Research Update: Bismuth-based perovskite-inspired photovoltaic materials

The effort to find new materials for low-cost thin film solar cells has recently been reinvigorated with the advent of lead-halide perovskites. Photovoltaics based on lead-halide perovskites have exhibited an unprecedented rate of efficiency rise from 3.8% in 2009 to a certified 22.7% in 2017.^1,2 Although the lead-halide perovskites originally investigated (methylammonium lead iodide/bromide) have limited air-stability,^1,3 recently-investigated compositions containing Cs, formamidinium and two-dimensional perovskites have demonstrated over 1000 h device stability.^4,5 However, there is debate over the environmental and commercial impact of the lead content,⁶ motivating efforts to find lead-free alternatives.⁷ 

Many groups searching for lead-free alternatives to lead-halide perovskites have investigated chemical substitution of Pb²⁺ for neighboring elements, such as Sn²⁺, Ge²⁺, Sb³⁺, and Bi³⁺. These efforts are well-documented in recent reviews, and the compounds are compositionally analogous to lead-halide perovskites, e.g., methylammonium tin iodide, methylammonium germanium iodide, and methylammonium bismuth iodide.^8–11 Another strong emphasis has been on finding compounds with the same crystal structure, resulting in numerous investigations into double perovskites, e.g., Cs₂AgBiBr₆, which is also well-documented in recent reviews and articles.^8,12–14 An approach that is contemporary to these is to find compounds that could replicate the tolerance of lead-halide perovskites to intrinsic defects. Computations of methylammonium lead iodide (CH₃NH₃PbI₃) have shown that it primarily forms shallow defects, which would lead to low rates of Shockley-Read-Hall recombination.¹⁵ This is thought to be an important factor for lead-halide perovskites to achieve long diffusion lengths enabling high efficiencies, despite their synthesis by defective solution-processing methods.¹⁶ A qualitative model to understand the defect tolerance of lead-halide perovskites is based on their electronic structure. The hybridization of Pb 6s and Pb 6p orbitals with I 5p orbitals leads to the formation of anti-bonding orbitals across the bandgap [Fig. 1(a)]. With this electronic structure, intrinsic defects (e.g., I vacancies) are likely to form close to the band edge or to be resonant within the valence band.^17–19 The energetic distance of defects from the band edge is further reduced by spin-orbit coupling due to the heavy Pb²⁺ cation, which results in greater band-dispersion. This contrasts with the electronic structure of traditional semiconductors, in which the bonding-antibonding orbital pair is formed across the bandgap [Fig. 1(b)]. Dangling bonds from defects in these traditional semiconductors (e.g., GaAs) are then likely to form transition levels close to mid-gap, which lead to high rates of Shockley-Read-Hall recombination.^17,18

It has been hypothesized that the defect-tolerant perovskite electronic structure could be replicated in materials consisting of a heavy metal cation with a stable pair of valence s electrons.¹⁸ Searches through the Materials Genome database for materials with a significant fraction of s orbitals in the density of states at the valence band maximum identified bismuth-based compounds as promising.²⁰ This is because bismuth forms a stable 3+ oxidation state, leading to stable 6s² valence electrons, giving the possibility of replicating the perovskite electronic structure.¹⁸ Bi³⁺ is also a heavy metal cation, and band dispersion due to spin-orbit coupling is expected. In addition, Bi³⁺, similar to Pb²⁺, is a large polarizable cation and consequently has a high Born effective charge, leading to high dielectric constants that are important for increased screening of charged defects.¹⁸ Bismuth has the additional important advantage of demonstrating very little evidence of toxicity.²¹ 

In this Research Update, we review recent work exploring bismuth-based compounds for photovoltaics (Fig. 2). We focus not solely on bismuth-based perovskites, but rather on bismuth-based materials that have been identified as promising based on recent defect-tolerance theory. These materials therefore do not necessarily have the perovskite crystallographic structure or a composition similar to lead-halide perovskites but rather are those predicted to replicate the perovskite electronic structure (e.g., BiOI and BiSI). We review computations on the electronic structure and defects of these materials to discuss their promise for photovoltaics (Fig. 3). We also evaluate methods to grow these materials and the strategies adopted to improve the morphology, phase-purity, and lifetime. In addition, we review the measured and computed band positions of the materials and early attempts to make solar cells from these new compounds. We conclude with a discussion of the outstanding challenges that need to be addressed as the exploration into bismuth-based photovoltaics matures.

Recent in-depth computations of bismuth-based compounds provide important feedback to the selection rules for identifying defect-tolerant absorbers, as well as further insight into the potential of specific materials as solar absorbers. High Born effective charges have been calculated in bismuth-based compounds.²² The compounds also have high calculated ionic (or static) dielectric constants (Table I). The partial density of states for three such compounds are shown in Fig. 3: BiOI, Cs₂AgBiBr₆, and BiI₃. In all three cases, Bi 6s states are found in the valence band maximum and they overlap with anion p orbitals. Additionally, Bi 6p orbitals are found in the conduction band minimum of all three materials, overlapping with the anion orbitals. These electronic structures are consistent with predictions and have also been found in the computed density of states of other bismuth-based compounds, such as BiSI¹⁸ and (CH₃NH₃)₃Bi₂I₉.³ However, the contribution of the Bi 6s orbitals to the valence band maximum in these materials is smaller than Pb 6s orbitals in CH₃NH₃PbI₃,¹⁸ resulting in less cation-anion orbital interaction. Since the dispersion of the valence band maximum depends on orbital interactions, the effective mass for holes is higher than in CH₃NH₃PbI₃, as shown in Table I, indicating that the mobility of holes is lower in bismuth-based compounds.¹⁶ This effect is particularly pronounced in Cs₂AgBiBr₆ double perovskite, in which the contribution of Ag 4d orbitals exceeds that of the Bi 6s orbitals (Fig. 3). The directionality of the Ag 4d orbitals results in the Ag 4d and Bi 6s states having high charge density at different k-points in the conduction band minimum, giving a flatter conduction band and wider bandgap than CsPbBr₃.¹² This results in higher effective masses.^12,23 The mixing of Br p orbitals with Ag d_z² orbitals at the X point in reciprocal space results in the valence band maximum occurring at the X point rather than at the L point (where the conduction band minimum is), giving an indirect bandgap. Calculations have shown that the lower absorption coefficients resulting from the indirect bandgap limit the maximum achievable efficiencies.¹² 

Defect calculations have also shown that not all bismuth-based compounds with s orbitals at the valence band maximum form shallow defects. Transition levels are Fermi energies at which there is a change in the charge of a defect and are energies within the bandgap at which a defect is expected to be able to trap electrons or holes. The concentration of a defect decreases with the exponent of formation energy and has an Arrhenius relationship with temperature. A formation energy above 1-1.5 eV can be considered high enough for the concentration of a defect to be neglected if the material is grown at temperatures of up to 600 K.¹⁶ In the case of BiOI, the formation energies of all transition levels computed were >1.5 eV [Fig. 3(d)].³⁹ This indicates that BiOI is tolerant to all anti-site and vacancy defects investigated since deep transition levels are not expected to be present. However, for Cs₂AgBiBr₆, low formation-energy transition levels have been found [Fig. 3(e)].²³ Although some of these defects are shallow (e.g., silver vacancies), other defects have transition levels close to mid-gap that would be expected to give high rates of Shockley-Read-Hall recombination. For example, the Ag_Bi anti-site defect has a deep −1/−2 charge transition level located 0.55 eV from the valence band maximum.²³ In part, these deeper traps are due to the reduced and more localized orbital overlap in the valence band maximum, which results in a wider bandgap.¹² Deep traps have also been found in BiI₃, such as the −1/−3 transition level, located >0.5 eV from the conduction band minimum, with a formation energy of <1 eV [Fig. 3(f)].¹⁶ 

The crystal structure also strongly influences the electronic structure, in particular the effective mass. Figure 4 compares three bismuth-based compounds with different dimensionalities. (CH₃NH₃)₃Bi₂I₉ has a crystallographic dimensionality of 0.5D because the [BiI₆]³⁻ octahedra occur as face-sharing pairs that form isolated Bi₂I₉³⁻ groups, which alternate with CH₃NH₃⁺ cations [Fig. 4(a)].^3,27 By contrast, a 0D material would have completely isolated octahedra, whereas a 1D material would have isolated ribbons of bismuth halide. The low dimensionality of (CH₃NH₃)₃Bi₂I₉ limits charge transport because electrons and holes tend to be localized in the isolated Bi₂I₉³⁻ groups, resulting in flat bands in the valence band maximum and low dispersion in the conduction band minimum [Fig. 4(b)].²⁸ Higher dispersion and lower effective masses occur in BiOI, which has a two-dimensional Matlockite structure [Fig. 4(c); Table I].^18,40 Cs₂AgBiBr₆, which has 3D crystallographic symmetry, has higher dispersion still, with the lowest effective masses of 0.14 for holes along the X-Γ direction or 0.37 along the R-X direction, and 0.37 for electrons along the L-W direction in reciprocal space.⁴¹ However, these effective masses are not as low as found in CH₃NH₃PbI₃, in part due to reduced orbital overlap. Recently, Yan et al. introduced the concept of electronic dimensionality to describe the connectivity of the orbitals in the upper valence band and lower conduction band. This accounts for the reduced orbital overlap in Cs₂AgBiBr₆, which gives an electronic dimensionality that is of lower order than the crystallographic dimensionality.²⁹ To illustrate this point, Yan et al. made computations for the more extreme case of Cs₂SrPbI₆, which has 3D crystallographic dimensionality but an electronic dimensionality of 0D because the [SrI₆]³⁻ octahedra do not contribute to the density of states at the band edge. This results in the Pb 6s and I 5p orbitals from the [PbI₆]³⁻ octahedra being isolated, giving flatter bands and a wider bandgap.²⁹ 

Although photovoltaic devices ultimately require the bismuth-based compounds to be synthesized as pinhole-free thin films, new materials are often initially investigated as single crystals. Single crystal diffraction allows the precise determination of the crystal structure and atomic position. Single crystals can in some cases also achieve lower trap densities, at the very least due to the absence of grain boundaries, allowing the mobility and lifetime limits to be investigated. These findings help to determine whether the material possesses suitable properties for photovoltaics before more research-intensive development of devices is undertaken. Two common methods for single-crystal synthesis are the vertical Bridgman method and the solvothermal method [Figs. 5, 6(a), and 6(b)]. In the former, a material is heated above its melting point and cooled slowly from one end of the container to aid slow recrystallization. A seed crystal can be used, the crystallographic orientation of which will determine the orientation of the resultant single crystal. For the solvothermal method, a solution of precursors is heated in a sealed autoclave. A temperature gradient across the chamber aids the crystallization of the product from solution as single crystals, due to the difference in solubility of the product in solution at different temperatures. Single crystal bismuth compounds have previously been grown using this method for application in scintillators, X-ray detectors, and as ferroelectric materials.^46–51 Here, we review the techniques that have been used to grow and characterize single crystals of a selection of bismuth-based compounds, before discussing thin film synthesis. The learnings from these materials can be applied to new bismuth-based compounds that researchers may wish to fabricate.

Single crystal diffraction methods have been used to identify the structure of methylammonium bismuth iodide, (CH₃NH₃)₃Bi₂I₉, single crystals. In early work by Jakubas et al., (CH₃NH₃)₃Bi₂I₉ single crystals were assigned to the space group P6₃/mmc using Weissenberg photographs.⁵² Eckhardt et al. recently confirmed this structure through single crystal diffraction.²⁷ Both groups deduced that (CH₃NH₃)₃Bi₂I₉ consists of isolated Bi₂I₉³⁻ clusters as opposed to the corner sharing metal-halide octahedra that are found in CH₃NH₃PbI₃. In temperature-dependent optical studies of (CH₃NH₃)₃Bi₂I₉,⁵³ Kawai et al. observed an excitonic peak centered at 2.51 eV, which they attributed to localized electronic states on the Bi₂I₉³⁻ clusters. They calculated an exciton binding energy of 300 meV, which is much larger than the exciton binding energy of 5 meV reported for CH₃NH₃PbI₃.⁵⁴ Abulikemu et al.⁵⁵ reported the synthesis of (CH₃NH₃)₃Bi₂I₉ single crystals from reaction precursors in concentrated HI, with recrystallization from an anti-solvent. An indirect bandgap of 1.96 eV was determined, although the excitonic nature of the material was not considered in the bandgap measurement. From space charge limited current measurements, a trap state density of 3.3 × 10¹¹ cm⁻³ was measured, an order of magnitude higher than for CH₃NH₃PbI₃ single crystals, but it is not known whether these traps are deep or shallow.⁵⁶ The high trap density, wide bandgap, and high exciton binding energy limiting free-carrier generation altogether may limit the application of (CH₃NH₃)₃Bi₂I₉ as a solar absorber.

Single crystals of bismuth sulphoiodide (BiSI), a chalcohalide with 1D crystallographic dimensionality,³⁴ have been fabricated by hydrothermal, solvothermal, Bridgman-Stockbarger, and vapor growth methods.^51,57–60 Indirect bandgaps of 1.55–1.61 eV have been reported for the rod-like crystals which are more suitable for single-junction solar cells than the wider bandgap (CH₃NH₃)₃Bi₂I₉.⁵⁷ BiSI has been assigned to the Pnam space group through X-ray diffraction studies, with the structure consisting of covalently bonded ([BiSI]_∞)₂ ribbons held together by van der Waals interactions between the ribbons.^33,34 Density functional theory (DFT) calculations predict anisotropic conductivity, with higher mobility along the ribbons (covalent bonds) than between them (weaker van der Waals interactions), suggesting controlled vertical growth of BiSI is key for good charge transport in devices.³⁴ However, there are limited experimental studies on the transport properties of single crystals, which would be useful to determine the defect tolerance of the material (as well as mobility and lifetime limits).

For materials with 2D crystallographic dimensionality, BiI₃ single crystals have been grown by several methods including the (modified) vertical Bridgman method and physical vapor transport.^37,61,62 Due to the indirect nature of the bandgap resulting in a low absorption coefficient near the band edge (10² cm⁻¹), measurements by ellipsometry have resulted in an overestimation of the bandgap where the direct transition was probed (∼1.9 eV).⁶¹ Through optical absorption measurements, an indirect bandgap at 1.67–1.73 eV has been observed.⁶¹ Measuring single crystals grown by an electrochemical gradient vertical Bridgman method, Brandt et al. reported a photoluminescence emission centred at 1.85 eV.³⁷ From time-resolved photoluminescence measurements, they estimated a minority carrier lifetime of 1.3–1.5 ns. This is above the 1 ns lifetime threshold for absorbers worth further investigation as thin films and devices.^16,63 Previously, Dmitriyev et al. reported an electron mobility-lifetime (μ_eτ_e) product of 9.5 × 10⁻⁶ cm V⁻¹ for BiI₃ single crystals grown by the Bridgman method, equating to a diffusion length of 4.9 μm which is on a comparable length to CH₃NH₃PbI₃ perovskite single crystals.^56,64 This value was larger than the μ_eτ_e product recorded for single crystals grown by physical vapor transport, for which Saito et al. measured values of 3.4–8.5 × 10⁻⁶ cm V⁻¹.⁶² Han et al. reported a route to improve the electron mobility of BiI₃ from 600 cm² V⁻¹ s⁻¹ to 1000 cm² V⁻¹ s⁻¹ via Sb-doping (determined from time-of-flight measurements).⁶⁵ They identified I-vacancies as the major defect in BiI₃ and using DFT calculations proposed that Sb-doping increases the formation energy of iodine vacancies, thus reducing iodine migration through BiI₃. However, BiI₃ has a large hole effective-mass of 10.39 (Table I), whilst also being known to suffer from severe hole trapping, thus its application in devices may be limited by poor hole transport.^18,65

BiOI is another 2D material that has generated interest as a solar absorber based on its promising activity as a photocatalyst.⁶⁶ BiOI single crystals have been grown by chemical vapor transport, where their application as a substrate in superconductor devices was studied.⁶⁷ Electrical measurements showed that BiOI increased in conductivity by 1.65 × 10⁻⁷ S at 275 K when excited by a laser with a wavelength of 640 nm excitation, demonstrating potential as a photoactive material.⁶⁸ 

The most common 3D bismuth-based compounds investigated are double perovskites. Volonakis et al. synthesized Cs₂BiAgCl₆ single crystals by solid-state synthesis in a heated ampoule.⁶⁹ A time constant of 100 ns was fitted to the tail of the photoluminescence decay measurements; however the material possessed an indirect bandgap of 2.3-2.5 eV, which is too large to achieve reasonable conversion efficiencies in a single-junction solar cell. Slavney et al. were able to reduce the bandgap through synthesis of the brominated analog, Cs₂BiAgBr₆, by heating the precursors in concentrated HBr.¹³ The material demonstrated weak photoluminescence emission, but the time resolved photoluminescence measurements exhibited a long tail, which was attributed to the fundamental lifetime of the material. Fitting the photoluminescence decay, the time constant of this tail was found to be 667 ns, which is comparable to the lifetimes of lead-halide perovskites, but we note that the intensity decreased by 1-2 orders of magnitude before reaching the tail in the decay. A recent investigation by Hoye et al. found that the photoluminescence in Cs₂AgBiBr₆ was weak (0.01% photoluminescence quantum efficiency) and red-shifted to the optical bandgap. Optical, transient absorption and photoemission spectroscopy measurements strongly suggested that the photoluminescence decay related to recombination to defect states in the band-gap, rather than across the indirect bandgap.111 While no standard protocol for stability measurements of single crystals exists, thermogravimetric analysis (TGA) measurements showed that the material was stable at temperatures up to 430 °C, while differential thermal analysis showed no phase transitions in this temperature range. Both factors are important as the operating temperature of solar panels can reach >60 °C. McClure et al.⁴¹ measured an indirect bandgap of 2.19 eV for Cs₂BiAgBr₆ by diffuse-reflectivity measurements and observed improved phase purity for single crystals synthesized from acid as opposed to solid-state synthesis. The bandgap of Cs₂AgBiBr₆ was further reduced to 1.4 eV (indirect) and 1.6 eV (direct) by alloying with Tl, with the composition Cs₂(Ag_1−aBi_1−b)Tl_xBr₆ (x = a + b).¹³ A Tl content of x = 0.074 gave the minimum bandgap. However, Tl is more toxic than Pb. Du et al.⁷⁰ also demonstrated bandgap tuning of the double perovskites by alloying Cs₂BiAgBr₆ with Inⁱⁱⁱ and Sbⁱⁱⁱ. Sbⁱⁱⁱ substitution resulted in an indirect bandgap reduction from 2.12 eV to 1.85 eV at 37.5% Sbⁱⁱⁱ. However, sub-bandgap peaks dominated the photoluminescence spectrum due to deep defect states introduced by the dopant. Inⁱⁱⁱ doping resulted in a stronger photoluminescence signal than the un-doped material; however, the bandgap increased from 2.12 eV for pure Cs₂BiAgBr₆ to 2.27 eV for 75% Inⁱⁱⁱ substitution. Calculations have shown that another method for reducing the bandgap of Cs₂AgBiBr₆ is through disorder between Ag⁺ and Bi³⁺ cations, which have similar ionic radii, although there is a reduction in band dispersion, and disordering may lead to reduced mobilities.⁷¹ 

The properties of thin films vary over a wide range, depending on their processing. In the exploration of bismuth-based compounds, much of the effort has focused on eliminating pinholes (which can act as shunt pathways) and controlling the structural properties (e.g., preferred orientation and achieving large grains), which influence the optoelectronic properties (e.g., mobility and lifetime). The fundamental principles of thin film growth are well-documented by Thompson and other authors,^72–75 and common solution-based and vapor-based growth methods are listed in Fig. 5.

For (CH₃NH₃)₃Bi₂I₉, Park et al. reported synthesis via a one-step spin casting method.³² Films demonstrated weak photoluminescence emission and had poor morphology, consisting of interlinked, overlapping grains. Substitution of CH₃NH₃⁺ with Cs⁺ resulted in an increase in the photoluminescence quantum yield; however, this caused the bandgap to increase. Öz et al. also found that (CH₃NH₃)₃Bi₂I₉ films processed by a one-step spin casting method had poor morphology.⁷⁶ Difficulty in achieving compact (CH₃NH₃)₃Bi₂I₉ films has been attributed to the fast crystallization and the poor solubility of BiI₃ in common organic solvents.⁷⁷ Hoye et al. proposed two-step fabrication processes to improve the quality of the films.³ BiI₃ was first spin-cast and annealed, followed by the diffusion of methylammonium iodide through either spin-casting on top or through exposure to methylammonium iodide vapor inside a vacuum oven. The films processed through vapor-assisted conversion demonstrated longer lifetimes. A further increase in lifetime to a value of 8.1 ns was achieved by exposing the thin films to pyridine.^3,16 The longer lifetimes may have been due to the passivation of surface states (reduced surface recombination) and possibly reduced structural defects in the bulk of the film when processed at higher temperatures through vapor reaction. Two-step growth of compact (CH₃NH₃)₃Bi₂I₉ thin films using evaporation steps has also been demonstrated.^30,78 However, photoluminescence lifetimes were limited at 2.84 ns. Additionally, the vacuum processing routes are limited by long processing times. Shin et al. recently presented a solution-processing route to compact phase-pure (CH₃NH₃)₃Bi₂I₉ films using a solvent engineering approach with antisolvent dripping to further aid recrystallization during the spin-casting process.⁷⁷ Infra-red measurements showed that adding dimethyl sulfoxide (DMSO) or 4-tert-butyl pyridine (tBP) as additives to solutions of BiI₃ and methylammonium iodide resulted in the formation of additive-BiI₃ Lewis adducts, which improved the solubility of BiI₃ and slowed down the crystallization process. (CH₃NH₃)₃Bi₂I₉ films had a similar grain size to those produced by a two-step process with a vacuum step by Ran et al.;⁷⁸ however, by comparing SEM images, the films produced by the solvent-engineering approach appear more compact and continuous, contributing toward higher device power conversion efficiencies (0.71% vs. 0.39%). Despite these efforts to achieve compact thin films, the short-circuit current densities have been below 2.95 mA cm⁻².³⁰ While this is partly due to the wide bandgap of 2.9 eV in (CH₃NH₃)₃Bi₂I₉,²⁹ the high effective masses (Table I) and high exciton binding energy could also limit the mobility and diffusion length of carriers (Sec. II).

Tuning the bandgap of the bismuth chalcohalides is possible by substituting the chalcogenide or halide species.³⁴ Kunioku et al.³⁵ reported the synthesis of BiSI thin films, with rod like grains 200–500 nm in size, by treating BiOI with H₂S at ambient pressure and 150 °C. The bandgap reduced from 1.96 eV to 1.59 eV, making it more suitable for a single-junction solar cell. However, careful control is needed when using H₂S due to its toxicity (exposure limit of 5 ppm). Additionally, the rods were randomly aligned. According to band structure calculations, vertically oriented crystals between charge transport layers should be optimal for charge transport due to the smaller electron and hole effective masses along this pathway.³⁴ Controlling the growth direction of BiSI will be important to achieve high device performance. Kunioku et al. also reported that the bandgap of BiOI can be further reduced to <1.37 eV by treatment with H₂Se gas to form BiSeI. However, H₂Se is more toxic than H₂S with an exposure limit of 0.05 ppm, which could restrict commercialization. It should be noted that the substitution of iodine with lighter halide species causes widening of the bandgap; therefore, BiOX where X = F, Cl, or Br are likely to be unsuitable for photovoltaic absorbers.⁴⁰ 

Brandt et al. reported the growth of compact BiI₃ thin films by physical vapor transport and solution-processing.³⁷ Physical vapor transport-grown films had large grains perpendicularly oriented to the substrate while solution processing produced smaller grains. Photoluminescence lifetimes of 180–230 and 190–240 ps were fit for physical vapor transport and spin-cast films, respectively, an order of magnitude shorter than the single crystal lifetimes. By growing films by physical vapor transport at higher temperatures (250 °C vs. 160 °C, with the homologous temperature increasing from 0.63T_m–0.77T_m, the same group measured an improvement in the effective lifetime to 7.3 ns.¹⁶ These efforts indicate that BiI₃ is not defect-tolerant, since a high-quality material needs to be synthesized in order to achieve lifetimes >1 ns, and this agrees with the defect calculations (Sec. II). It has also been challenging to achieve compact, pinhole-free thin films of BiI₃.³⁷ Two contemporary investigations of BiI₃ devices based on solution-processed thin films reported open-circuit voltages >1.3 V below the bandgap with low shunt resistances, and pinholes in the thin films may contribute to this shunting.^38,79

As well as single crystals, thin films of BiOI, with an indirect bandgap ∼1.9 eV,^35,39 have been synthesized for photovoltaic and photocatalytic applications by methods such as successive ionic layer adsorption and reaction [SILAR, Fig. 6(b)]⁸⁰ and the solvothermal method.⁸¹ Films fabricated by SILAR were non-compact, with crossed grains, which exposed leakage pathways, whilst those produced by the solvothermal method consisted of microspheres containing small BiOI particles. In both cases, poor photo-currents and voltages were reported in solar cells. We recently reported the growth of BiOI by chemical vapor transport [Fig. 6(d)].³⁹ Films were more compact than those fabricated by SILAR, and a photoluminescence lifetime of 2.7 ns was measured. These BiOI thin films were also shown to be stable in air for the duration of the 197-day experiment, when the relative humidity varied from 45% to 67%.³⁹ 

There are few reports of double perovskite thin films, owing to difficulties in dissolving the precursors in a common organic solvent. The first report is from Greul et al.,⁴² who fabricated Cs₂BiAgBr₆ by a one-step spin casting method. A photoluminescence decay which had a tail with a time constant of 220 ns was measured, which, whilst shorter than that measured for single crystal Cs₂BiAgBr₆, is up to two orders of magnitude longer than the lifetimes reported for most thin film bismuth-based compounds. But again we note that a recent investigation has found that the photoluminescence in Cs₂AgBiBr₆ originates from defect emission rather than across the band-gap. Hoye et al. directly tracked the decay of all photo-carriers in their Cs₂AgBiBr₆ thin films by transient absorption spectroscopy and found the lifetime of carriers recombining across the indirect band-gap to be 1.4 μs. This was two orders of magnitude slower than the photoluminescence decay.¹¹¹ Whilst the bandgap is too large for single-junction solar cells, it is suitable for a top-cell in 4-terminal tandem devices on silicon, but the low absorption coefficient may limit the achievable efficiencies.⁸² 

AgBiS₂ thin films have also been investigated for photovoltaic applications. Although the lowest-energy room-temperature polymorph of AgBiS₂ is the hexagonal matildite structure, nanocrystals of AgBiS₂ have been reported to form a rock-salt structure.⁴⁵ AgBiS₂ has been employed in semiconductor-sensitized solar cells as a replacement for PbS. Thin films of AgBiS₂ have been fabricated by SILAR,⁸³ electrochemical atomic layer deposition,⁸⁴ and from solution processing of nanocrystals synthesized by the hot injection method.⁸⁵ For the former two, the devices had low open-circuit voltages and fill factors due to incomplete coverage of the electron transport layer exposing recombination pathways, while the latter used insulating ligands to cap AgBiS₂ nanocrystals.⁴⁵ Recently, Bernechea et al.⁴⁵ showed that through tetramethylammonium iodide treatment of nanocrystal thin films, the average device efficiency improved threefold from 1.5% to 4.8% (with a champion certified device efficiency of 6.3%, Table I), the highest recorded efficiency for any of the bismuth compounds discussed. This work highlights the importance of exploring strategies to passivate defects when investigating new bismuth-based compounds.

The architectures investigated for bismuth-based photovoltaics are often borrowed from lead-halide perovskite devices. These device structures are summarized in recent reviews,^10,11 and typically comprise of n-i-p and p-i-n planar stacks [Figs. 7(a) and 7(b)].⁸⁶ The n-type electron-transport layer (ETL) and p-type hole transport layer (HTL) typically have wider bandgaps than the intrinsic absorber (i layer) to avoid parasitic absorption. The advantage of these architectures is that the internal electric field occurs over the entire absorber layer [Fig. 7(c)]. This aids charge separation and extraction through drift, which is particularly important for absorbers with short diffusion lengths. By comparison, in heterojunction devices, drift occurs in the depletion region and diffusion in the quasi-neutral region.⁸⁷ The highest reported device efficiencies of a selection of bismuth-based photovoltaics and their device structures are given in Table I. Some of the reviews detailing device optimization efforts are given in Refs. 8, 10, and 11 but it is clear that in most cases the efficiencies are not close to reaching or exceeding 10%. While some of the reasons for the limited photovoltaic performance are intrinsic to the materials (discussed in Secs. II and III), other reasons are the sub-optimal band-alignment between the bismuth-based absorbers and charge extraction layers.

Efficient device operation requires that the hole and electron transport layer charge extraction levels are energetically matched with the valence and conduction bands of the absorber, respectively. The conduction (valence) band of the hole transport layer (electron transport layer) must also block electrons (holes). In bismuth-based compounds, this is challenging owing to their deep valence band maxima. The ionization potentials in bismuth-based compounds tend to be higher than in CH₃NH₃PbI₃ because the higher atomic number of Bi³⁺ attracts the 6s orbital to deeper energies than occurs with Pb²⁺. The computed band positions of some bismuth-based compounds are shown in Fig. 8, with experimentally measured positions in Fig. 9. For bismuth-based materials, the valence band maximum ranges from −5.2 eV to −7.03 eV relative to a vacuum. By comparison, CH₃NH₃PbI₃ has a valence band maximum at −5.43 eV.⁸⁸ (CH₃NH₃)₃Bi₂I₉, with a valence band maximum at −5.9 eV, is one of the most commonly investigated bismuth-based compounds.³ Shin et al.⁷⁷ recently demonstrated that the efficiency of (CH₃NH₃)₃Bi₂I₉ solar cells could be improved by replacing spiro-OMeTAD (2,2′,7,7′-tetrakis(N,N-di-p-methoxyphenyl-amine)9,9′-spirobifluorene), an organic hole transport layer commonly used in lead-halide perovskite devices, with PIF8 TAA (poly-indenofluoren-8-triarylamine), which has a deeper highest occupied molecular orbital (HOMO) level which is more closely aligned with the valence band maximum of (CH₃NH₃)₃Bi₂I₉. BiOI is another example where the device performance has been shown to be limited due to sub-optimal band alignment, in which downward band bending of BiOI occurs at the interface with the NiO_x hole transport layer.³⁹ In a study by Greul et al.⁴² on Cs₂AgBiBr₆ double perovskite devices, the large band offset between the spiro-OMeTAD (−5.13 eV) and Cs₂AgBiBr₆ (−6.04 eV) has been highlighted as one of the main barriers for limited V_OC and inefficient charge extraction. It is therefore important to find new hole transport materials for bismuth-based compounds. However, commonly used organic hole transport layers have non-ideal HOMO levels (−5.2 to −5.3 eV).⁷⁶ Furthermore, common additives, such as 4-tert-butylpyridine, can dissolve the absorber.⁸⁹ Organic materials are also prone to degradation (potentially limiting device lifetime), are expensive,^90,91 and have been found to be the source of failure in fracture toughness testing.⁹² Possible inorganic alternatives include p-type NiO_x doped with Li or Cu.^93,94 MoO_3, an n-type compound with a conduction band maximum at −6.4 eV can also be used for hole extraction,⁹⁵ although MoO₃ has been reported to degrade CH₃NH₃PbI₃.⁹⁶ Another class of potential hole transport layers are two-dimensional metal chalcogenides, such as MoS₂ and WS₂. These have the advantage of high carrier mobilities (approx. 200 cm² V⁻¹ s⁻¹ and 50 cm² V⁻¹ s⁻¹ for MoS₂ and WS₂, respectively)^97,98 and tuneable work functions.^99–101 

For electron transport layers, bismuth-based photovoltaics commonly use n-type oxides, such as TiO₂, SnO₂, and ZnO, which have given the most efficient lead-halide perovskite devices.¹⁰² The conduction band minima of these n-type oxides (between −3.4 eV and −4.5 eV) are well aligned with the conduction band minima of the bismuth-based absorbers. Additionally, the oxide electron transport layers can be constructed into mesoporous scaffolds. Meso-superstructured devices, particularly those using mesoporous TiO₂, have enabled record efficiencies in lead-halide perovskites.¹⁰³ The advantage of mesoporous scaffolds is that the transport length required for photogenerated electrons from the absorber to reach the electron transport layer is reduced since the nanocrystalline oxide penetrates into the bulk of the absorber layer.¹⁰⁴ This advantage is clearly exhibited in the investigation by Zhang et al. where planar TiO₂ with (CH₃NH₃)₃Bi₂I₉ yielded a device efficiency of 0.14%, whereas mesoporous TiO₂ on top of a compact layer exhibited an efficiency of 0.42%.¹⁰⁴ In light of this, mesoporous TiO₂ continues to be the most widely used electron transport layer, and the efficiency of (CH₃NH₃)₃Bi₂I₉ devices has reached 1.6% utilising this workhorse electron transport material.³⁰ Amongst other oxides, planar ZnO has been exploited in devices based on BiOI³⁹ and AgBiS₂.⁴⁵ ZnO has a conduction band minimum located at −3.8 eV, but it is yet to be fully explored with new classes of bismuth-based absorbers.

An important limitation of bismuth-based compounds for use in single-junction solar cells is the wide bandgap exhibited by most materials (Table I). This could be addressed by doping or alloying to tune the band positions through changes in the composition of the density of states at the band edges. For example, it has recently been found that alloying (CH₃NH₃)₃Bi₂I₉ with sulfur can reduce the bandgap.¹⁰⁵ However, it should be noted that the bandgaps of many un-doped bismuth-based absorbers are close to the optimum for top-cells in tandems with silicon (1.7 eV for two-terminal tandems; 1.9 eV for four-terminal tandems).^4,82,110 For these applications, bismuth-based absorbers may be advantageous over lead-halide perovskites which are not photostable for bandgaps above 1.65 eV.¹⁰⁶ 

A significant amount of effort has been invested in optimizing the morphology of bismuth-based thin films, but this has only partly addressed low device efficiencies. The intrinsic limitations in mobility and defect-tolerance of some bismuth-based absorbers (discussed in Sec. II) should be considered. Further investigation into alternative charge transport layers with improved alignment with the deep valence band maximum of bismuth-based compounds is also needed. Another important consideration is that careful selection of the contact layers is needed to achieve air-stable devices. For example, 1000 h device stability with lead-halide perovskites has been achieved under constant 1-sun illumination by using indium-tin-oxide and atomic-layer-deposited SnO₂/zinc tin oxide as the top electrode.⁴ Although it has been shown that many bismuth-based absorbers are more air-stable than CH₃NH₃PbI₃,^3,13,39 there are no reports of 1000 h device stability. There is also no standard approach for testing and reporting the stability of bismuth-based materials or devices. For example, different storage gases (nitrogen or air), relative humidity levels, and illumination levels are used. There is also no standard test duration, which makes a comparison of the stability of materials or devices grown and tested by different research groups difficult. It is therefore important to follow protocols in stability testing. The International Summit on Organic Photovoltaics Stability (ISOS) has established protocols for the purpose of enabling accurate comparisons of reported stability and lifetime data for organic solar cells.¹⁰⁷ These protocols have been adopted by some groups working on lead-halide perovskites^108,109 and can be extended to bismuth-based absorbers. Research groups can utilize one of the 5 different test types prescribed in the standards (dark, outdoor, laboratory weathering testing, thermal cycling, and solar-thermal-humidity cycling) independently or in conjunction with each other for a comprehensive set of stability data. Each of the test types can be performed at three different test levels, from basic to advanced with increased levels of stringency. For example, a research group might employ the laboratory weathering protocols at the basic level, ISOS-L-1 whilst another group might use the same protocol but at an advanced level, ISOS-L-3. This allows for research groups with varying access to equipment to still report standardized measurements. For commercialization, cells will need to pass accelerated device stability tests outlined by the International Electrotechnical Committee (IEC) for terrestrial photovoltaic modules (IEC 61215-1-3:2016, Ed. 1, 2016). These include 1000 h testing at 85 °C and 85% relative humidity under constant 1-sun illumination.

Compounds containing Bi³⁺ have been proposed as having the potential to mimic the defect-tolerance of lead-halide perovskites. Detailed calculations have confirmed the predictions of high dielectric constants for many Bi-based materials, although calculations also suggest that not all compounds are defect tolerant. The recent work has shown the importance of considering electronic dimensionality, which describes the overlap between orbitals in real space and influences the effective mass and bandgap. The number of Bi-based compounds being grown as thin films is increasing, but there is still wide scope for improvements in their processing to achieve improved morphology, longer carrier lifetimes, and higher device performance. Despite this, the photoluminescence lifetimes of most of the materials have been reported to be <10 ns. There is also much scope for engineering the acceptor levels of charge transport layers to improve charge extraction. This particularly applies to hole transport layers, which need to match the deep valence band maximum positions of bismuth-based compounds.

The authors would like thank Robert A. Jagt for useful discussions. L.C.L. would like to acknowledge funding from the EPSRC Centre for Doctoral Training in New and Sustainable Photovoltaics. T.N.H. acknowledges funding from the EPSRC Centre for Doctoral Training in Graphene Technology (No. EP/L016087/1). R.L.Z.H. acknowledges support from Magdalene College, Cambridge. All authors acknowledge support from the Winton Programme for the Physics of Sustainability.