photovoltaics

: AgBiI 4 powder, crystals, and polycrystalline ﬁ lms were synthesized by sealed tube solid state reactions, chemical vapor transport (CVT), and solution processing, respectively, and their structural, optical and electronic properties are reported. The structure of AgBiI 4 is based unambiguously upon a cubic close packed iodide sublattice, but it presents an unusual crystallographic problem: we show that the reported structure, a cubic defect-spinel, cannot be distinguished from a metrically cubic layered structure analogous to CdCl 2 using either powder or single crystal X-ray crystallography. In addition, we demonstrate the existence a noncubic CdCl 2 -type polymorph by isolation of nontwinned single crystals. The indirect optical band gap of AgBiI 4 is measured to be 1.63(1) eV, comparable to the indirect band gap of 1.69(1) eV measured for BiI 3 and smaller than that reported for other bismuth halides, suggesting that structures with a close-packed iodide sublattice may give narrower band gaps than those with perovskite structures. Band edge states closely resemble those of BiI 3 ; however, the p-type nature of AgBiI 4 with low carrier concentration is more similar to MAPbI 3 than the n-type BiI 3 . AgBiI 4 shows good stability toward the AM1.5 solar spectrum when kept in a sealed environment and is thermally


INTRODUCTION
Lead halide based perovskites and perovskite-related semiconductors, particularly methylammonium lead iodide (MAPbI 3 ), are of great interest due to their remarkable performance in photovoltaic devices. This stems from their possession of a suitable band gap as set by the Shockley-Queisser limit, 1 high absorption coefficients suitable for thin film technology, 2 good carrier mobilities, 3−5 and low recombination rates. 4,6,7 Furthermore, films can be cast from solution 8 or vapor deposited 9 and contain inexpensive, earthabundant elements. The main issues associated with these materials are low stability against thermal and photoinduced degradation (e.g., MAPbI 3 is sensitive to light and heat, 10,11 moisture, 12 and oxygen 12 ) and the potential environmental impact associated with scale up of new lead-based technologies, a concern which is shared with other families of functional materials such as ferroelectrics. 13 Studies exploring the chemical diversity of this class of materials have investigated substituting the methylammonium cation with other organic cations, 14−17 the Pb(II) cation with Sn(II) 14,18,19 or Ge(II), 20,21 and replacing or mixing the halides between Cl, Br, and I. 22−25 Attempts to substitute Pb(II) with Sn(II) or Ge(II) have produced compounds that oxidize rapidly to Sn(IV) 14 and Ge(IV) 20 in air and more recently the similar chemistries of Pb(II) and Bi(III), which are isoelectronic 6s 2 cations, has led to interest in developing Bi(III) halide based semiconductors for this purpose. Research has concentrated on producing direct analogues of the hybrid lead perovskites, with 3D networks of bismuth halide octahedra expected to produce narrower band gaps. 26 Several ternary and quaternary Bi-based compounds based on a perovskite-type halide sublattice have been investigated as potential solar absorbers, including (CH 3 NH 3 )BiI 4 , 27 (NH 4 ) 3 Bi 2 I 9 , 28 A 3 Bi 2 I 9 (A = K, Rb, Cs), 29 Cs 2 AgBiX 6 (X = Cl, Br) 30 and (CH 3 NH 3 ) 2 KBiCl 6 , 26 but all have been found to exhibit band gaps that are too wide to be used in single junction solar cells (1.90−3.04 eV). In contrast, BiI 3 , 31,32 whose structure is based on a hexagonal close-packed iodide sublattice, has a smaller band gap of 1.69(1) eV, 33 similar to that reported for AgBi 2 I 7 , 34 which is thought to be close packed. 35 This indicates that selection of a suitable iodide packing array may be an important step in the development of bismuth based halide solar absorbers.
Several Bi-rich compounds based on a close-packed iodide sublattice have been reported in the AgI−BiI 3 (Ag 1−3x Bi 1+x I 4 ) phase field. 36 The earliest study of Ag 1−3x Bi 1+x I 4 isolated the bismuth-rich composition Ag 0.58 Bi 1.14 I 4 (x = 0.14), which was reported to crystallize in a cubic unit cell. 37 Subsequent reports suggested that this cubic phase is stable across a range of stoichiometries, as it is observed independently for AgBiI 4 36,38 (x = 0) and for the compositions Ag 0.73 Bi 1.09 I 4 −Ag 0.58 Bi 1.14 I 4 39 (x = 0.09−0.14) using different synthetic routes. The structure of the cubic phase was solved using crystals of AgBiI 4 grown by a solvothermal method 36 and is reported as a cubic closepacked iodide lattice where edge-sharing octahedra are occupied by either an Ag + or Bi 3+ cation, giving rise to an extended 3D network. It can be considered as a spinel type structure with all of the tetrahedral sites vacant (Figure 2b). Here we use three different routes to synthesize AgBiI 4 as powder, single crystals, and solution-deposited films and evaluate its use as a lead-free solar absorber by detailed characterization of its structure, composition, and electronic and optical properties.

EXPERIMENTAL SECTION
2.1. Synthesis Procedures. Commercial BiI 3 powder (Alfa Aesar, 99.999%) was used for the preliminary syntheses (compositional screening) of Ag 1−3x Bi 1+x I 4 but was found to contain BiOCl as a crystalline impurity and was not used thereafter. The BiI 3 powder used for the synthesis of high quality AgBiI 4 samples (powders, crystals, and films) was synthesized directly from the elements: 0.43 mmol of Bi powder (Alfa Aesar, 99.999%, kept in a He-filled drybox) and 0.75 mmol of I 2 powder (Sigma-Aldrich, 99.8%), corresponding to a 16 mol % excess, were combined in a borosilicate tube of length 150 mm, inner diameter of 7 mm, and wall thickness of 2 mm and sealed at a pressure of 10 −4 mbar by using a liquid nitrogen bath to prevent the sublimation of I 2 . The tube was placed vertically inside a furnace with enclosed heating elements and heated to 427°C to a melt overnight with slow heating and cooling rates of 1°C min −1 due to the high vapor pressure of elemental I 2 . The resulting BiI 3 powder was found to be phase pure by PXRD, and the composition was measured as Bi 0.98(3) I 3.00(3) by scanning electron microscopy energy-dispersive X-ray spectroscopy (SEM EDX). The AgI starting material (Alfa Aesar, 99.999%) was found to be pure by PXRD with the composition Ag 1.00(2) I 1.00 (2) as measured by SEM EDX and used throughout the study.
All powder samples were synthesized in sealed 200 mm long quartz tubes, with 1 mm thick walls and a 6 mm internal diameter. The tubes were vacuum sealed at 10 −4 mbar and placed vertically inside a furnace with enclosed heating elements, and the temperature was ramped at 5°C min −1 to 610°C, held for 1 day, and then cooled at a rate of 5°C min −1 to 350°C and held for a further 5 days. The tubes were then quickly taken out of the furnace, and the bottom half (containing the powder) was quenched to room temperature in a water bath to suppress the formation of BiI 3 and Ag 2 BiI 5 , which form when the reaction is cooled slowly (see Figure S1). AgBiI 4 single crystals were grown by chemical vapor transport (CVT). CVT growth was carried out by placing BiI 3 (0.33 mmol) and AgI (0.25 mmol) in a 150 mm long quartz tube, with 1 mm thick walls and a 6 mm internal diameter, which was then vacuum sealed at 10 −4 mbar. The tube was placed horizontally in the middle of a horizontal two-zone tube furnace between temperatures 363 and 350°C for 2 weeks, with a temperature gradient of 0.8°C cm −1 , the powder being at the hot end. Black octahedral-faceted crystals of dimensions approximately 0.5 mm × 0.5 mm × 0.5 mm and black elongated plate crystals of faces approximately 0.1 mm × 0.5 mm suitable for single crystal X-ray diffraction studies were isolated.
Optimized films were deposited on 15 mm × 15 mm glass substrates, which had been prepared by the following protocol: they were cleaned with soap, sonicated in acetone for 10 min and methanol for 5 min, rinsed with DI water, dried under a nitrogen gas flow, sonicated in isopropanol for 10 min, and finally dried again under a nitrogen gas flow. BiI 3 (0.117 mmol) and AgI (0.058 mmol) were dissolved in 4 cm 3 DMSO by heating and sonicating at 50°C for 30 min. The solution was then dropped onto a glass substrate until full coverage was achieved. The deposited solution was dried by placing the glass slide in an oven and heating to 200°C at a rate of 3.4°C min −1 . Once at 200°C, the oven was switched off, allowing the films to cool back to room temperature.
2.2. Characterization Methods. 2.2.1. Structural. Initial compositional screening, light sensitivity, and solution casting experiments made use of powder X-ray diffraction (PXRD) data measured on a Panalytical X'Pert Pro diffractometer using Co Kα 1 radiation (λ = 1.7890 Å) in Bragg−Brentano geometry and an X'Celerator detector. Phase identification was carried out using the X'Pert HighScore Plus (Version 2.2a) 40 with the PDF-2-ICDD database. PXRD data of AgBiI 4 in capillaries used for photostability assessment were measured on a Bruker D8 Advance diffractometer using monochromated Mo Kα 1 radiation (λ = 0.7093 Å).
PXRD patterns used for detailed structural analysis were collected using the MAC detectors on the I11 beamline at the synchrotron at Diamond Light Source (RAL, Oxfordshire, U.K.) with a wavelength of λ = 0.825898 Å at room temperature. Samples were mixed with 80 vol % amorphous boron to reduce absorption effects and contained within 0.3 mm diameter borosilicate capillaries. For stability measurements the sample was measured every 10°C on heating from 60 to 350°C, using a position sensitive detector for rapid data collection. Topas Academic (Version 5) 41 was used to perform Le Bail fittings and Rietveld refinements of the data. VESTA 42 was used for graphical representation of the structures.
Single crystal X-ray diffraction (SCXRD) data were collected at 100 K on a Rigaku MicroMax-007 HF diffractometer with a molybdenum rotating anode microfocus source and a Saturn 724+ detector using Rigaku Crystal Clear v2.0. Unit-cell indexation, data integration, and reduction were performed using Rigaku CrysAlisPro v171.38.43. 43 The structure was solved and refined using SHELX-2013, 44 implemented through Olex2. 45 2.2.2. Compositional Analysis. SEM EDX was used to measure the composition as a direct elemental analysis technique. Measurements were carried out using a Hitachi S-4800 SEM and an Oxford Instruments model 7200 EDS X-ray detector with the quantification carried out using the microanalysis suite of the Inca Suite software (Version 4.15). Measurements on powders consisted of taking the spectra of nine different areas. Measurements on crystals consisted of taking the spectra at three different areas on each crystal; however, crystals used for structural analysis were subject to six measurements each. Powders of AgI and BiI 3 were used as EDX standards (see Figure  S2). All powders and crystals were sputtered with 15 nm Au to limit charging effects. Errors on reported average compositions correspond to the 1σ standard deviation of the measured distribution.
2.2.3. Optical and Electronic Properties. UV−visible spectra were taken on a Shimadzu UV-2600 instrument with an integrating sphere in both reflection and transmission mode to measure the absorbance, diffuse reflectance, and transmittance. The measurements on the powder sample consisted of a small amount of powder pressed between two glass slides that was exposed to light between wavelengths 600−1400 nm. The Kubelka−Munk function F(R) was obtained from the diffuse reflectance measurements R using the equation F(R) = (1 − R) 2 /2R. Tauc plots were then plotted using (hνF(R)) 1/n vs hν where n = 1/2 and 2 for direct and indirect band gaps, respectively. The absorption coefficients of the films deposited on glass slides were calculated using α = −ln(T/(1 − R))/d where α is the absorption coefficient (cm −1 ), T is the transmission, R is the reflection, and d is the average film thickness (cm). The thickness of each film was measured using a WYKO NT1100 Optical Profiling System. A strip of the film was wiped off the glass substrate to produce a step profile, which was measured in 16 areas for each film. The mean of these step heights is reported as the film thickness with errors corresponding to the standard deviation.
2.2.4. X-ray Photoelectron Spectroscopy Measurements. XPS measurements were conducted on powders of BiI 3 and AgBiI 4 using a SPECS monochromatic Al Kα (1486.6 eV) X-ray source and a PSP MCD5 analyzer. Further details, including spectrometer calibration, can be found elsewhere. 46 Use of 54.7°atomic scaling factors (ASFs) 47 applied to the Ag 3d 5/2 , Bi 4f 7/2 , and I 3d 5/2 peak areas gave

Chemistry of Materials
Article approximate surface compositions Bi 1.0 I 2.9 and Ag 0.2 Bi 1.0 I 3.1 . Charge neutralization at the surface was achieved by means of a low energy electron flood gun and subsequent correction of the binding energy scale to the adventitious C 1s peak (284.8 eV).
2.2.5. Resistivity and Thermopower. Seebeck coefficient and resistivity were measured in a Quantum Design PPMS Dynacool system using the thermal transport and electrical transport options, respectively. The Seebeck coefficient was measured on a cold pressed pellet in a two probe configuration using disc shaped copper leads attached with conductive epoxy. The resistivity was measured on a bar shaped sample in a two probe configuration with gold wires attached using silver paint, achieving ohmic contacts.
2.2.6. Photostability. Two experiments were performed to measure the stability of the material toward the solar spectrum: the first under ambient conditions in an open system and the second under controlled atmospheres in sealed vessels. For both experiments, a Solar Light Model 16S-300-002 Solar Simulator was used which has a spectral output that complies with air mass 1.5 (AM1.5) per the ASTM standard definition. The combination of neutral density filters and lamp-to-sample distance allowed for the tuning of the intensity of the incident light to 1000 W m −2 as measured by a Solar Light Pyranometer PMA2144 and datalogging radiometer PMA2100. For measurements under an ambient atmosphere, AgBiI 4 powder was sprinkled onto a glass slide. Sample temperatures were monitored using a T-type thermocouple and were found to stay below 35°C. Second, for measurements in sealed atmospheres, AgBiI 4 was loaded into thin-walled (0.01 mm wall thickness) borosilicate capillaries under ambient air, dry synthetic air, and helium, and these were sealed with a gas-oxygen torch. The capillaries were then placed in the solar simulator and subjected to the full solar spectrum at an intensity of 1000 W m −2 .
2.3. Computational Details. Density functional theory calculations were performed with a plane-wave basis under periodic boundary conditions using VASP. 48 A plane-wave cutoff energy of 400 eV was used throughout, and core-electrons were treated using the projector augmented wave method. 49 Structural optimization was performed using the optB96b-vdW functional 50 without the inclusion of spin−orbit coupling until atomic forces were less than 0.001 eV Å −1 . This was found to reproduce the experimental cell parameters of BiI 3 to within 1% 51 ( Figure S3). Electronic structure calculations (band structure and density of states, DOS) were then performed using the PBE 52 functional with spin−orbit coupling included. The inclusion of spin−orbit coupling has been found to be necessary for the correct description of BiI 3 . 33 Structural models of AgBiI 4 were constructed in rhombohedral cells with initial lattice parameters of a = b = c = 14.93 Å and α = β = γ = 33.56°, containing 8 metal cations and 16 iodide anions. This cell was chosen such that both the proposed spinel and the CdCl 2 structures could be modeled in the same cell. With each structure, the symmetrically unique configurations of four Ag + and four Bi 3+ ions were constructed, resulting in nine configurations in the spinel structure and six in the CdCl 2 structure. The cell and atomic positions of every configuration were then relaxed and the electronic structure calculated using a 7 × 7 × 7 k-point grid. Of all 15 calculations, one of the two lowest energy configurations had the defect-spinel structure and one the CdCl 2 structure, the defect-spinel structure being more stable by only 18 meV per AgBiI 4 formula unit (using optB86b-vdW energies, Table S1). The crystal and electronic band structures of these two lowest energy configurations are shown in Figures S4 and S5.

RESULTS
3.1. Synthesis and Composition. An initial compositional screening of the Ag 1−3x Bi 1+x I 4 system was performed by reacting AgI and BiI 3 following the powder synthesis procedure described in Section 2.1, with a range of nominal compositions corresponding to 0 ≤ x ≤ 0.19, which covers the range of reported compositions resulting in compounds with the defectspinel structure (x = 0 36 and 0.09 37 −0.14 39 ). It was necessary to quench the tubes from 350°C to room temperature to prevent the formation of BiI 3 and Ag 2 BiI 5 impurities. For all starting compositions, this route resulted in a bulk powder phase at the bottom of the tube and a small quantity of fragile flat dendritic crystals, with widths and lengths varying between 50 and 300 μm, at the top of the tube. These crystals were found to be silver-poor AgBi 2 I 7 (measured by SEM EDX), and their yield increased with decreasing AgI:BiI 3 ratio. The powder retrieved from reactions with nominal compositions of x = 0.05 and 0.07 appear phase pure by PXRD, with the exception of the BiOCl impurity present in the purchased BiI 3 ( Figure S6).
Pure BiI 3 synthesized directly from Bi and I 2 was used with the same synthetic procedure and a nominal composition of x = 0.07 to synthesize the phase pure powders used for the remainder of this study. SEM EDX measurements show the resulting powder has an average composition of Ag 0.97(8) Bi 1.06(5) I 4.00 (10) , within the error of AgBiI 4 (Figure 1), which we will use in the rest of the paper. The composition is slightly Bi and I deficient compared to the starting mixture which is consistent with the small silver-poor crystals forming at the top of the tube. The spread in the composition of the AgBiI 4 powder observed by SEM EDX is larger than that of the standards AgI and BiI 3 ( Figure S2), showing a degree of inhomogeneity that is comparable to a previous report on this system. 39 Efforts to homogenize the sample by a second firing at 350°C for 5 days and quenching to room temperature further increased the spread of the composition along the charge balance line ( Figure S7). This powder synthesis was found to be reproducible with subsequent reactions producing the same average compositions and spreads ( Figure S8).
Single crystals were grown by CVT, resulting in two sets of black crystals with different habits; octahedral-faceted crystals and elongated plate crystals ( Figure S9). SEM EDX measured compositions of four octahedral-faceted crystals and four plate crystals are shown in Figure 1b and Table S2. An octahedralfaceted crystal of composition Ag 1.16 (6) Bi 0.93(6) I 4.00 (5) and a platelike crystal of composition Ag 1.06(5) Bi 0.97(5) I 4.00 (5) , both within 3σ of AgBiI 4 , were used for determination of the crystal structure by SCXRD.
Solution processing of Ag 1−3x Bi 1+x I 4 was carried out by the method described in Section 2.1. This procedure resulted in polycrystalline films with PXRD patterns consistent with that of AgBiI 4 powder (Figure S10), with a set of low-intensity peaks corresponding to a BiI 3 impurity. The extremely intense (00l), l = 4n, peaks indicate that the polycrystalline films are textured. Two optimized films were used for measurements. A Le Bail fit to laboratory PXRD data gives a cubic lattice parameter for films 1 and 2 of a = 12.2104(3) Å and 12.2135(6) Å, respectively, which are consistent with that obtained from bulk powder samples (a = 12.21446(4) Å, see Table S6). The compositions of the films 1 and 2 were measured by SEM EDX as Ag 0.79(10) Bi 0.90(9) I 4.00 (10) and Ag 0.72(13) Bi 1.06(6) I 4.00 (10) , respectively ( Figure 1b). SEM micrographs of the polycrystalline films are shown in Figure S11. Measurements gave a mean thickness of 221(97) nm for film 1 and 343(116) nm for film 2 with a representative film profile shown in Figure S12.
3.2. Structure. The plate-like crystal (Section 3.1) was found by SCXRD at 100 K to be a single untwinned crystal of therefore distinct from the previously reported cubic structure 36 and would be distinguishable by PXRD, as shown by the simulated powder pattern in Figure S13. The structure was solved by direct methods and was found to belong to the layered CdCl 2 structure type: 53 statistically disordered Ag + and Bi 3+ ions occupy every other ⟨111⟩ layer of octahedral interstices in the close packed iodide sublattice, forming the ordered rock-salt structure with layers of edge-sharing Ag + /Bi 3+ octahedra (Figure 2a The previously reported cubic structure also consists of a face-centered cubic iodide sublattice but with differing occupation of the octahedral interstices, such that edge-sharing octahedra are occupied by statistically disordered Ag + and Bi 3+ cations giving rise to an extended 3D network with a defectspinel structure where tetrahedral sites are entirely vacant (Figure 2b). The cell of the cubic defect-spinel structure in its trigonal setting has the same length along the c axis as the CdCl 2 structure, but is doubled in the a and b directions ( Figure 3) and has thus four times the volume of the CdCl 2 cell.
The SCXRD data of the octahedral-faceted crystal (Section 3.1) could be indexed according to two different solutions. Initially, the data were reduced in a cubic cell and the structure was solved in the Fd3̅ m space group. This produced the defectspinel structure previously published 36 with a = 12.1075(1) Å (Table S4). On closer inspection of the data for the octahedralfaceted crystal, the indexed reflections in an Ewald sphere projection show systematic absences throughout the crystal lattice ( Figure 4). If the cubic unit cell is correct, these absences must be coincidental based on the structure factor. Alternatively, the SCXRD data can be indexed to four twinned metrically cubic trigonal cells with cell dimensions corresponding to the CdCl 2 structure, where the c axis (001) of each trigonal cell points along the body diagonal of the cubic defectspinel cell. Fitting the reflections using the twinned trigonal cells removes the zero-intensity cubic reflections highlighted in Figure 4 by reducing the size of the unit cell. The structure was reduced and solved in the R3̅ m CdCl 2 structure and refined against the outputted combined twin reflection listing, with all four twins giving an equal fractional volume contribution (Table S5) (4)). SCXRD is thus unable to determine whether the octahedral-faceted crystal has the defect-spinel structure or a twinned CdCl 2 structure. SCXRD weighted R parameters for the defect-spinel (wR 2 0.0377) and twinned CdCl 2 (wR 2 0.0898) models show that both are valid solutions, and neither model may be disregarded on this basis (the difference in values is due to the deconvolution of the overlapped reflections for each rhombohedral cell in the twinned refinement and that the twinned refinement is fitted against 7 times the number of data points). We note that these models could in principle be distinguished by performing electron diffraction on a single domain; however, these samples were found to be too unstable in an electron beam to achieve this.
Analysis of PXRD data of AgBiI 4 powder samples also cannot distinguish the two models. As for SCXRD data, Figure 2 shows a series of symmetry allowed reflections for the cubic defectspinel model which do not correspond to observed Bragg intensity; these reflections are not allowed in the rhombohedral CdCl 2 model. Rietveld refinements of the two candidate models, shown in Figure 2 and Table S6, produce equivalent fits to the data (χ 2 = 1.229 for the cubic defect-spinel model, versus 1.227 for the CdCl 2 model) and equivalent refined cell parameters: the refined rhombohedral model is metrically cubic with a = 4.31844(1) Å, c = 21.15553(8) Å, and c/2a = 2.44944(1), and no peak broadening due to a rhombohedral distortion is resolved.
To test whether the distinct defect-spinel and CdCl 2 models could be distinguished from a statistical combination of the two, a new cell was constructed which accommodates both cation ordering motifs by using equivalent cell settings (the

Chemistry of Materials
Article trigonal setting of the cubic defect-spinel model and the trigonal setting of the CdCl 2 cell with doubled a, b parameters) ( Figure S14). A series of Rietveld refinements was then carried out at fixed spinel:CdCl 2 ratios, as defined by the occupancies of their characteristic octahedral sites in the close packed iodide sublattice. These refinements show a monotonic increase in χ 2 up to a maximum at the 50:50 ratio ( Figure S15), showing clearly that the structure must adopt either the defect-spinel structure or the CdCl 2 structure but is not a statistical mixture of the two.
Density functional theory (DFT) calculations were performed to investigate the physical plausibility of the two candidate models. For both defect-spinel and CdCl 2 models, the disorder of the Ag + and Bi 3+ cations was accounted for by computing energies for all possible Ag/Bi distributions within a supercell. No significant separation was found between the calculated energies of the defect-spinel and CdCl 2 models, suggesting that the two models are expected to have a similar level of stability (Table S1): neither one of the structures can be disregarded on computational grounds.
In summary, it is shown unambiguously by SCXRD that AgBiI 4 can crystallize in a layered CdCl 2 -type structure which is metrically rhombohedral. The existence of a second polymorph is also demonstrated, but its structure is more enigmatic: it must be either a metrically cubic equivalent of the layered CdCl 2 -type rhombohedral structure or the cubic defect-spinel structure which has been reported previously. 36 3.3. Optical and Electronic Properties. UV−visible spectroscopy measurements of powder and thin-film samples show that AgBiI 4 absorbs visible and near-infrared light ( Figure  5a), with a band gap suitable for single junction solar cells. Tauc plots using diffuse reflectance data give indirect band gaps of 1.63(1) eV for powder samples of AgBiI 4 and 1.75(1) eV and 1.73(1) eV for the polycrystalline films 1 and 2 (Figure 5b). Tauc plots for direct band gaps, which lie in the range 1.73(1)− 1.80(1) eV, are shown in Figure S16. Similar measurements for BiI 3 powder gave an indirect band gap of 1.69(1) eV, consistent with previous reports. 31−33 A small red shift in the optical absorbance is observed for AgBiI 4 when compared with BiI 3 . Indirect band gaps are calculated in similar materials such as CdI 2 (2.9 eV) 54 and AgBi 2 I 7 . 35 Absorption coefficients for films 1 and 2 were found to lie in the range 10 5 −10 6 cm −1 at energies greater than the band gap (see Figure 5c), which is comparable to typical Pb-based perovskite systems, 55 and implying that similar film thicknesses would be required in a AgBiI 4 -based device.   Figure S3). The top of the valence band is dominated by I 5p states and the bottom of the conduction band by a mixture of Bi 6p and I 5p states. As the main contribution of the Ag 4d states is to the bottom of the valence band and the Ag states are absent from the conduction band, we have used a combination of the PBE functional 48 and spin−orbit coupling to study the states around the band gap. This method reproduces the electronic structure of BiI 3 49 well but performs less well when describing more localized Ag d states (the computed band gap of AgI of 1.13 eV with this approach is much narrower than the experimental band gap of 2.85 eV 56 ). As a result, both structural models have calculated band gaps which are underestimates of the experimental values for (CdCl 2 -type AgBiI 4 : 0.51 eV, defect-spinel AgBiI 4 : 0.81 eV), though we expect this to have little effect on the nature and dispersion of the states close to the valence and conduction band edges. 57 Using a similar method, Xiao and co-workers computed a band gap of 1.04 eV for AgBiI 4 in the defect-spinel structure, which increased to 1.70 eV when using a hybrid functional. 35 Hole and electron effective masses for both the defect-spinel and CdCl 2 -type models were calculated by fitting the curvature of the bands along high symmetry directions (Figures S4 and   S5). The defect-spinel model showed little difference in either direction, with hole effective masses of 1.0 m e and 0.9 m e and electron effective masses of 0.6 m e and 0.8 m e along the Γ−Z and Γ−L directions, respectively. In contrast, for the CdCl 2type model the effective masses of holes (1.3 m e ) and electrons (1.8 m e ) are much higher along the Γ−Z direction than along the Γ−L direction where the hole and electron effective mass is 0.4 m e . This is expected since the Γ−Z direction represents transport normal to the Ag + and Bi 3+ filled planes in the structure. The same effect is seen for hole transport within BiI 3 , where we compute an effective mass of 1.5 m e along Γ−Z and 0.9 m e along Γ−L, though surprisingly not for electrons where the effective mass is lower (1.0 m e ) along Γ−Z than along Γ−L (1.2 m e ).

Chemistry of Materials
At 300 K, bulk AgBiI 4 has a large positive Seebeck coefficient and resistivity of 500 μV/K and 1 MΩ·cm, respectively ( Figure  7), comparable to that of MAPbI 3 (820 μV/K and 38 MΩ· cm), 58,59 which suggests a small hole majority carrier concentration and shows that AgBiI 4 behaves as a p-type band semiconductor with low dopant concentration. This is in comparison to BiI 3 which has been characterized as an n-type semiconductor. 60 Resistivity presents Arrhenius behavior between 190 and 300 K, giving an impurity level with an activation energy of 0.4 eV similar to those found in BiI 3 (0.43 eV) 61 and MAPbI 3 (0.48 eV). 58 The measured thermal  conductivity of 0.6 W/K·m is also comparable to that of MAPbI 3 (0.3−0.5 W/K·m). 58 3.4. Stability. Stability toward light and heat in ambient conditions is a major factor in the technical viability of a solar absorbing material under working conditions in a solar cell. A series of PXRD patterns collected in situ on heating from 60 to 350°C show that the first indication of structural decomposition of AgBiI 4 occurs at 90°C where Ag 2 BiI 5 appears as a minority phase as a shoulder to the main phase [111] c /[003] r peak, and this remains the sole minority phase until a major irreversible decomposition process occurs above 290°C ( Figure S17). No structural phase transitions (i.e., changes to the structure of AgBiI 4 itself) are observed in this temperature range, as shown by the linear expansion of the unit cell up to 100°C ( Figure S18). This compares favorably to the thermal stability of Pb hybrid perovskites; e.g., the organic component of MAPbI 3 , CH 3 NH 3 + , decomposes into HI and CH 3 NH 2 at 85°C even in inert atmospheres, 62 while at lower temperatures (45−55°C) this compound undergoes a tetragonal to cubic phase transition which may affect longterm device performance.

Chemistry of Materials
The photostability of AgBiI 4 was investigated under ambient conditions (with a sample sprinkled on a glass slide in an open atmosphere) and under controlled gastight conditions (in thinwalled borosilicate capillaries sealed under air or inert gas), using a solar simulator as described in Section 2.2.6. AgBiI 4 was found to decompose partially to form crystalline AgI under ambient atmospheric conditions after 144 min of exposure at 1000 W m −2 and showed a similar decomposition when a 500 nm band-pass filter was used; however, it did not decompose under these conditions when a 600 nm band-pass filter was used ( Figure S19), implying stability to longer wavelengths. The sealed samples were found to be considerably more stable, with no apparent decomposition after 180 min exposure to the full solar spectrum at 1000 W m −2 , while exposure for an extended period of 150 h produced less extensive decomposition than that observed in the open system ( Figure S20). A control sample, kept in the dark under ambient conditions, was

Chemistry of Materials
Article found to be stable for at least 6 weeks ( Figure S21). These results imply that while wavelengths of 600−761 nm (approximately 21% of the AM1.5 solar spectrum) would be accessible to the photovoltaic process without causing structural decomposition of AgBiI 4 regardless of atmospheric conditions, the sealed environment of a photovoltaic cell is likely to afford stability over the full solar spectrum for extended periods.

DISCUSSION
The results presented in Section 3.2 show that bulk powders, polycrystalline films, and octahedral-faceted crystals of AgBiI 4 cannot be assigned unambiguously to either the cubic defectspinel structure or the rhombohedral (metrically cubic) CdCl 2type structure but must be one of them and not a statistical combination of the two. However, the SCXRD of an untwinned plate-like crystal associates a metrically rhombohedral cell with the CdCl 2 structure type, with no ambiguity. The two structures both have face-centered cubic iodide sublattices and can be described as different vacancy ordered derivatives of the rock-salt structure with half of the cation sites vacant. Both    orderings can be treated as frozen modes associated with the L (1/2,1/2,1/2) point in the Brillouin zone of Fm3̅ m and indeed both are products of the L 2 − irreducible representation. These distortions can be more explicitly labeled, in the format "Irreducible Representation (order parameter direction in representation space) Sub group (basis) (origin shift)" as L 2 − (a, 0, 0, 0) R3̅ m(1/2, −1/2, 0), (0, 1/2, −1/2), (2, 2, 2) (0, 0, 1/2) and L 2 − (a, a, −a, a) Fd3̅ m (2, 0, 0), (0, 2, 0), (0, 0, 2) (0, 1/2, 0) for CdCl 2 -type and defect-spinel, respectively. 63,64 The difficulty in distinguishing these two models derives from the homometry 65 of the cubic structure and the twinned rhombohedral structure: in the absence of a detectable rhombohedral strain they possess identical pair distribution functions and thus powder diffraction patterns. Essentially, they are related as alternative cation decorations of an expanded face-centered cubic iodide sublattice. It can be shown that the bond lengths and angles are the same for the two structures (Tables S7, S8) with the first coordination spheres of Bi/Ag (octahedral coordination) and I (trigonal coordination) shown in Figure S22. It should be noted that this sort of ambiguity is Bismuth is shown in purple and cations in yellow. BiI 3 shows uninterrupted hexagonal close packed iodide sublattice, and both the CdCl 2 and defectspinel structures of AgBiI 4 show uninterrupted cubic close packed iodide sublattices. In the double perovskites Cs 2 AgBiX 6 (X = Br, Cl), one-fourth of the anions in the close packed iodide sublattice are replaced by large Cs + cations, in the perovskite-type structure. The same level of cation substitution is observed in hexagonal iodide layers within (CH 3 NH 3 )BiI 4 , (NH 4 ) 3 Bi 2 I 9 , A 3 Bi 2 I 9 (A = K, Rb, Cs), and (CH 3 NH 3 ) 2 KBiCl 6 . All compounds with structures where cations replace I − anions within the hexagonal layers have optical gaps above 1.9 eV, [26][27][28][29][30]32 where those with pure iodide layers have optical gaps of 1.6−1.8 eV.

Chemistry of Materials
Article common in modulated structures where only first order satellites are observed; 66 in this case the choice of the L point as the modulation vector precludes observation of higher order satellites. The defect-spinel and CdCl 2 structures of AgBiI 4 are computed to have similar energies, suggesting that it may be possible to access the different structures by chemical substitution. We note that the silver rich compound Ag 2 BiI 5 has unambiguously been shown to have the CdCl 2 structure 39 and that CuBiI 4 is reported to have a defect-spinel structure. 67 The structure of BiI 3 consists of a hexagonal close-packed iodide sublattice, in which layers of edge-sharing octahedra occupied by 2/3 Bi 3+ cations and 1/3 vacancies alternate with entirely vacant layers (Figure 3a). 51 Taking into account the unoccupied layers, 1/3 of the octahedral sites in the structure are filled. The inclusion of silver in AgBiI 4 increases the coverage of octahedral sites to 1/2 and changes the closepacking of the iodide sublattice from hexagonal to cubic (Figures 3b,c).
Apart from BiI 3 , other recently reported bismuth halide semiconductors have wider band gaps not suitable for single junction devices which have led to them being suggested as useful for tandem devices ( Table 1). The structures of all of the compounds with band gaps above 1.9 eV are perovskite-like, with large cations replacing 1/4 of the iodide anions within a close packed lattice ( Figure 8, Table 1), whereas AgBiI 4 and BiI 3 with indirect band gaps of 1.63(1) and 1.69(1) eV, respectively, both have fully occupied close-packed iodide sublattices. The indirect band gap of 1.63(1) eV for AgBiI 4 is to date the most suitable band gap for a bismuth halide solar absorber in single junction solar cells but is still wider than the ideal 1.1−1.5 eV range suggested by the Shockley−Queisser limit, 1 suggesting that further reduction of the optical gap in bismuth halides is desirable. This study suggests that materials with uninterrupted close-packed halide sublattices are more promising candidates for bismuth halides with band gaps in the 1.1−1.5 eV range than those based on the perovskite structure of MAPbI 3 and related lead halides.
The electronic properties of AgBiI 4 are more similar to those of MAPbI 3 than to those of BiI 3 , in spite of the closer structural similarity with the latter. Previous reports of the resistivity of BiI 3 at room temperature give values of 100−1000 MΩ·cm 60,68 with n-type charge carriers, whereas AgBiI 4 has an increased conductivity of 2−3 orders of magnitude with a small number of p-type charge carriers, as does MAPbI 3 . 59

CONCLUSION
The lead-free semiconductor AgBiI 4 has been synthesized and characterized. SCXRD of a plate crystal shows that the CdCl 2 structure is accessible, and for the octahedral-faceted crystals and powder samples another polymorph exists that adopts either a metrically cubic CdCl 2 structure or a cubic defect-spinel structure but not a statistical mixture of both. The iodide sublattice is cubically close packed in AgBiI 4 in comparison to the hexagonally close packed sublattice of BiI 3 . We suggest that retaining a close packed iodide sublattice is the best way to keep the band gap narrow because other bismuth halide compounds reported have significantly wider band gaps and all have cations replacing 1/4 of the anions within the anion sublattice. Although the band edge states closely resemble those of BiI 3 , the p-type nature of AgBiI 4 with low carrier concentrations is more similar to the hybrid perovskites than the n-type BiI 3 . These properties suggest that for bismuth halide based semiconductors other structure types, rather than perovskite, may better mimic the properties of the hybrid leadbased systems. AgBiI 4 is accessible via different synthesis routes: solid state sealed tube reactions, chemical vapor transport growth, and solution processing. The thermal stability of AgBiI 4 up to 90°C is above photovoltaic operating temperatures, but light sensitivity intrinsic to the lead and bismuth halides remains a key factor that must be addressed in order to secure the technical viability of these materials in the next generation of solar cells. AgBiI 4 shows structural stability to red light corresponding to energies above the band gap that should be accessible to the photovoltaic process, and its enhanced photostability in sealed, controlled atmospheres improves its potential for device application. ■ ACKNOWLEDGMENTS