Beyond methylammonium lead iodide: prospects for the emergent field of n s 2 containing solar absorbers

The field of photovoltaics is undergoing a surge of interest following the recent discovery of the lead hybrid perovskites as a remarkably eﬃcient class of solar absorber. Of these, methylammonium lead iodide (MAPI) has garnered significant attention due to its record breaking eﬃciencies, however, there are growing concerns surrounding its long-term stability. Many of the excellent properties seen in hybrid perovskites are thought to derive from the 6s 2 electronic configuration of lead, a configuration seen in a range of post-transition metal compounds. In this review we look beyond MAPI to other n s 2 solar absorbers, with the aim of identifying those materials likely to achieve high eﬃciencies. The ideal properties essential to produce highly eﬃcient solar cells are discussed and used as a framework to assess the broad range of compounds this field encompasses. Bringing together the lessons learned from this wide-ranging collection of materials will be essential as attention turns toward producing the next generation of solar absorbers.


Introduction
As modern society continues to consume natural resources at an ever-increasing rate, there is a growing demand for a clean energy source capable of providing indefinite and sustainable economic growth. 1 Arguably the most abundant renewable energy resource is sunlight, with over 1500 exawatt-hours of energy falling incident on the earth in the form of solar radiation each year. 2 The enormity of this resource is apparent when one considers that the total known reserves of oil, gas and coal amount to just 9 exawatt-hours.A year's worth of sunlight, therefore, provides almost two hundred times the energy of the world's entire known supply of fossil fuels. 3Of the technologies available to harness solar radiation, photovoltaic cells are perhaps the most promising due to their ability to convert light directly into electricity.Indeed, coverage of less than 0.3% of the earth's surface would be sufficient to meet the world's energy needs using the latest commercially available solar panels.However, in order for photovoltaic cells to compete with fossil fuels in utility-scale power generation, it is necessary to reduce the total cost of solar energy, either through increased efficiencies or lower cost per photovoltaic cell. 4he current photovoltaic market is dominated by crystalline silicon solar cells, which, having benefited from six decades of research, possess power conversion efficiencies (PCEs) around 21%. 5 However, the performance of these cells is limited by the indirect band gap of silicon and the last five years have seen little improvement to their efficiencies.Currently, the highest performing single-junction solar cells are those containing GaAs.These devices have shown efficiencies approaching the Shockley-Queisser limit of 33% 6 but due to the high raw materials cost of GaAs (B7 times that of crystalline silicon), 7 their viability is limited to extraterrestrial applications where efficiency presides over cost. 8ecently, hybrid halide perovskite solar cells have emerged as a serious contender to silicon-based devices thanks to an unprecedented rise in efficiency-reaching 22% in 2016, overtaking all other third-generation technologies. 9,10CH 3 NH 3 PbI 3 (MAPI) is the champion hybrid halide perovskite absorber and has, as such, attracted an enormous volume of research interest.8][29] However, attempts to replace methylammonium with other organic and inorganic cations outside of those already known in the literature has proved exceedingly challenging. 30,31Whilst the tolerance factor metric has traditionally been applied as a predictor of perovskite stability, it performs poorly across the range of known iodide perovskites. 32,33Recently, a revised tolerance factor method has been developed that takes into account the greater covalency seen in some metal-halide bonds and is able to accurately predict the stability of the majority of halide perovskites. 34Strikingly, the report suggests that only a handful of halide perovskites remain to be discovered, with most likely to possess band gaps unsuitable for photovoltaic applications.It is therefore essential, now more than ever, that the search for the next generation of solar absorbers be extended beyond the cubic perovskite motif.
The meteoric rise in the efficiency of MAPI has fuelled intense interest among a broad community of physicists, chemists, and engineers and has brought together the lessons learned in over 20 years of development of related dye-sensitised and organic photovoltaic cells. 35,36Brandt et al. have recently proposed several key properties likely to give rise to highly efficient and defect-tolerant solar absorbers, including a large dielectric constant, small effective masses, a valence band maximum composed of antibonding states, and high levels of band dispersion. 37Materials containing post-transition metals with an ns 2 electronic configuration (i.e. an N-2 oxidation state) possess many of these properties due to their soft polarisabilityleading to high Born effective charges-and large spin-orbit effects, which act to increase the bandwidth of the conduction band. 38,39As such, a wide range of compounds comprising Pb 2+ , Sn 2+ , Ge 2+ , Sb 3+ , and Bi 3+ cations are currently of interest for their solar absorber ability.
In this Review we focus on this emerging field of ns 2 containing solar absorbers.The ideal properties needed to produce highly efficient solar cells are discussed and used as a framework to assess the broad range of compounds this field encompasses.Initially, group 14-based materials-those containing lead, tin and germanium-are examined, with both well established and novel absorbers considered.The second half of this Review concerns materials containing the group 15 post-transition metals antimony and bismuth.Throughout, particular attention is given to the relationship between structure and properties, specifically the effect of dimensionality on stability and carrier transport.Lastly, we look towards the future of next generation solar absorbers.

Desired solar absorber properties
The performance of novel solar absorbers is hard to predict in practice, due to the dependence on many external conditions such as the method of deposition, quality of precursor and device architecture.However, analysis of high performance solar materials reveals several key properties likely to beneficially David O. Scanlon David O. Scanlon received his degree in Computational Chemistry (2006) and PhD in Chemistry (2011) from Trinity College Dublin, Ireland.After a Ramsay Fellowship in the Department of Chemistry at UCL, he was appointed to a Lectureship (2013) and then as a Reader (2016) in Computational, Inorganic and Materials Chemistry at UCL and at Diamond Light Source.He currently leads the Materials Theory Group, which focusses on the use of Computational Chemistry techniques to understand and predict the behaviours of solid state materials, primarily for renewable energy applications.The group is currently working on novel materials for photovoltaics and photocatalysis, Li-ion batteries, thermoelectrics, and optimising materials for thin film displays.
This journal is © The Royal Society of Chemistry 2017 affect device efficiencies. 37Crucially, many of these properties, whilst difficult to measure experimentally, can be obtained relatively cheaply from theoretical methods, thus highlighting the importance of a combined theoretical/experimental approach in screening new materials.

Magnitude and nature of the band gap
Arguably the most important property of a solar absorber is its band gap, as it determines the maximum theoretical PCE possible for the material.The best performing absorbers possess band gaps in the 1.10-1.55eV range, as quantified by the well-known Shockley-Queisser limit, 6 which takes into account the antagonistic dependence of short-circuit current ( J sc ) and open-circuit voltage (V oc ) on the band gap and solar spectrum (Fig. 1a).While this does not preclude materials with band gaps outside this range being examined, in order to maximise efficiency, a band gap close to 1.3 eV is highly advantageous.

Strength of optical absorption
Strong optical absorption is particularly crucial for solar absorbers: many compounds with ideal band gaps are poor absorbers.Indeed, low absorbance due to an indirect band gap is one of the primary reasons to look beyond crystalline silicon as an absorber.As recently stressed by Yu and Zunger, strong absorption requires a direct band gap transition; however, materials with indirect band gaps may still perform well if a direct transition of suitable energy is also available. 40Loss in absorption can further result if the fundamental band gap is dipole disallowed, widening the optical band gap relative to the fundamental gap, as often results in centrosymmetric materials. 41,42Ideally, strong absorption is characterised by a steep absorption edge in the absorption coefficient, a, just above the band gap, up to 10 4 -10 5 cm À1 or higher (Fig. 1b). 43

Charge carrier effective mass
High charge carrier mobilities can be particularly useful in photovoltaics for establishing electron-hole separation and improving device performance.Mobility, m, is dependent on the dispersion of the band edges in a material-theoretically quantified by the effective mass of a carrier-with greater dispersion giving rise to smaller effective masses and in turn, enhanced carrier mobilities (Fig. 1c).The primary limiter of carrier mobility is through scattering by defects, phonons and other charge carriers.We note that mobility is not the only important transport property: minority-carrier lifetimes, t, have recently been proposed as an essential metric for screening novel PV materials, due to their role in Shockley-Read-Hall recombination (trap-assisted non-radiative recombination). 37,46,47ndeed, MAPI's excellent performance is dependent on both its high mobilities and very long carrier lifetimes and diffusion lengths. 17,48

Defect tolerance
Defect tolerance is the ability for semiconductors to retain strong optoelectronic properties, particularly power conversion efficiency, regardless of the presence of defects, including point defects and grain boundaries. 49One of the proposed mechanisms by which defect tolerance may occur is the presence of antibonding interactions at the valence band maximum (Fig. 1d)-as a result, defects are confined to shallow states at the band edges, rather than deep gap states that may act as traps and recombination centres. 44As Brandt et al. recently highlighted, ions such as Sn 2+ and Bi 3+ are highly likely to have this bonding composition due to the active ns 2 lone pair, and so present excellent candidates for defect tolerant compounds. 37

Dielectric constant and ferroelectric behaviour
Electric response can also be vital to a photovoltaic absorber.A large static dielectric constant has perhaps the most obvious benefit, particularly with regard to some of the aforementioned properties-it confers a high degree of charge screening, resulting in smaller defect charge-capture cross-sections, and inhibits radiative electron-hole recombination.Furthermore, for 'hydrogenic' defects, a large dielectric constant enables smaller defect binding energies promoting shallower defect states. 50As with defect tolerance, large highly polarizable cations like Pb 2+ are likely to lead to high dielectric properties.Ferroelectric behaviour has also been of considerable interest with regards to MAPI's hysteresis; 38,51,52 while current evidence suggests it may not be the primary cause of hysteresis, 53 the prospect of useable photoferroic devices, and high photovoltages from the anomalous photovoltaic effect, remains enticing. 16,546 Rashba splitting 'Spintronics' is an emergent field in condensed matter physics, of which MAPI has seen its share of interest.Recently, Rashba splitting in the MAPI electronic band structure has been implicated in strongly reducing radiative recombination and is considered a possible cause of its high carrier lifetimes.55,56 Indeed, the spin-split indirect gap seen in MAPI is thought to reduce the recombination rate by a factor of more than 350% compared to direct band gap behaviour (Fig. 1e).45 Non-centrosymmetric structures with heavy elements such as bismuth and lead could easily demonstrate similar effects due to due their strong spin orbit coupling.Likewise, multivalley band structures are also thought to increase charge carrier mobilities through separation of charge carriers, in addition to ensuring a high density of states at the band edges, leading to higher absorption coefficients.40,57

Alignment with commonly used contact materials
During heterojunction cell construction, care must be taken to ensure close band alignment of the absorber with its neighbouring materials, such as buffer layers and contacts materials.Efficient band alignment prevents loss of V oc and enables facile carrier transport throughout the cell (Fig. 1f).While layers can be tuned to ensure this, an absorber with typical band positions that align well with ubiquitous components like F-doped SnO 2 would be highly advantageous to reducing the cost and difficulty of manufacture and distribution.

Perovskite structured
As previously mentioned, stability is a major concern limiting the use of hybrid halide perovskites in commercial photovoltaic devices, primarily as longevity is crucial to reach energy payback times. 582][63] Additionally, MAPI possesses poor thermal stability and is known to rapidly decompose at temperatures above 85 1C, 64,65 with research indicating that the MAPI structure is intrinsically unstable with respect to phase separation into CH 3 NH 3 PbI 3 and PbI 2 . 25,66This instability has recently been attributed, with the aid of computational studies, to the low formation energy of MAPI. 67,68As such, modifications of the MAPI formula that are able to increase stability have become highly desirable. 69ne method of tuning MAPI's electronic properties is through changing the organic cation (Fig. 2a).So far, only methylammonium (MA) and formamidinium (FA) have been successfully incorporated into the perovskite structure, with larger cations resulting in lower dimensionality structures due to disruption of the three-dimensional (3D) Pb-I cage. 31,70,71eplacing MA with FA to yield CH(NH 2 ) 2 PbI 3 (FAPI), results in films with a slightly smaller band gap of 1.48 eV, long photoluminescence (PL) lifetimes, lower rates of recombination, high PCEs, and enhanced thermal stability. 72,73Unfortunately, the synthesis of FAPI is complicated by the formation of a thermally accessible hexagonal d-phase, whose large band gap adversely affects device performance. 74,75FAPI films incorporating up to 20% MA show considerable stabilisation of the black a-phase during synthesis and possess long exciton lifetimes and high efficiencies, 76,77 however, the long term stability of these films has not yet been addressed.Alternatively, replacing the organic component of MAPI with an inorganic cation to produce an all-inorganic perovskite is considered a possible route to enhanced stabilities. 78,79Eperon et al. have recently demonstrated a working CsPbI 3 based device by preventing the formation of the weakly absorbing yellow non-perovskite phase. 80,81heir cells showed remarkable thermal stability up to 300 1C but possessed poor efficiencies of only 2.9% and extreme sensitivity to ambient conditions.MAPI can also be tuned through the replacement of iodide with other halides (Fig. 2a). 82In many early cells, using a PbCl 2 precursor incorporated a very small proportion of the chlorine, leading to improved film morphologies due to better distributed heterogeneous nucleation. 48,83,84On the other hand, inclusion of bromide in a solid solution, allows for a tunable band gap 85 and lower levels of hysteresis. 86Based on this Rehman et al. evaluated  This journal is © The Royal Society of Chemistry 2017 the mixed Br/I system, FAPb(Br x I 1Àx ) 3 , for use in tandem solar cells. 87It was noted that the composition needed to form an ideal top-cell band gap of B1.7-1.8 eV (namely x = 0.3-0.5),resulted in apparently ''amorphous'' phases with reduced charge-carrier diffusion lengths, high levels of energetic disorder, and poor optical absorption.This is also been observed in theoretical calculations of the MAPb(Br x I 1Àx ) 3 system, in which the region between 0.3 o x o 0.6 is unstable with respect to spinodal decomposition at 300 K. 88 Recently, McMeekin et al. have demonstrated a mixed-cation mixed-halide perovskite system, in which the phase instability region is subjugated through partial substitution of FA with Cs (Fig. 2b-d). 89The resulting thin films, with the composition FA 0.83 Cs 0.17 Pb(I 0.6 Br 0.4 ) 3 , possessed excellent crystallinity, an optical band gap of 1.74 eV, a high open circuit voltage (V OC ) of 1.2 eV and efficiencies competitive with those seen in MAPI based solar cells (up to 17.9%).The same composition was employed as a semitransparent cell along with a crystalline silicon module, in order to assess the potential of these films for tandem architectures, with an efficiency of 19.8% reported.

Reduced dimensionality perovskites
Layered perovskites have recently become of interest as a route to increased stabilities. 92As previously discussed, incorporating large organic cations in the synthesis of MAPI results in lower dimensionality structures, where the cations cannot fit within the perovskite cage. 32,33In these structures, the large cations are sandwiched between slabs of PbI 6 octahedra, effectively ''capping'' the perovskite layers. 93][96][97] Similar to MAPI, the physical and structural properties of these compounds can be fine tuned through tailoring of the organic cation, halide and metal components. 31,70By adjusting the stoichiometric quantities of the bulky organic cation versus lead iodide (PbI 2 ) and methylammonium iodide (CH 3 NH 3 I, MAI), the number of layers of perovskite octahedra in each slab (n) for the series PEA 2 (CH 3 NH 3 ) nÀ1 Pb n I 3n+1 , can be controlled. 98In this way, the limit n = N represents a cubic 3D perovskite, n = 1 corresponds to a 2D layered structure and n 4 1 describes ''quasi-2D'' perovskite structures (Fig. 2e).
98][99][100][101][102][103][104] Generally, upon moving from the 3D perovskite structure to the 2D n = 1 phase, the optical band gap widens considerably and n-type conductivity decreases.This is partially a result of the high exciton binding energies (often 4300 eV) 101,105 seen in these structures, which arise due to spatial restrictions and dielectric mismatch between the organic and inorganic layers. 106,107In the context of solar cells, this combination of properties is undesirable as it limits V oc and is likely to result in impaired performance.Stability, on the other hand, is found to be highest for the 2D structures and decrease with increasing values of n.As such, the primary challenge presented by reduced dimensionality perovskites is to optimise stoichiometry, so as to balance stability against electronic and optical performance.
One route to producing suitable 2D perovskites is through introducing enhanced halogen or hydrogen bonding at the organic-inorganic interface, which results in a red shifting of the band gap due to a more disperse valence band maximum. 108,109he first layered perovskites put to use in solar cell applications adopted a different approach, instead tuning the band gap through judicious choice of organic cation along with careful stoichiometric control.Smith et al. employed bulky C 6 H 5 (CH 2 ) 2 NH 3 + (PEA) as the organic cation, producing films with 3 perovskite layers (n = 3), an optical band gap of 2.1 eV and reasonably small exciton binding energies (40 meV). 110Their cells displayed efficiencies up to 4.73% and were stable after 40 days in air with 52% humidity.Similarly, Cao et al. produced n = 3 devices containing CH 3 (CH 2 ) 3 NH 3 + via a simple one-step spin-coating process, that exhibited strong light absorption, smooth film surfaces and efficiencies of 4.02%. 111he initial attempts at producing 2D perovskite devices suffered from poor charge carrier extraction that limited their efficiencies. 110,111Recently, Quan et al. have demonstrated the first certified hysteresis-free planar perovskite solar cell, containing a 2D absorbing layer. 96Using a combination of density functional theory (DFT) calculations and complementary studies on physical and optoelectronic properties, the number of perovskite layers was optimised to 40 o n o 60.As expected, van der Waals interactions between the PEA molecules were found to drive increased stability, with the forces acting to reduce the desorption rate of MAI (a key decomposition pathway in MAPI) [112][113][114] by 6 orders of magnitude.Interestingly, the trend in optical band gap (decreasing with increasing n) 93 was found to result from a gradual shift in the conduction band minimum, with the valence band maximum effectively fixed in energy.The resulting quasi-2D devices combined the stability of 2D perovskites with the excellent charge-carrier transport and optical properties of MAPI, resulting in a PCE of 15.3% for the composition where n = 60. 96Another recent paper has succeeded in producing layered perovskite films orientated such that the sheets form perpendicular with respect to the substrate. 115This further motivates the development of layered perovskite as it enables clear carrier extraction pathways from the absorber layer to the device contacts.
The last year has seen the substitution of iodine for thiocyanate (SCN À ) emerge as an avenue to improving the stability of MAPI whilst retaining high efficiencies. 71,116,117Inclusion of small amounts of SCN À , a pseudo-halide with an ionic radius similar to that of I À (217 pm vs. 220 pm), 118 has been shown to produce cells with larger crystal sizes and fewer trap states. 119urthermore, the reaction of MAI and Pb(SCN) 2 produces the compound (CH 3 NH 3 ) 2 Pb(SCN) 2 I 2 (MAPSI), 120 which is similar in structure to the n = 1 layered perovskites but with Pb 2+ octahedrally coordinated to four equatorial I À and two axial SCN À ions (Fig. 2e). 121A debate exists as to the size of MAPSI's optical band gap: several studies have reported a direct optical gap of 1.57 eV, 119,121 however a recent paper by Xiao et al. demonstrated a thin film possessing direct and indirect optical band gaps of 2.04 eV and 2.11 eV, respectively. 122Similarly, Umeyama et al. also report a larger band gap for MAPSI, noting a red-to-black piezochromic response upon compression (Fig. 2g). 91he exact cause of this discrepancy has not yet been elucidated, however it has been suggested that contamination of MAPSI with MAPI may play a role.Similar to other 2D perovskites, the layered structure results in an increased enthalpy of formation, bestowing MAPSI with increased chemical stability. 68Devices containing MAPSI have recorded efficiencies up to 8.3%, with a larger V oc and enhanced stability in air with 95% humidity than comparable MAPI films. 120Additionally, it should be possible to tune MAPSI through incorporating other pseudo-halides (such as OCN and SeCN), halides (Cl, Br, I), and metals (Sn, Ge), opening up the possibility for a range of MAPSI structured analogues. 68

Photoactive organic cations
When the size of the organic cation is increased further still, interconnectivity of the lead iodide octahedral becomes difficult due to large steric effects, and zero-, one-, or twodimensional (0D, 1D, or 2D) structures result. 104,1235][126][127] An emerging route to overcoming such properties is the use of photoactive organic cations, primarily those containing aromatic moieties, which can contribute to the optical response of the material. 128,129Here, shrewd choice of cation is essential to ensure a significant interaction between the organic and inorganic components, and result in broad band spectral absorption. 130,131aughan et al. have recently employed the aromatic cycloheptatrienyl ([C 7 H 7 ] + , tropylium) ion 128 due to its strong absorption, electronic sensitivity (allowing for optical finetuning) and chemical stability, which make it likely to interact favourably with an inorganic system. 132,133The resulting compound, C 7 H 7 PbI 3 , forms a nanowire-type structure composed of edge-sharing PbI 6 octahedra, with the tropylium ions distributed between the wires (Fig. 3a).DFT calculations show that the band gap results from charge transfer between the inorganic and organic components, however, its band gap of 2.15 eV is too large for a high efficiency solar absorber.A similar system has been demonstrated using a significantly bulkier organic ligand, N,N 0 -di(4-pyridyl)-1,4,5,8-napthalene diimide (DPNDI), in which nanowires of face-sharing lead iodide octahedra are encased by an interwoven network of protonated DPNDI molecules (Fig. 3b). 134Beneficial hydrogen bonding and anion-p interactions act to reduce the distance between the ligand and This journal is © The Royal Society of Chemistry 2017 nanowire, enabling a material with an ideal band gap of 1.27 eV and unusually long-lived charge-separated states.The efficient charge-carrier separation is expected to lead to reduced recombination losses, however, the loss in dimensionality results in high effective masses and therefore is likely to adversely affect electron mobilities. 135aterials comprising mixed inorganic and organic connectivity have been suggested as one method of countering poor mobilities.By designing systems in which charge carriers are mobile in both the inorganic network along with an overlapping network of p-stacked photoactive organic ligands, multiple pathways for carrier extraction can be developed. 136his has recently been realised in systems containing 1D lead iodide chains coordinated to large aromatic dipyrido-and benzodipyrido-phenazine compounds, which form eclipsed channels connecting the inorganic chains. 137It was found that photoexcitation effects charge transfer from the lead iodide component to the organic moiety, however the large band gaps of the materials (2.13 eV and 2.52 eV) prevents their use in photovoltaic applications.While it is clear that this system highlights the viability of the technique, further optimisation of the organic component to produce the desired optical response and structure directing effects is essential.

Lead(II) chalcogenides
0][141][142][143] Furthermore, PbS possesses a large Bohr exciton radius of 18 nm, enabling QDs with excellent charge carrier mobilities and a highly tunable band gap. 144,145The band alignment of PbS QDs can be engineered through the use of different ligand treatments 146,147 and has enabled rapid advancement in PbS solar cells, with a maximum certified efficiency of 9.2%. 148bS has also been employed in a core/shell configuration with CdS, which aids surface passivation and has enabled efficiencies up to 5.6% with an enhanced open circuit voltage of 0.66 V. 149 The highly tunable band gap of PbS (from 0.7 eV to 1.5 eV) means it is possible to produce QDs that absorb in the nearor mid-infrared portion of the solar spectrum, which, when combined with hole conducting polymers that absorb in the UV-visible range, allows for broad spectral coverage.150 Lu et al. produced PbS polymer:nanocrystal composites, based on a donor-acceptor polymer, leading to efficiencies of 4.8% with efficient charge separation from PbS to the polymer and enhanced colloidal stability.151 PbS has similarly been employed in perovskite solar cells, functioning simultaneously as a co-sensitiser and hole transporting material, enabling cells with a panchromatic response, high photocurrent densities up to 24 mA cm À2 and efficiencies of 3.6%, despite low open circuit voltages of 0.34 V. 152 By tuning the band alignments of the PbS QDs, a dramatically increased V OC of 0.86 V was achieved, resulting in cells with a PCE of 7.5%.153 Remarkably, due to favourable lattice matching, preformed PbS quantum dots have been incorporated whole into the MAPI structure, producing epitaxially aligned ''dots-in-a-matrix'' crystals (Fig. 3c).138 The resulting heterocrystals possessed a modulated optical response, which, when decomposed, indicated that the properties of the QDs and perovskite matrix remained otherwise unaltered.However, the small band gap of the QDs employed (1 eV) and the highly efficient transfer of photo-excited charge carriers from the perovskite to the PbS nanocrystals, while useful for infra-red applications, limits the applicability of the system with regard to solar power generation.
The closely related compounds, PbSe and PbTe, are also of intense interest as a quantum dot solar absorbers, partly due to their even larger Bohr radii of 46 nm 154 and 152 nm, 155 which when coupled with well established and finely size-tunable synthesis methods, enables precise control over the band gap from 0.4-1.4eV. 156Pb-based QD solar cells are additionally attractive due to their enhanced stability in air when compared to other solar absorbers.Indeed, the maximum recorded efficiency for a PbSe-based QD solar cell of 6.2% was obtained in devices fabricated in an ambient atmosphere, demonstrating the stability of these devices. 157Furthermore, previous studies on PbS have demonstrated its stability in air over 1000 hours of light illumination without device encapsulation. 158As such, it is clear that lead salts such as PbS and PbSe have an important role to play in the future of colloidal quantum dot photovoltaics.The low cost of raw materials 7 combined with their significant versatility guarantees that Pb-based quantum dot cells will remain a highly studied class of materials.

Tin absorbers
Lead-based photovoltaics, while generally the highest performing ns 2 solar absorbers, have come under significant scrutiny due to concerns over the toxicity of lead. 159It is important to note that within the EU, commercial solar cells are exempt from the Restriction of Hazardous Substances Directive and, as such, are not subject to regulation over their lead content. 160Regardless, the intrinsic instability of MAPI 64 presents the risk that toxic material may be released into the environment unless devices are sufficiently encapsulated, necessitating an increase in the cost of device fabrication. 59,161,162The last few years have therefore seen a concerted effort to eliminate the use of lead in photovoltaic devices.In this regard, tin is of considerable interest as it is both cheap and non-toxic, 163 and, being a group 14 metal, is expected to show much of the same coordination chemistry and electronic properties seen in the lead analogues.Consequently, materials containing tin have long been sought after as a route to an earthabundant and non-toxic solar absorber.

Perovskite structured
Due to the successes of the lead hybrid perovskites there has been a significant research effort towards synthesising and evaluating the tin analogues.Unfortunately, the propensity for Sn 2+ to undergo oxidation to Sn 4+ has proved challenging to overcome and still represents a major decomposition mechanism in the tin hybrid perovskites. 39,167Initially, partial substitution of Pb with Sn to form CH 3 NH 3 Sn 0.25 Pb 0.75 enabled a reduction in band gap relative to MAPI, with devices showing efficiencies up to 7.37%. 168,169In an attempt to counteract the low photocurrent densities resulting from poor film coverage, PbCl 2 was added to produce planar CH 3 NH 3 Sn n Pb 1Àn I 3Àx Cl x based cells with 10.1% efficiency. 170In 2016, mixed Pb-Sn perovskite films with the composition CH 3 NH 3 Sn 0.5 Pb 0.5 I 3 achieved efficiencies of 13.6%, demonstrating the potential of devices with reduced lead content to achieve high performance. 171t was originally thought that the presence of Pb was essential to prevent undesirable oxidation of Sn, however, in 2014, two groups were successful in yielding solar cells based on the completely lead-free CH 3 NH 3 SnI 3 . 172,173These devices, with efficiencies of 5.23% 172 and 6.4%, 173 were made possible through rigorous glovebox protocols and ultra-high purity starting materials.CH 3 NH 3 SnI 3 possesses an optical band gap of 1.30 eV 39,74 and, similarly to MAPI, is intrinsically p-type, with relatively high carrier densities (10 19 cm À3 ). 174Devices based on a CH 3 NH 3 SnI 3 /SiO 2 nanocomposite architecture have shown impressive open circuit voltages up to 0.88 V, only 350 mV less than the thermodynamic limit. 173However, small carrier diffusion lengths (30 nm) along with poor long-term stability present a significant obstacle to overcome before these devices can find practical applications. 90n 2015, Koh et al. revealed that inclusion of SnF 2 in the synthesis of formamidinium tin iodide (HC(NH 2 ) 2 SnI 3 , FASI) counteracts the facile oxidation of tin by promoting the reduction of Sn 4+ to Sn 2+ , however, this increased resilience to oxidation occurs at the cost of decreased conductivity.This technique enabled, for the first time, the production of FASI-based solar cells with power conversion efficiencies up to 2.1%.FASI has an ideal band gap of 1.4 eV and, unlike the analogous lead compound, FAPI, does not possess any other thermally accessible phases that can cannibalise device performance. 74,77,176Indeed, the stability of FASI has been demonstrated up to 200 1C, well above maximum device operating temperatures. 175Recently, this technique has been extended to prevent phase segregation on the surface of the films through the formation of a SnF 2 -pyrazine complex, yielding devices with power conversion efficiencies of 4.8% (Fig. 4a).These films displayed smoother surface morphologies and exhibited longer recombination lifetimes than reference cells prepared in the absence of pyrazine.Impressively, encapsulated devices stored in ambient conditions showed stable performance for over 100 days, with a loss in efficiency of only 2% over this period.Currently, however, a combination of low open circuit voltage (0.32 V) and fill factor (63%) prevent these devices from achieving performance comparable with other perovskite technologies. 164he issue of unwanted oxidation has also plagued the all-inorganic perovskite, CsSnI 3 .First used as a solid-state hole transporting material in dye-sensitised solar cells, 177 CsSnI 3 is a p-type semiconductor with a band gap of 1.30 eV. 178,179The first planar heterojunction devices employing CsSnI 3 as the absorber material demonstrated power conversion efficiencies up to 0.9%, but were inhibited by poor stability and the polycrystalline nature of the films. 180Subsequent studies have, similarly to the other tin perovskites, employed SnF 2 as an additive to control the oxidation state of tin, enabling efficiencies of 2.0% and high photocurrents of 22.7 mA cm À2 . 165,181In these cells, performance was severely restricted by low open circuit voltages of 0.24 V, about 4 times less than those seen in typical MAPI devices. 182,183This has been partially mediated through the substitution of iodine with bromine to form CsSnI 3Àx Br x , enabling higher open circuit voltages due to blue shifting of the band gap (Fig. 4b). 165Unfortunately, unlike in the lead hybrid perovskites where inclusion of bromine has been shown to increase stability in humid conditions, 85,184 no such resistance is conferred in the case of CsSnI 3 , with the oxidation of Sn 2+ to Sn 4+ remaining the primary chemical instability.Additionally, alongside problems relating to unwanted oxidation, CsSnI 3 also possesses an alternative 1D yellow phase, with a band gap of 2.6 eV, formed under exposure to air or organic solvents. 178,185,186As such, it is clear that significant work is needed to improve the stability of CsSnI 3 based devices if they are to find practical applications.
This journal is © The Royal Society of Chemistry 2017

Tin(II) chalcogenides
8][189][190][191] However, despite these properties and over 20 years of active research, 192 the efficiency of SnSbased devices has yet to reach 5%. 193,194Nonetheless, the abundance of tin and sulphur, combined with the ease of materials processing and a maximum predicted efficiency of 32% (as per the Shockley-Queisser limit), 195 make SnS attractive as a candidate photovoltaic absorber.0][201] SnS is intrinsically p-type, with low hole effective masses (m h * = 0.2 m 0 in the a and b directions; m h * = m 0 along c) and is generally employed in a p-n junction architecture using ZnO or CdO as the n-type layer. 190Unlike PbS, the small Bohr radius of SnS (7 nm) precludes its use in quantum dot solar cells. 202everal distinct phases of tin(II) monosulfide are known, 188,[203][204][205] with the most stable room temperature structure being herzenbergite, a distorted rock salt structure (a-SnS, Pnma). 166The distortion of the lattice from a cubic to orthorhombic structure is due to the strong Sn 5s 2 lone pair (following the revised lone pair model) 206,207 which repel and result in the formation of a layered system held together by weak covalent interactions (Fig. 4c).This 2D structure is the source of the anisotropy seen in the hole effective masses. 190Thin films of SnS have been produced by a wide range of deposition techniques, including chemical bath deposition (CBD), [208][209][210] atomic layer deposition (ALD), 211,212 spray pyrolysis, [213][214][215] and thermal evaporation, 216,217 of which the latter two techniques are most common.Choice of deposition method can have a profound impact on the optoelectronic properties of the films, with CBD, ALD and vapor transport deposition generally producing films with the largest carrier concentrations and highly mobile holes. 188,218][221][222] Sinsermsuksakul et al. have reported the most efficient SnSbased solar cell to date, producing a cell with a PCE of 4.4%. 193heir device, along with the other highest performing SnS solar cells, employed Mo as the metallic contact and ZnO as the n-type transparent electrode, with the addition of an optimised Zn(O,S) buffer layer necessary to achieve high performance. 223,224nterestingly, Burton et al. have shown Mo to be unsuited as a back contact material due to the small ionisation potential of SnS (4.7 eV), instead suggesting metals with lower workfunctions, such as Ti, W, or Sn, as optimal contact materials. 225Other reports, based on impedance spectroscopy measurements, indicate Mo performs poorly due to increased series resistance and high rates of tunnelling assisted recombination and have recommended Cu as a viable alternative. 226here has been some controversy surrounding the reports of zincblende structured SnS (F% 43mm) deposited as microparticles and thin-films (Fig. 4c). 188,194,203,227Such a material would allow for extended compatibility with existing II-VI and III-V tetrahedral semiconductor technologies due to favourable lattice matching at the interface between the absorber material and window layers.However, there have been few subsequent reports of SnS adopting the zincblende structure and theoretical calculations reveal the phase to be unusually high in energy, with Burton et al. proposing that such structures have been mis-assigned zincblende instead of distorted-rocksalt. 166ecently, Nair et al. have provided evidence for a cubic SnS structure which they suggest helps to characterise those structures previously assigned as zincblende. 228Based on a previous single-crystal electron diffraction report that demonstrated cubic SnS (P2 1 3), 229 the authors have produced SnS thin films with a direct band gap of 1.66-1.72eV that can be formed via a number of synthetic routes.Nair et al. have also reported an analogous cubic SnSe compound, with a direct optical band gap of 1.4 eV, considerably larger than the 0.95-1.11eV indirect gap of the rocksalt SnSe structure, and in the ideal range for a highly efficient solar cell. 230As of yet these cubic phases have yet to be incorporated into functioning photovoltaic devices, however, development of these materials is of obvious interest to the field.

Perovskite structured
Germanium, another group 14 metal, has received relatively little attention as a replacement to lead, in part due to its strong lone pair effect, which acts to produce lower dimensionality structures with poor conductivities.Regardless, germanium is non-toxic and relatively earth-abundant and is therefore an ideal candidate for novel solar absorbing materials. 2313][234][235] They crystallise in a rhombohedral structure (space group R3m) with band gaps of 3.1 eV, 2.3 eV, and 1.6 eV for X = Cl, Br, and I, respectively. 236,237nterestingly, CsGeI 3 does not possess any other thermally accessible phases and is stable up to 350 1C, 238 with devices recording efficiencies up to 3.2%. 239However, the stability of Ge 2+ is significantly reduced due the low binding energy of its 4s 2 electrons, which increase its sensitivity to oxidation. 90As such, rigorous glovebox protocols and suitable device encapsulation are essential for efficient device performance. 240ecently, two groups have reported a range of germaniumbased hybrid perovskite materials with properties suitable for solar cell applications (Fig. 5a). 238,240It was found that replacement of the alkali metal acts to increase the band gap in line with the size of the cation, from B1.9 eV and B2.2 eV in the case of CH 3 NH 3 GeI 3 (MAGI) and HC(NH 2 ) 2 GeI 3 (FAGI), up to 2.8 eV when trimethylammonium was employed ((CH 3 ) 3 NHGeI 3 ). 240,242ncorporation of cations larger than formamidinium results in highly distorted structures due to the stereochemical activation of the 4s 2 lone pair. 240This reduces the bandwidth of the frontier bands and leads to a significant widening of the band gap, resulting in semi-transparent materials with extremely poor absorption.The large band gaps seen in these materials are not expected to produce efficient standalone solar cells and are instead more relevant as top layer absorbers in tandem devices.Despite this, MAGI has been incorporated into a functioning photovoltaic device but showed poor performance due to low open circuit voltages (0.15 V) thought to result from formation of Ge 4+ , with a maximum efficiency of 0.2% obtained. 238Overall, if germanium perovskite solar absorbers are to succeed, significant development of chemical methods to prevent unwanted oxidation is essential.One possibility is through layered perovskite structures, which, analogously to the lead equivalents, may gain some resistance to oxidation through van der Waals interactions acting on the surface of the GeI 6 octahedra. 101

Germanium(II) chalcogenides
The germanium chalcogenides, GeS and GeSe, are also of interest as ns 2 solar absorbers due to their ideal band gaps of 1.6-1.7 eV and 1.1-1.25][246] Whilst both compounds, similarly to their tin analogues, possess indirect fundamental band gaps, their direct gaps are only slightly larger (B0.08 eV) due band structures which contain multiple band extrema close in energy (Fig. 5b). 247,248Both are intrinsically p-type semiconductors with large hole mobilities of 110 cm 2 V À1 s À1 , 249,250 a similar magnitude to those found in MAPI single crystals (90 cm 2 V À1 s À1 ). 48Furthermore, they are both solution processable which allows for the possibility of cheap and low cost device manufacturing through roll-to-roll printing. 245However, while GeSe has recently gained attention as a possible record breaking thermoelectric material, 251 there are, to our knowledge, no reports of functioning germanium chalcogenide-based solar devices in the literature.
This journal is © The Royal Society of Chemistry 2017 However, this precludes direct integration into the AMX 3 formula-instead, we review the wide variety of antimony compounds that have been studied for photovoltaics.

Antimony chalcogenides
The binary antimony chalcogenides have seen a great deal of research interest over the past 35 years, yet despite this interest, cell efficiencies remain below 10%.Both Sb 2 S 3 and Sb 2 Se 3 share a crystal structure consisting of 1D nanowires formed from Sb 4 Ch 6 units, with the nanowires bound by weak Ch-Ch interactions.The Sb s 2 lone pair is stereochemically active, forming a square pyramidal coordination around Sb, with the lone pair occupying space in-between the nanowires. 254,255This low dimensionality may be expected to reduce the overall functionality as an absorber material, however despite this, Sb 2 S 3 and Sb 2 Se 3 have both seen significant interest from the PV community.
1][262][263] Since, Sb 2 S 3 -sensitized cells have seen much work, as Seok and others have made recent strides in bringing efficiencies up to 7.5% [264][265][266][267] in inorganic-organic heterojunction cells.This has largely been through the reduction of trap sites, however, there still remains a large V oc deficit limiting the growth of these cells.Measurements at low light have seen higher efficiencies of 8-9%, however the lack of comparable measurements with other materials makes this difficult to compare. 268With a high near-direct band gap of near 1.7 eV, and a high absorption coefficient, 269,270 and multiple deposition techniques available, 271,272 Sb 2 S 3 remains a promising candidate material.First principles calculations have replicated the experimental band gap well, using hybrid DFT and meta-GGA functionals, 273,274 and have identified that the valence electrons remained confined to the individual chains, but are diffuse within them: this dispersion indicates that the charge carriers should remain mobile. 275,276espite the relative success of Sb 2 S 3 , within the last few years, there have been movements to develop the isostructural antimony selenide instead.Sb 2 Se 3 has a lower band gap, around 1.1-1.30][281] This high absorption may disguise a fundamental indirect band gap, which is only slightly lower in energy, as observed via DFT, 282 quasiparticle band structure calculations, 276 and UV-vis 283 and so, like the sulfide, Sb 2 Se 3 has been referred to as a 'effectively direct band gap' semiconductor.Additionally, theoretical band alignment with TiO 2 indicates that with improvements in V oc , the lower band gap (and thus high possible J sc ) of Sb 2 Se 3 compared to Sb 2 S 3 could lead to higher attainable efficiencies in pnictogenchalcogenide sensitised solar cells. 284ther than a o1% efficiency cell in 2002, 278 almost all advances in Sb 2 Se 3 cells began in 2014, with the publication of a TiO 2 -sensitized inorganic-organic cell of 3.2% from Choi et al. 285 and a 1.9% thin film CdS/Sb 2 Se 3 cell. 286Since then, Tang and co-workers have driven the development of the thin film architecture, reaching 3.7% by the end of 2014, 287,288 and 5.6% in 2015. 243,289Additionally, they have indicated that vertically aligned growth of the 1D structure of Sb 2 Se 3 means that high carrier transport can be maintained throughout the Sb 2 Se 3 layer, while grain boundaries in the material will form parallel to the chains, making them inherently benign, reducing a major source of losses in polycrystalline cells (Fig. 6a). 243This applies to many other antimony and bismuth(III) materials, including Sb 2 S 3 but also SbSI and the other members of the V-VI-VII series, where the stereochemically active lone pair enforces a 1-D topology-this presents a major advantage for the future development of these materials in photovoltaics.
In addition to the binary sulfides, a smaller section of work has examined the solid solution between the sulfide and selenide-with a tunable band gap between 1.3-1.7 eV, 290 and high absorption in both binary systems, this presents an opportunity for tunable photovoltaic materials.Despite low initial efficiencies below 1%, 291 recent work has used the advances in deposition techniques and the adoption of poly(3-hexylthiophene) (P3HT) as an efficient hole transporting material to attain an efficiency of 6.6%, close to the most successful Sb 2 S 3 cells. 292

Antimony chalcohalides
The ternary antimony chalcohalides adopt a 1D structure very similar to that of the simple chalcogenides mentioned above: the antimony atoms occupy a pseudo-square pyramidal configuration with sulfur and iodine in a dimeric unit that infinitely repeats along a, with the stereochemically active lone pair occupying the final position in the Sb octahedron. 293What distinguishes SbSI from the binary chalcogenides, however, is its photoferroicity: SbSI undergoes a ferroelectric phase transition, 294 giving a phase with spontaneous polarization, allowing for anomalous photovoltaic effects beyond that of usual semiconductors, with the potential for very high V oc .Walsh and co-workers discuss this in their 2015 work, 54 also calculating its electronic structure-like Sb 2 S 3 , it possesses a slightly indirect band gap of around 1.9 eV, in concordance with experiment, 295 but also low effective masses.Despite this, as the authors highlight, SbSI and SbSeI have so far been passed over for other materials.In 2016, Walsh and co-workers revisited the V-VI-VII compounds, including SbSBr, using quasiparticle GW theory, obtaining mostly similar band gaps in addition to band alignments-this should allow for semiconductor contacts to be tailored for these materials. 296

Copper antimony chalcogenides
In contrast, the ternary copper antimony chalcogenides have seen a great deal of interest in photovoltaic applications.CuSbS 2 has seen perhaps the most attention-Nair and coworkers investigated p-type CuSbS 2 in combination with intrinsic Sb 2 S 3 to create a p-i-n device in 2005, 297 however the early stage of development meant both the J sc and V oc were relatively low.Fundamentally, however, CuSbS 2 displays suitable electronic properties, with a band gap measured around 1.5 eV 298 and good carrier mobilities of 20 to 49 cm 2 V À1 s À1 , although high conductivity has been found to be strongly dependent on the precise composition. 253,2991][302] Its crystal structure, together with CuSbSe 2 , is related to the 3D chalcopyrite structure of the successful photoabsorber CuInSe 2 , however the greater stereochemical effect of the Sb s 2 lone pair causes a distortion, resulting in a layered structure (Fig. 7a). 303Walsh et al., using DFT calculations, have discussed the electronic and structural effect of the lone pair in these materials: in concordance with the revised lone pair effect model for posttransition metal compounds, 207 the antimony s states mix with the sulfur p states to allow further mixing with antimony p states, leading to antibonding states close to the top of the valence band and the structural distortion (Fig. 6b).This study also finds the band gaps to be indirect, although like the binary chalcogenides, only by 0.1-0.2eV.This has been supported by subsequent theoretical work, 304,305 which have additionally evaluated high absorption coefficients (410 5 cm À1 ).Yu et al. corroborated these results, and used them to indicate that CuSbS 2 and CuSbSe 2 , by possessing relatively flat band edges with a high Density of States there, will give stronger absorption than the well-known absorber CuInSe 2 . 57Using their metric of spectroscopically limited maximum efficiency (SLME), which improves upon the Shockley-Queisser limit by including non-radiative effects and film absorption, the SLMEs of CuSbS and CuSbSe 2 were 23% and 27% respectively, comparable to or higher than that of CuInSe 2 .Cell efficiencies have, so far, have been unable to reach this level: CuSbS 2 has improved significantly from that initial cell with multiple cells, either with CdS or sensitizing TiO 2 , now exceeding 3%, [306][307][308] and CuSbSe 2 has matched this, with cells of 1.3% and 3.5% 309,310 published so far.However, recently two strategies have seen improvements in CuSbS 2 cells: annealing in Sb 2 S 3 vapour followed by KOH etching was seen to almost double J sc , V oc and fill factors, while using a co-evaporation fabrication method lead to the highest recorded open-circuit voltage of 526 meV. 311,312With the suitable optical and electronic properties seen above, and Yang et al. finding few deep defect levels to act as recombination centres, further growth in this area may well be possible (Fig. 7b).perovskite unit are vacant, creating layers of corner-sharing Sb-I octahedra with a small inter-layer separation.This structure was found theoretically and experimentally to have an indirect 2.05 eV band gap, with carrier effective masses close to 0.5m 0 and a high absorption coefficient.However, their prepared thin film cells displayed a V oc of B0.3 eV and low efficiencies, theorised to be a result of multiple deep defect levels effecting facile non-radiative recombination.Study of the methylammonium hybrid antimony iodide, (CH 3 NH 3 ) 3 Sb 2 I 9 , has been primarily limited to spectroscopic studies on the dynamics of the methylammonium cation, 314,315 however 2016 has seen the first production of a cell by Hebig et al., demonstrating an efficiency of 0.5%, improving upon that of the cesium compound. 316The cell demonstrated a much higher V oc of 0.90 eV, however a low J sc of 1.0 mA cm À2 plus an Urbach tail energy of 62 meV indicating significant disorder in the amorphous absorber layer, suggests optimization of synthesis and cell construction will be required to produce viable cells.Beyond this, work on hybrid antimony iodides remains limited: while some lower dimensionality perovskites have been reported, electronic characterisation is scarce. 317,318The antimony sulfides Cs 3 Sb 8 S 13 and (MA) 2 Sb 8 S 13 have also come to attention recently in the work of Yang et al., 273 who examined these two compounds with hybrid DFT.Both compounds demonstrate significant similarities with the parent Sb 2 S 3 phase, such as the stereochemically active lone pair, similar, if higher band gaps, and a multivalley band structure-which should lead to low radiative recombination, while retaining a high optical absorption.

Bismuth absorbers
Bismuth is particularly notable as a heavy metal as, unlike its neighbours of lead, polonium and thallium, it demonstrates very little evidence of toxicity. 319,320As such, it has seen increasing use in catalysis and organic synthesis as 'green chemistry' becomes ever more crucial; 321,322 the rise of the lead halide perovskites in the perovskite community has lead to some considering whether bismuth compounds can reach a similar level of success. 129

Bismuth sulfide
Bismuth sulfide, like the binary antimony chalcogenides, has had a long research history within the solar community, with the first thin films and photoelectrochemical cells being developed in the 1980's. 323,324Most research on bismuth selenide has focused on its rhombohedral topological insulator phase, 325 as the 1.25 eV orthorhombic phase has only recently been isolated in a single-phase. 326The use of chemical deposition techniques to produce bismuth sulfide has seen considerable interest since, [327][328][329] as offering a low temperature, facile synthesis is highly desirable for device manufacture, although the quality and crystallinity of films may suffer compared to evaporation techniques. 330,331Nevertheless, chemically deposited thin films of Bi 2 S 3 have been used in combination with lead chalcogenides to produce cells of 0.5% efficiency in 2011, 332 and 2.5% in 2013. 333With an optical band gap measured between 1.3-1.6 eV experimentally 332,334 and through GW calculations, 276 a high absorption coefficient, and low toxicity elements, Bi 2 S 3 is a promising material for PV (Fig. 8a).Bi 2 S 3 is intrinsically n-type, 335 and hence has been particularly successful in heterojunctions with traditional p-type materials, such as crystalline silicon, 336 or PbS quantum dots; the latter devices have neared 5% efficiency with the 'bulk-nano' heterojunction architecture leading to much longer carrier lifetimes and quadrupled J sc in comparison with a bilayer architecture. 337,3382][343] In particular, hybrid solar cells containing nanocrystalline Bi 2 S 3 in combination with the organic absorber P3HT have seen great progress from the group of Konstantatos and others, with cell efficiencies rising from below 1% in 2011 to 3.3% in 2015, [344][345][346] and the exploration of size-dependent tunability and passivation of surface defects. 347,348

Bismuth iodide
Like the antimony and bismuth chalcogenides and chalcohalides, bismuth iodide, BiI 3 , was recently highlighted by Brandt et al. as a potential solar absorber material due to its suitable band gap and the likelihood of good defect tolerance. 37rior to this, its involvement in photovoltaics has been limitedmost research has focused on its use as a radiation detector, 352,353 with a single report of its use as a hole-transport layer in organic solar cells, where its performance was observed to be comparable to PEDOT:PSS. 354Occupying a layered rhombohedral crystal structure, 355 its band gap was the subject of considerable variance, with values ranging from 1.99 eV with ellipsometry to 1.43 eV with DFT calculations (Fig. 8b). 356,357Podraza et al. were able to rationalise this through the low optical absorption obscuring the actual indirect band gap of 1.67 eV, in close agreement with other calculations. 358,359With a suitable band gap for PV applications, it has thus been the subject of study by Seshadri and co-workers, producing a cell with 0.3% efficiency; the V oc of this cell was noted to be low due to the poor alignment of BiI 3 with the hole and electron transporting layers. 349,350Brandt et al. have also further studied this compound, finding a much higher absorption coefficient than previously recorded, however photoluminescence revealed an estimated, short carrier lifetime of 180-240 ps which, in combination with a high resistivity of B10 8 O cm À1 , is likely to severely hinder conductivity. 360Recent computational study of BiI 3 monolayers, however, predicted them to be stable, and show significantly improved IR-absorbance in a bilayer with graphene, raising the possibility of photoactive heterostructures. 361

Bismuth chalcohalides
With analogous 1D crystal structures to their antimony counterparts, 362-364 together with suitable, slightly indirect band gaps of 1.59 eV and 1.29 eV respectively, 365 it is understandable that BiSI and BiSeI should also be considered for photovoltaic applications (Fig. 8c).However, despite interest in its photoconduction and photovoltaic behaviour in the 1960's, 366,367 it is only very recently that the bismuth chalcoiodides have been thoroughly assessed in this manner.In 2012, Mullins and co-workers published two studies on this topic-the first of these was a thorough characterization study of BiSI thin films, finding a high absorption coefficient (5 Â 10 4 cm À1 ), photocurrents of up to 5 mA cm À2 under AM1.5G illumination and a maximum V oc of 370 mV with a I À /I 3 À couple. 368The second examined the effect of selenium doping, allowing for a decrease in band gap of 0.15 eV, and attempted both photoelectrochemical and solid-state cells, with 0.25% and 0.01% efficiency respectively, 369 however BiSI was observed to degrade under electrochemical conditions.Recent theoretical work has explained the poor solid-state behaviour: the conduction band alignment of BiSI means that the choice of p-CuSCN as the hole transporting material limits V oc considerably, and there will be poor electron transfer to the FTO electrode, as used by Mullins and co-workers. 351As such, there should be great potential for improvement-with antibonding states at the valence band maximum, and a high static dielectric constant recently evaluated, 370 BiSI films should possess the high degree of defect tolerance identified by Brandt et al. as a highly desirable PV property. 37hile the bismuth oxyhalide family (BiOX, X = Cl, Br, I) have been investigated for a variety of applications, most notably as photocatalysts for organic pollutant decomposition or water splitting, 371-374 a small number of reports have examined BiOI for photovoltaic applications.BiOI has the lowest band gap of the family, measured between 1.6 eV and 2.0 eV, 372,375,376 meaning that while it demonstrates the weakest photoelectrochemical behaviour, 377 it is also likely the best candidate for photovoltaics.The first attempt at a cell in 2009, using a iodide couple in ethylene glycol attained a moderate V oc of 461 mV, but a very low short circuit current (20.4 mA cm À2 ), likely due to the poor conduction between the BiOI and its chitosan matrix.Directly coating the FTO electrode with BiOI was observed to increase J sc by a factor of 10, 379 yet the most significant improvement was seen recently with a similar cell architecture, with a J sc of 3.8 mA cm À2 and an efficiency of 1.03%. 380Interestingly, BiOI has also seen use recently in combination with Bi 2 S 3 as the absorber layer-while short-circuit currents were again above 1 mA cm À2 , low fill factors meant efficiencies were low.Attempts to utilise BiOCl and BiOBr in other DSSCs have also met with low currents and efficiencies. 382

Noble metal bismuth chalcogenides and halides
CuBiS 2 , like the copper antimony chalcogenides, has seen substantial theoretical interest within the last four years, 40,304,[383][384][385][386] with much attention paid to the structural effect of the bismuth lone pair, and the high absorption coefficient (B10 5 cm À1 )-the band gaps obtained (1.5-1.6 eV) also correlate well with experimental measurements (1.6-1.8 eV). 387,388Additionally, CuBiS has been shown to demonstrate moderate hole mobilities of 410 cm 2 V À1 s À1 , 389,390 and although previously assigned as an n-type conductor, 387,388 recent theoretical and experimental work strongly suggests it is p-type. 385,390 be passivated in a heterojunction with In 2 Se 3 , 393,394 and its excitonic behaviour through Raman and Photoluminescence measurements. 3957][398] Most recently, a photoelectrochemical energy conversion efficiency of 1.28% was obtained in a Cu 3 BiS 3 -sensitized TiO 2 cell, giving a strong indication of the potential of these materials. 399Cu 4 Bi 4 S 9 has also seen particular success as a sensitizer to In 2 Se 3 , obtaining cells with efficiencies between 3.9% and 6.2% with a variety of oxide contacts, with the Cu 4 Bi 4 S 9 addition noted to particularly improve the surface photovoltages. 400ost notably, however, AgBiS 2 has seen a massive stride in photovoltaic efficiency very recently, much more so than its copper analogues.Also unlike CuBiS 2 , nanocrystalline-and high temperature bulk-AgBiS 2 occupies a 3D cubic disordered rocksalt structure, 401,402 which has seen previous interest due to its high temperature order-disorder transition and its effect on thermoelectric behaviour. 403,404While the bulk has been recorded to have an optical band gap of 0.9 eV, 405 strong quantum confinement effects have meant that quantum dot thin-films have band gaps reported at 1.0-1.3][407] Despite this, initial AgBiS 2 QD-sensitized TiO 2 cells reported low efficiencies of 0.53% and 0.95%, both of which suffered from low fill factors and low current densities, 402,408 while AgBiS 2 QDs have also been seen to improve on Pt as a contact material. 409Recent work by Bernechea et al. however, using tetramethylammonium iodide-treated AgBiS 2 nanocrystals in a p-i-n junction between PTB7 and ZnO, has recently given a record cell of 6.3% efficiency, equaling many of the record antimony and bismuth-based cells, with a high J sc of 22 mA cm À2 and FF of 0.63. 410A relatively low V oc of 0.45 eV was attributed to trapassisted recombination, and so improvements in synthesis could easily lead to significantly easier charge transport and even higher efficiencies in the future.
Also notable is the recent production of a cell based on cubic AgBi 2 I 7 , a compound that has only been sparsely characterised previously. 411Sargent and co-workers investigated the AgI-BiI 3 solid solution, producing a thin film of AgBi 2 I 7 , and reporting a direct band gap of 1.86 eV. 412A subsequent nanocomposite cell, with AgBi 2 I 7 between mesoporous TiO 2 and P3HT, gave an efficiency of 1.12%, with a high fill factor of 0.67, but limited by relatively lower current density and voltage.As with many of the above compounds, further attention could lead to significant improvements in the future.

Cesium and hybrid bismuth iodides
The cesium bismuth iodide Cs 3 Bi 2 I 9 and its methylammonium counterpart, (CH 3 NH 3 ) 3 Bi 2 I 9 , have come under attention similar to the antimony analogue.While past studies have performed detailed study concerning the low temperature (B200 K) ferroelastic phase transition of Cs 3 Bi 2 I 9 , [413][414][415] and the vibrational properties of (CH 3 NH 3 ) 3 Bi 2 I 9 , 416 thorough electronic characterization of the room temperature phases has only been performed recently.This structural behaviour may still be relevant, however, given that exciton-phonon coupling in the bromide analogue, Cs 3 Bi 2 Br 9 , has been shown to lead to a much stronger exciton binding energy than the lead halide perovskites. 417nlike the layered Cs 3 Sb 2 I 9 , the bismuth compounds occupy the 'zero dimensional' Cs 3 Cr 2 Cl 9 structure, with isolated [Bi 2 I 9 ] 3À dimers of face-sharing octahedra surrounded by the free Cs + or (CH 3 NH 3 ) + cations, [418][419][420] raising the possibility of lower carrier mobilities.Indeed, Lehner et al.'s theoretical and experimental characterisation of Cs 3 Bi 2 I 9 found much lower band dispersion in comparison with MAPI, indicative of low mobilities, together with a relatively high indirect optical band gap of 1.9 eV (Fig. 8d). 349Nevertheless, Cs 3 Bi 2 I 9 was used as an absorber on mesoporous TiO 2 by Park et al. to give a best cell with 1% efficiency and a V oc of 0.85 eV, 421 a significant improvement on the equivalent (CH 3 NH 3 ) 3 Bi 2 I 9 cells which had efficiencies below 0.2% and much lower V oc and J sc .Other attempts to use (CH 3 NH 3 ) 3 Bi 2 I 9 since, with alternate organic hole transporting layers, have seen small improvements in J sc 422,423 and fill factor, 424 but efficiencies remain low, despite a similar band gap of 2.1 eV and strong absorption. 425,426uonassisi and co-workers also studied the photoluminescence behaviour of (CH 3 NH 3 ) 3 Bi 2 I 9 , finding a decay time of 760 ps-up to 10 3 lower than the highest recombination lifetime in MAPI films, however it is noted that a o1 ns lifetime is comparable to the initial values recorded for popular absorbers such as Cu 2 ZnSnSe 4 and SnS. 425A low PL quantum efficiency of 0.4% was also recorded and attributed to non-radiative recombination pathways; Park et al. found defect states in the XPS of Cs 3 Bi 2 I 9 that suggest it may be similarly limited. 421It is evident that the photovoltaic behaviour of both materials may be enhanced if further work to reduce the impact of these pathways is performed, and, as both studies also found that the bismuth iodides were stable in air over a month, these compounds do present a potential avenue for hybrid bismuth PV materials.(CH 3 NH 3 ) 3 Bi 2 I 9 is not the only inorganic-organic bismuth iodide to have been examined for photovoltaic activity, with the one-dimensional hexanediammonium bismuth iodide, (H 3 NC 6 H 12 NH 3 )BiI 5 , recently trialled in a mesoporous TiO 2 architecture.Like the bismuth iodides above, these devices also possessed low V oc and J sc , and a band gap close to 2 eV, however they also showed greater film coverage and thermal stability than MAPI. 429

Double perovskites (elpasolites)
Very recently, there have been significant efforts to move beyond the A I M II X 3 perovskite formula to the double perovskite A I 2 M I M III X 6 .Crucially, this allows the expansion of the perovskite motif beyond Group 14 ions on the M site while retaining the successful 3D perovskite structure (Fig. 9a).While double perovskites in the oxide family are well known as fuel cells, 433,434 only a few relevant halides such as Cs 2 NaBiCl 6 , with absorption much higher in energy than the ideal photovoltaic range, are known. 435,436In 2015, however, Giorgi and Yamashita proposed a simple substitution of Tl + and Bi 3+ , in place of Pb 2+ , into the tetragonal MAPI structure using DFT, finding a theoretical bandgap of 1.68 eV, compared with MAPI's 1.62 eV using the PBE functional. 437However, as the authors admit, Tl presents an environmental risk and its possible replacement by its neighbour In I would likely lead to instability issues, similar to Sn II .
Instead, at the beginning of 2016, multiple separate groups have published on the Cs 2 AgBiX 6 family.Slavney et al. synthesised Cs 2 AgBiBr 6 and discuss its photoluminescence behaviour, finding a long decay time of 660 ns which is attributed to a long recombination lifetime dominated by non-radiative recombination, similar to that found in MAPI, and estimating the indirect band gap at 1.95 eV (Fig. 9b). 428McClure et al. were able to synthesise both the bromide and chloride, finding indirect bandgaps of 2.19 eV and 2.77 eV respectively from diffuse reflectance spectroscopy and supporting these observations with DFT calculations, finding hole effective masses slightly lower than those in the analogous methylammonium lead bromide. 452dditionally, both studies examined the stability of these materials: Slavney et al. found the bromide stable up to 430 1C and no evidence of degradation under either moisture or light after 30 days, however, while McClure et al. found no significant decomposition in the chloride, the reflectance and XRD pattern of Cs 2 AgBiBr 6 was observed to degrade when stored in light and air.Volonakis et al. have performed a wider theoretical screening of the double perovskites using hybrid DFT, examining M I = Cu, Ag, Au and M III = Bi, Sb and additionally synthesising Cs 2 BiAgCl 6 to support their observations. 427Of those predicted, a linear decrease is observed in all cases down the halide group, while, for M I , the Ag compounds have higher band gaps, and Au the lowest; as such, Cs 2 AgSbI 6 , Cs 2 BiCuI 6 and Cs 2 BiAuBr 6 are those observed to have band gaps within the optimal 1.0-1.5 eV range.Further work from the Giustino group has seen the accurate prediction of the band gaps of Cs 2 BiAgCl 6 and Cs 2 AgBiBr 6 using quasiparticle GW theory. 453Recent work by the Cheetham and co-workers has seen the synthesis of the methylammonium double perovskites (MA) 2 KBiCl 6 and (MA) 2 TlBiBr 6 : however, (MA) 2 KBiCl 6 's indirect band gap of 3.04 eV is too high for efficient use in photovoltaics, and while the 2.16 eV bandgap of (MA) 2 TlBiBr 6 is direct, the inclusion of toxic thallium, as discussed above, would minimize the benefit of removing lead. 454,455It is clear that while the double perovskite motif allows the replacement of Pb, their wide, indirect band gaps and the potential issue of stability may be problematic for further PV applications.
This split-ion approach for producing bismuth doubleperovskites has similarly been proposed on the anion site, with a mix of chalcogenides and halides: DFT calculations predicted the hypothetical CH 3 NH 3 BiSI 2 to have a similar band gap, dielectric behaviour and effective masses to MAPI. 456However, another wide-ranging theoretical study on the analogous inorganic AM(Ch II X I ) 3 (A = Cs, Ba, M = Sb, Bi, Ch = O, S, Se, Te and X = F, Cl, Br, I) compounds has found that all were unstable to decomposition, and the authors' attempts to synthesise these compounds found no evidence that the target phases were produced. 28As Hong et al. reflect, it is clear that when such alternatives, particularly quaternary compounds with a larger number of competing phases, are predicted, assessments of their stability should not be neglected.

Further inorganic-organic bismuth compounds
Further to those previously mentioned, there exists a massive library of extant hybrid inorganic-organic bismuth and antimony compounds due to the surge in interest in the 1990's and early 2000's.Several reduced dimensionality bismuth iodides exist, yet are relatively under-characterised: these include guanidinium, iodoformamidinium and a variety of alkyl and photoactive-cation bismuth iodides, many of which were synthesised by David Mitzi and his co-workers. 100,103,317,457-461Indeed, there are still many new bismuth compounds being discovered even this year, 462 some of which may already be promising PV candidate materials. 463,464Given the diversity of coordination, dimensionality and organic character in these examples alone, it is clear there already exists a wide parameter space of potential inorganic-organic bismuth compounds.A full survey of these materials is beyond the scope of this review, yet a number of excellent reviews cover this section of bismuth chemistry already, particularly low-dimensionality structures. 71,465,466evertheless, the field of hybrid-organic bismuth compounds would appear to open up an opportunity: a new frontier of nontoxic, optoelectronic materials waiting to be examined and explored by the photovoltaic community.

Outlook and conclusion
Table 1 gives the highest recorded efficiencies and device specifications for all ns 2 solar absorbers discussed in this Review.Lead-based devices currently outperform all alternativesthis is, in part, due to the enormous research effort these materials have experienced but also to the excellent properties of lead, highlighted in our introduction.Despite the similarities in the electronic properties of the group 14 metals and the long-standing research interest in tin solar absorbers, tin-based devices still lag behind their lead counterparts.This can be attributed the sensitivity of Sn 2+ (and similarly, Ge 2+ ) to The highest recorded efficiency for a hybrid halide lead perovskite solar cell is 22.1%, however, only limited technical specifications for this device have been released. 467 such, we report details for the next highest efficiency device here.
This journal is © The Royal Society of Chemistry 2017 Chem.Commun., 2017, 53, 20--44 | 37 oxidation, which hampers device fabrication and leads to efficiency losses.It is therefore essential that alternative schemes to stabilise tin-based devices are developed.Antimony devices have seen persistent improvements to their efficiencies since 2009 and show the second highest performance of any material behind lead.Their earth-abundant composition make them attractive as emerging solar absorbers and with increased attention, efficiencies should continue to rise.Analogously, there has been an increase in reports of bismuth-based devices over the last few years, spurred on by its non-toxic nature and ideal electronic properties, which are expected to be ideal for solar cell applications.As can be seen in Fig. 10, bismuth containing absorbers have recently begun to show efficiencies comparable with the best antimony and tin alternatives.
Several materials highlighted in this review show excellent promise as solar absorbers.The layered lead hybrid perovskites have seen a sharp rise in efficiency in recent monthscomparable with the rise of MAPI-due to the development of a ''quasi-2D'' structure.This approach combines the beneficial increase in stability seen in the purely layered structures, with the excellent optical and electronic properties possessed by the 3D perovskites.In this way, stability and optical properties can be finely tuned to enable moisture tolerant devices with exceptional absorber characteristics.The bismuth containing material AgBiS 2 has also seen rapid advances in efficiencies recently.Its 3D-connected cubic structure draws parallels with the hybrid perovskites and should allow for comparable charge transport properties.
The rise of the hybrid perovskites and emergence of MAPI has galvanised the photovoltaic community with a renewed focus.That an unknown and little studied material can overtake all other third-generation solar absorbers in such a short number of years firmly demonstrates the potential of an earthabundant and cost-effective alternative to silicon technologies.Such a dramatic rise in efficiencies has relied on a multidisciplinary approach and has brought together a decade's worth of advancements in the photonics, engineering and synthetic chemistry communities.Nonetheless, MAPI clearly possesses a fortuitous combination of properties that make it tolerant of a wide range of synthesis conditions and an ideal solar absorber.For this reason, the rise of MAPI should be seen as an outlier and not a precedent to be expected of all emerging absorbers.Instead, it is likely that the next generation of solar materials will require many years of development and optimisation before they can reach comparable efficiencies.Regardless, the emergent field of ns 2 solar absorbers shows particular promise for the future of photovoltaic energy generation.

Fig. 1
Fig. 1 Desired solar absorber properties: (a) ideal range of solar absorber band gaps (shaded) projected onto the AM1.5 solar spectrum (yellow) and the Shockley-Queisser limit (blue); (b) importance of a direct band gap transition-crystalline silicon (c-Si, purple) has dramatically weaker absorption relative to GaAs (green) and MAPI (red); (c) schematic band structure indicating how greater band dispersion (curvature) gives rise to smaller hole and electron effective masses; (d) the impact of bonding structure on defect tolerance, with antibonding states at the top of the valence band maximum giving rise to shallower defects; (e) schematic of absorption and recombination processes in Rashba spin-split systems; (f) band alignment in a heterojunction solar cell.Efficient alignment between materials results in a larger maximum obtainable open circuit voltage (V oc ).The Fermi level of the n-type transparent conducting oxide (TCO) and p-type hole transporting material (HTM) layers is denoted by E F,n and E F,p , respectively.In panels (c) and (e) conduction bands and valence bands are shown in orange and blue, respectively; (b), (d), and (e) adapted with permission from ref. 43-45.

Fig. 3
Fig. 3 (a) Band decomposed charge densities of tropylium tin iodide and tropylium lead iodide, indicating the highest occupied band (HOB) and lowest unoccupied band (LUB) for each material.Reprinted with permission from ref. 128.Copyright 2015, American Chemical Society.(b) (Pb 2 I 6 )Á(H 2 DPNDI)Á(H 2 O)Á(NMP) viewed along the a and b axes, illustrating the 1D lead iodide nanowires and DPNDI network.Reprinted with permission from ref. 135.(c) Quantum-dot-in-perovskite model.Favourable lattice matching between the PbS quantum dot and lead iodide perovskite allows for efficient heteroepitaxial growth in both the X-Y and X-Z planes.Reprinted with permission from ref. 138.Copyright 2015, Nature Publishing Group.

Fig. 4
Fig. 4 (a) J-V curves and trend in band gap demonstrating the effect of pyrazine and concentration of SnF 2 of FASnI 3 perovskite solar cells.Adapted with permission from ref. 164.Copyright 2016, American Chemical Society.(b) Trend in band gap and J-V curves for CsSnI 3Àx Br x perovskite solar cells, where x = 0 to 3. Adapted with permission from ref. 165.Copyright 2015, American Chemical Society.(c) Crystal structures of tin monosulfide.The lowest energy structure is herzenbergite (Pnma), a distorted rocksalt structure (Fm % 3m).Several reports have identified tin monosulfide in the zincblende crystal structure (F % 43m), however, DFT calculations reveal this phase to be high in energy.Adapted with permission from ref. 166.Copyright 2012, American Chemical Society.

Fig. 6 (
Fig. 6 (a and b) The role of grain boundaries (GBs) as recombination centres in CdTe and Sb 2 Se 3 solar cells.(a) The 3D structure of CdTe results in dangling bonds at GBs (illustrated by red rods), which act as defects that can cause unwanted recombination of charge carriers.(b) In contrast, the 1D ribbon structure of Sb 2 Se 3 , if oriented in the [001] direction, permits benign GBs due to saturation of the terminal atoms (red spheres).(a and b) Reprinted with permission from ref. 243.Copyright 2015, Nature Publishing Group.(c) Schematic of the orbital interactions that result in the formation of the stereochemical lone pair in PbO (top panel) and the corresponding molecular orbital diagram (bottom panel).Reprinted with permission from ref. 207.
Despite this substantial effort towards characterisation, however, detailed study of defects and production of a cell remains an open challenge.Alternate copper bismuth sulfides, on the other hand, have seen a little more success.Interest in Cu 3 BiS 3 from the photovoltaic community mostly began in the 2000's: Estrella et al. characterised thin films of Cu 3 BiS 3 in 2003, finding a band gap within 1-1.6 eV, a high optical absorption coefficient (410 5 cm À1 ) and a photocurrent response; 391 Gerein and Haber were able to replicate the band gap (1.3 eV) and absorption coefficient, in addition to finding low resistivity in high quality annealed films. 392Since then, experimental studies have examined its photovoltaic behaviour, finding that surface trap states may This journal is © The Royal Society of Chemistry 2017

Fig. 9
Fig. 9 (a) Crystal structure of Cs 2 BiAgBr 6 .Adapted with permission from ref. 427.Copyright 2015, American Chemical Society.(b) Time-resolved room temperature photoluminescence (PL) of single-crystal and powder Cs 2 BiAgBr 6 , including fits for the PL decay time (t).Reprinted with permission from ref. 428.Copyright 2016, American Chemical Society.