Tin perovskite/fullerene planar layer photovoltaics: improving the e ﬃ ciency and stability of lead-free devices †

We report the ﬁ rst demonstration of orthorhombic CsSnI 3 ﬁ lms prepared from solution at room temperature that have defect densities low enough for use as the light harvesting semiconductor in photovoltaic devices even without using excess Sn in the preparative method, and demonstrate their utility in a model p – i – n photovoltaic device based on a CuI | CsSnI 3 | fullerene planar layer architecture. We also report an e ﬀ ective strategy for simultaneously improving both the e ﬃ ciency and stability of these devices towards air exposure based on the use of excess of SnI 2 during CsSnI 3 synthesis from CsI and SnI 2 . A combination of photoelectron spectroscopy, contact potential measurements and device based studies are used to elucidate the basis for this improvement and role of the excess SnI 2 . The open-circuit voltage in these lead-free photovoltaic devices is shown to be strongly dependent on the degree of alignment between the perovskite conduction band edge and the lowest occupied molecular orbital (LUMO) in the fullerene electron transport layer. Furthermore, the energetics at the perovskite – fullerene interface are shown to be a function both of the LUMO energy of the fullerene and the nature of the interaction at the heterojunction which can give rise to a large abrupt vacuum level shift across the interface. A champion open-circuit voltage of (cid:1) 0.55 V is achieved using indene-C 60 bis-adduct as the electron extraction layer, which is twice that previously reported for a CsSnI 3 based PPV. champion device V oc , FF and h are the highest that have been reported using CsSnI 3 as the light harvesting semiconductor in a PPV by a considerable margin. We have also demonstrated a simple and e ﬀ ective strategy for improving the stability of PPV devices based on CsSnI 3 and have rationalised the reason for this improvement using a combination of photoelectron spectroscopy, contact potential measurements and device based studies. This work shows that if tin-halide perovskites are prepared in a tin rich environment, in order to minimise the density of Sn vacancy defects, then consideration must be made as to the charge transport properties of the source of the excess tin.


Introduction
Over the past 4 years there has been dramatic progress in the eld of perovskite photovoltaics (PPVs), most of which has focused on using methylammonium (MA) lead iodide (MAPbI 3 ) 1-4 and chloride substituted analogues, MAPbI 3Àx Cl x , [5][6][7] as the light harvesting semiconductor. This is due to the high absorption coefficient, high charge carrier diffusion length (L D > 100 nm) 8 and low band gap (E g ) of 1.5-1.6 eV 5,9 in these materials. The highest power conversion efficiency (h) achieved to date in this eld is over 19%. 10 The majority of PPV devices reported to date use the perovskite in a distributed heterojunction device architecture, in which n-type mesoporous TiO 2 serves as the electron-transport layer (ETL) and a p-type organic semiconductor is used as the hole-transport layer (HTL). There are also a smaller number of reports of planar-heterojunction PPVs based on a p-i-n structure, including those in which TiO 2 is replaced with organic semiconductors such as fullerenes, which have achieved h > 17%. [11][12][13] Unlike TiO 2 , fullerene ETLs do not require annealing which removes the need for the substrate electrode to be able to withstand high temperatures. Planar device architectures using organic charge transport layers may also offer the best prospects for accommodating the very large thermal expansion exhibited by some metal halide perovskites. 14 There is also interest in identifying low cost organic and inorganic HTLs, most recently the Cu(I) compounds CuI and CuSCN. [15][16][17][18] The primary drawback of Pb perovskites for photovoltaic applications is their susceptibility to degradation in the presence of water to produce PbI 2 , [19][20][21] which is a partially water soluble compound. 21,22 Soluble Pb compounds are well known to be toxic to both humans and the natural environment 23 and so there is growing interest in the development of alternative, narrow E g perovskites that use Sn instead of Pb, 24 such as CsSnI 3 [25][26][27][28] and MASnI 3 . 29,30 At room temperature and in the absence of air CsSnI 3 adopts a 3D orthorhombic perovskite structure which is known as the black phase. The E g of this semiconductor is near optimal for PV applications ($1.3 eV 14 ), 31 which combined with its high optical absorption coefficient ($10 4 cm À1 in the visible region) 32 and low exciton binding energy ($18 meV), 33 makes CsSnI 3 attractive as a light harvesting semiconductor in PPVs because photoexcitation results in free carrier generation for light with wavelengths as long as $950 nm. 14 However, the material is prone to the formation of a high density of Sn cation vacancy defects, which serve as p-type dopants and recombination centres. 14,34 Sn perovskites also have the drawback of air sensitivity, with the black phase converting to yellow phase CsSnI 3 , which in turn oxidises to Cs 2 SnI 6 in which Sn is in its more stable 4+ oxidation state. 14,35 Unlike Sn, Pb is most stable in the 2+ oxidation state due to the relativistic lowering of the Bohr radius of s-orbitals more than that of other atomic orbitals, which stabilises them towards oxidation, an effect that is most pronounced in heavy p-block elements like Pb. 36,37 In practice the ingress of oxygen and moisture into PV devices is inevitable and so a central challenge in Sn-PPV research is to develop strategies that can be used in conjunction with device encapsulation to reduce the rate of oxidation of Sn perovskite enough for practical applications to be possible. To date there have been six literature reports relating to the use of Sn perovskites in PPVs which collectively demonstrate the potential to achieve high h. [25][26][27][28][29][30] CsSnI 3 was rst used by Chung et al. 25 as a HTL in PPVs in which a ruthenium dye was the primary light harvesting element and the CsSnI 3 was synthesised using a vacuum melt process at 450 C. That pioneering work was quickly followed by Chen et al., 26 who demonstrated a Schottky contact type PPV based on a vacuum evaporated layer of CsSnI 3 formed by sequential deposition of alternating layers of CsI and SnCl 2 followed by annealing at 175 C to form the perovskite, although the device ll factor (FF) was extremely low ($0.22) due to a very low shunt resistance and high series resistance. Concurrently to this Zhou et al. 38 reported that CsSnI 3 could be synthesised at room temperature from a mixed solution of CsI and SnI 2 , although lms formed using that method were not tested for optoelectronic applications. In 2014 Noel et al. reported the highest h to date for a Sn PPV of $6%, achieved using MASnI 3 as the light harvesting semiconductor. 29 However those devices were also shown to exhibit a very large variation in performance and poor stability with deeply coloured devices becoming completely transparent minutes aer measuring. 29 That work was published only days before a report by Hao et al. 30 who showed that h close to 6% could be achieved using MASnIBr 2 and the open-circuit voltage (V oc ) could be tuned by varying the ratio of iodide to bromide as a result of a change in E g . Shortly aerwards Kumar et al. 27 reported that CsSnI 3 processed at 70 C and with 20 mol% SnF 2 additive could be used as the light absorber in a PPV device based on a distributed heterojunction with TiO 2 , achieving photocurrents >22 mA cm À2 . Unfortunately the V oc and ll factor (FF) were limited to $240 mV and $0.37 respectively which constrained h to 2.02%. Notably, devices fabricated without SnF 2 did not function and whilst it was shown that uoride was not incorporated into the CsSnI 3 lattice, the possible optoelectronic functionality of the relatively large quantity of SnF 2 present in the perovskite lm was not discussed. 27 Kumar et al. concluded that the V oc in CsSnI 3 PVs is limited by the high background carrier density due to Sn-vacancy defects even when prepared under Sn-rich conditions, and is not set by the difference between the conduction band minimum of the ETL and the valence band maximum (or highest occupied molecular orbital (HOMO)) of the HTL. The same group went on to show that substitution of iodide by bromide in CsSnI 3 is an effective means of increasing the V oc in Sn-PPV achieving a V oc of 0.41 V. However, the improvement in V oc was at the expense of photocurrent due to the increase in E g . 28 Herein we report: (i) the rst demonstration of orthorhombic CsSnI 3 lms prepared from solution at room temperature with defect densities low enough for use as the light harvesting semiconductor in PPV devices, even in the absence of excess Sn; (ii) a new strategy for improving device stability towards oxidation in air, which does not complicate the process of device fabrication; (iii) the results of a study which unambiguously show that the energetics at the interface between CsSnI 3 and ETL are a key determinant of device V oc .

Results
Model p-i-n PPVs were fabricated with the structure: indium tin oxide (ITO)|CuI|CsSnI 3 |fullerene|bathocuproine (BCP)|Al, as schematically illustrated in Fig. 1a. The CsSnI 3 layer was prepared from 8 wt% CsSnI 3 in N,N-dimethylformamide (DMF) using SnI 2 and CsI. CuI was chosen as the p-type material due to its proven effectiveness as a HTL in PPVs. 15 CuI is also amenable to deposition by both thermal evaporation and by solution processing. Since thermal evaporation enables a very high degree of control over lm thickness and morphology it is the preferred method in this study. Likewise C 60 was selected as the n-type material because it can also be deposited by thermal evaporation. There are also a number of simple derivatives of C 60 commercially available with modied frontier orbital energies. For these reasons C 60 serves as a model n-type charge extraction layer in this work. The CsSnI 3 layer was deposited by spin casting directly onto freshly evaporated CuI at room temperature to give lms with a thickness of $50 nm, as measured using cross-sectional AFM histogram analysis ( Fig. 1b(ii)). In contrast to all other reports relating to the use of Sn-perovskites in PPVs these lms were used without postdeposition annealing. Fig. 1b(i) shows a representative SEM image of a CsSnI 3 lm deposited from 8 wt% CsSnI 3 in DMF onto a CuI coated ITO substrate. The lm is made up of small crystals approximately 50-100 nm in diameter, which is comparable to the lm thickness determined from a step edge analysis using atomic force microscopy ( Fig. 1b(ii) and (ESI) Fig. S1a-c †). It is therefore likely that these crystallites extend from the top to the bottom of the lm as schematically depicted in Fig. 1a. Increasing the loading of CsSnI 3 to 15 wt% resulted in an increase in lm thickness to $80 nm with a larger mean crystallite size and very few pin holes ( Fig. 1c and ESI Fig. S1d-f †). Whilst there are still some gaps in lms of this thickness the coverage is exceptionally high for such a thin layer of metal halide perovskite. 11 Furthermore, lms of comparable quality could also be prepared on glass, ITO and Au coated substrates which proved useful for fundamental measurements of lm properties.
A representative X-ray diffraction (XRD) pattern of a CsSnI 3 lm spin cast onto a glass substrate pre-coated with 10 nm of evaporated CuI is shown in Fig. 2b. The very thin CuI layer is included to ensure the structure of the perovskite lm probed is identical to that used in PPV devices. Since there is no evidence of CuI in the XRD pattern it can be concluded that the buried CuI layer is essentially amorphous and/or too thin to be observed using this technique. The XRD pattern in Fig. 2b is consistent with the orthorhombic perovskite structure expected for the black phase of CsSnI 3 , as reported in the literature 14 and simulated in Fig. 2a. The prominence of the peaks at $29 in the XRD pattern of the lm as compared to the simulated powder pattern is indicative of preferential crystal orientation in the spin cast lm on CuI. The small reections at $26.3 ,$27.7 and $37.8 (starred) are associated with the supporting substrate and sample holder. Aer 16 hours air exposure the CsSnI 3 lm gives the XRD pattern Fig. 2e, which is consistent with the formation of the Cs 2 SnI 6 , in which Sn is in the more stable 4+ oxidation state, as is evident from the good agreement with the simulated reections for Cs 2 SnI 6 shown in Fig. 2d. 39 Xray photoelectron spectroscopy (XPS) measurements of the oxygen 1s binding energy shows that those Sn atoms that do not form part of Cs 2 SnI 6 structure are most likely present as SnO 2 (ESI Fig. S2 †). The absence of reections in the XRD pattern indicative of crystalline SnO 2 indicates that it is present as an amorphous or nano-crystalline phase.
It has recently been proposed that due to the ease with which Sn-cation vacancies form in CsSnI 3 it must be prepared in a Snrich environment or at high temperature to ensure the density of these defects is low enough for its use as a light harvesting material in PPV devices. 27,34,35 However we nd that devices based on CsSnI 3 lms prepared using stoichiometric mixtures of CsI and SnI 2 with no excess of Sn and processed entirely at room temperature generate a short circuit current density (J sc ) of $10 mA cm À2 under 1 sun simulated solar illumination, which is a very high current for such a thin semiconductor layer ($50 nm) (Fig. 3). The incident-photon-to-converted-electron (IPCE) efficiency (ESI Fig. 3 †) conrms that CsSnI 3 is the primary light absorbing component in these devices, since wavelengths as long as $930 nm are converted into electrons in the external circuit, which corresponds to sub-band gap photon Fig. 1 (a) Schematic of the device architecture used in this work; (b(i)) SEM image of a CsSnI 3 film prepared with a 10 mol% excess SnI 2 and spin cast at 4000 rpm from 8 wt% solution onto an ITO glass substrate coated with a 100 nm layer of CuI; (b(ii)) AFM step height of a scored CsSnI 3 film spin cast from 8 wt% solution onto glass; (c) SEM image of a CsSnI 3 film spun from 15 wt% CsSnI 3 solution with 10 mol% SnI 2 excess using the same deposition process and substrate as for (b(i)). energies for C 60 . This spectral response is also consistent with E g for CsSnI 3 reported in the literature of $1.3 eV. 14,35 As shown in Fig. 3, devices based on CsSnI 3 lms prepared using an excess of SnI 2 and C 60 ETL exhibit up to $30% increase in J sc and V oc as compared to those without. The optimal SnI 2 excess is $10 mol% which results in a doubling of h from $0.75% to $1.5%. All reections in the XRD pattern for lms prepared using 10 mol% excess SnI 2 (Fig. 2c) can be attributed to CsSnI 3 , except the weak reection at $28.5 . The latter cannot be assigned to SnI 2 or its oxidised form; SnI 4 , which indicates that the excess SnI 2 is nano-crystalline or amorphous. High-resolution XPS measurements of the surface chemical composition of freshly prepared lms with a 10 mol% SnI 2 excess show that there is very little SnI 2 and no evidence for SnI 4 at the CsSnI 3 surface (Fig. S4 †), since $95% of the photoelectrons originate from the top $7 nm of the sample. The XPS data therefore provides evidence that the 10 mol% excess SnI 2 phase separates to the buried interface upon solvent evaporation. Notably, since these lms were unavoidably exposed to the laboratory for $1 minute when transferring from the nitrogen lled glove box to the spectrometer vacuum system there is also evidence for signicant oxidation of the surface of the CsSnI 3 to Cs 2 SnI 6 in the XPS spectra.
If the excess SnI 2 is primarily located at the buried CsSnI 3 |CuI interface it must be capable of transporting holes from CsSnI 3 into CuI since PPV devices incorporating up to 15 mol% SnI 2 perform comparably to those with 5% SnI 2 excess (Fig. 3). To our knowledge there are no previous reports of SnI 2 being used as a semiconductor in an organic or hybrid electronic device and so its charge transport properties are not known. To investigate the transport properties of SnI 2 we have estimated the hole-mobility in SnI 2 from the space charge limited current in unipolar diodes to be 2.2-5.1 Â 10 À6 cm 2 V À1 s À1 (ESI, Fig. S5 †). Whilst this value is low as compared to that of Sn and Pb halide perovskites it is comparable to that of many amorphous organic semiconductors 40 and offers an explanation as to why device FF begins to deteriorate with increasing SnI 2 excess beyond 10 mol%. To determine the accessibility of the SnI 2 valence band for hole transport we have measured the ionisation potential (I p ) of SnI 2 (I p $ 5.64 eV AE 0.05 eV) and CuI lms (I p $ 5.84 eV AE 0.05 eV) using ultraviolet photoelectron spectroscopy (UPS). Full details of the UPS study are given in the (ESI Fig. S6 and S7 †). These energy levels, along with the E g for CuI determined from electronic absorption measurements (ESI, Fig. S8 †) and E g for SnI 2 taken from the literature, 41 are given in Fig. 4a. The susceptibility of CsSnI 3 lms to oxidation in air prevented us from making a reliable measurement of the CsSnI 3 I p using UPS, and the values reported in the literature for argon sputtered CsSnI 3also measured using UPSvary from 4.9 eV 25 to 5.7 eV. 41 In order to estimate the position of the valence and conduction band edges in CsSnI 3 with condence we have measured the work function of CsSnI 3 lms without air exposure, or argon sputtering, using a Kelvin probe located in the same glove box as the spin coater used to deposit the perovskite lm. On the basis of this measurement, combined with the known E g ($ 1.3 eV) and knowledge of the tendency of this material to form p-type defects, 27,34,35 we have estimated the possible range of energies for the valence and conduction band edges in CsSnI 3 . These energy levels along with the work function of each material measured using the Kelvin probe technique are given in Fig. 4a. It is clear from the location of the Fermi level in CuI that it is heavy p-doped which is consistent with literature reports 43 and can be attributed to Cu atom array vacancies. In the absence of such intrinsic defects CuI would block the extraction of holes from the CsSnI 3 valance band due to its very deep lying valance band. However, the high density of electron acceptor states near to the valence band edge should ensure that there is no barrier to hole- Fig. 3 Scatter graphs of PPV performance characteristics for devices with the structure: ITO|CuI|CsSnI 3 |C 60 |BCP|Al using CsSnI 3 prepared from solutions containing either 0, 5 or 10 mol% excess SnI 2 . The mean values are shown as red squares. extraction from CsSnI 3 when thermodynamic equilibrium is established across the interface. Similarly, spontaneous electron transfer from the SnI 2 into CuI due to the difference in Fermi level would be expected to remove the relatively small barrier to holeextraction from SnI 2 into CuI. On the basis of the energy levels given in Fig. 4a there would however be expected to be a signicant barrier to hole-transport from the CsSnI 3 valence band into the SnI 2 valence band. However, since the current-voltage characteristics of PPV devices fabricated with an excess of SnI 2 in the CsSnI 3 layer (up to 15 mol%) are not consistent with a high device series resistance it is likely that there is a favourable vacuum level shi at the interface due to the formation of an interfacial dipole.
The improvement in J sc and V oc in PPV devices using a CsSnI 3 layer prepared with excess SnI 2 can be rationalised in terms of a reduction in the background carrier density which gives rise to recombination loses, since it is known that the density of Snvacancy defects (i.e. the primary source of the background carrier density) is suppressed when the perovskite is synthesised in a Sn-rich environment. 27,34,35 However, to investigate the possibility that the energetics at the perovskite|C 60 interface might also plays a role in controlling V oc we have measured the change in energy between the perovskite Fermi level and the surface vacuum level (vac) as a function of the C 60 layer thickness using the Kelvin probe method. This energy difference, 3 vac f , corresponds to the work function of C 60 when the lm thickness is sufficient for the establishment of Fermi level alignment across the interface. If thermodynamic equilibrium is not achieved then 3 vac f corresponds to the work function of the perovskite with a modied surface potential. 42,43 Changes in 3 vac f can therefore be interpreted in terms of a vacuum level shi which give rises to a comparable shi in the energy of the HOMO and LUMO in the C 60 with respect the valance and conduction band edges in CsSnI 3 . The results of this experiment are summarised in Fig. 4b. At the interface between CsSnI 3 prepared with 0% excess SnI 2 there is no signicant change in the energy of the vacuum level. Conversely for lms prepared with 10 mol% excess SnI 2 there is a large abrupt positive vacuum level shi of $0.3 eV meV. Abrupt positive vacuum level shis at the interface between inorganic materials and organic semiconductors are relatively rare 42,43 and are indicative of ground state charge transfer from the substrate into the rst layer of molecule forming an interfacial dipole. The reason for the stronger interaction between C 60 and CsSnI 3 with fewer Sn-vacancy sites is not yet clear, although the existence of this abrupt positive vacuum level shi offers an alternative explanation for the increase in V oc , since it would operate to reduce the loss in electron potential energy when moving from the perovskite conduction band into the fullerene LUMO.
To further investigate the extent to which V oc in these devices is controlled by the degree of alignment between the LUMO of the ETL and the conduction band edge in CsSnI 3 , we have systematically reduced the LUMO energy of the ETL from $4.00 eV 45,46 to $3.91 eV 46 and nally to $3.74 eV 46 below the vacuum level by substituting C 60 with the C 60 derivatives [6,6]phenyl-C 61 -butyric acid methyl ester (PC 60 BM) and indene-C 60 bis-adduct (IC 60 BA) (Fig. 4a). Whilst the exact magnitude of these values remain the subject of debate 45 the differences in LUMO energy between these derivatives is rmly established. 45,46 Champion J/V characteristics in the dark and under 1 sun simulated illumination are given in Fig. 5 and the full data set is given in the ESI (Table S1 †). It is clear from Fig. 5 and the full data set in (Table S1 †) that the increase in device V oc correlates with an increase in the LUMO energy of the fullerene ETL which is consistent with a reduction in the potential energy lost by an electron moving from the CsSnI 3 into the ETL as seen in organic PVs. 46 The highest h is achieved for devices using ICBA as the ETL at 2.76%, which is signicantly higher than the best previously reported for a CsSnI 3 based PPV (2.02%). 27 Collectively these data show that the performance of PPVs based on CsSnI 3 depends strongly on the energetics at the interface between the CsSnI 3 and the ETL, much like that seen at donor/ acceptor heterojunctions in OPVs. 46 For practical applications the stability of the light harvesting semiconductor towards oxidation in air is critically important.
Encouragingly PV devices based on CsSnI 3 prepared in a Sn-rich environment have been reported to be stable for more than 250 hours when stored in glove box, 27 althoughto our knowledgethe stability of Sn perovskite PVs in air has not be quantitatively reported. We nd that the use of excess SnI 2 in the preparation of CsSnI 3 dramatically improves device stability. Fig. 6 shows the key performance parameters of unencapsulated PPV devices (using C 60 as the ETL) as a function of SnI 2 excess in the CsSnI 3 layer. The data points in the hashed region correspond to device performance immediately aer fabrication and aer storage for 10 days in a nitrogen lled glove box (<5 ppm O 2 and <0.1 ppm H 2 O). For devices based on CsSnI 3 lms prepared using stoichiometric amounts of CsI and SnI 2 the h of is reduced by $70% aer this ten day period of storage due to the combined effect of deteriorated J sc , V oc and FF. Conversely devices based on CsSnI 3 with 10 mol% excess SnI 2 exhibit a relatively small reduction in h of $10% due primarily to a reduction in FF. These devices were then bought into the laboratory and the performance tested as a function of time. Devices with 0 mol% excess SnI 2 stopped working altogether aer $30 minutes air exposure due to a complete collapse in J sc . Conversely devices with 10 mol% excess SnI 2 exhibit dramatically improved stability such that even aer 14 hours in air without encapsulation they still function, albeit with a J sc of one h the starting value. Both the FF and V oc are essentially stable during this period.
It is likely that the deterioration in device performance upon exposure to the laboratory environment results from a number of parallel degradation mechanisms some of which are not related to the perovskite layer, such as oxidation of the Al cathode and/or doping of the C 60 layer by oxygen. 47 However, since it is evident from our XPS studies (ESI, Fig. S4 †) that oxidation of CsSnI 3 in air to form Cs 2 SnI 6 is a very rapid process and that Cs 2 SnI 6 in known to absorb light very weakly as compared to CsSnI 3 , 41 it is likely that conversion of CsSnI 3 into Cs 2 SnI 6 is major contributor to the Fig. 5 Dark and light JV curves of best performing PPV devices with the structure ITO|CuI|CsSnI 3 + 10 mol% SnI 2 |fullerene|BCP|Al, using different fullerene ETLs. For the case of ICBA the device V oc was found to be strongly dependent on the concentration of the ICBA solution used: 3 mg ml À1 ICBA (green), 5 mg ml À1 ICBA (blue), although was always higher than that achieved in devices using PCBM when all other parts of the device were identical. Fig. 6 Key device performance characteristics for unencapsulated PPV devices with the structure ITO|CuI|CsSnI 3 |C 60 |BCP|Al using CsSnI 3 prepared from solutions containing either 0, 5 or 10 mol% excess SnI 2 . The data in the hashed region correspond to device performance immediately after fabrication and then again after storage in a nitrogen atmosphere (<5 ppm O 2 and <0.1 ppm H 2 O) for 10 days. These devices were then bought into laboratory and the performance was measured as a function of time stored in air.
continuous deterioration in J sc . This conclusion is supported by the evolution of the absorption spectra of CsSnI 3 lms on glass, with and without 10 mol% SnI 2 , as a function of time exposed to air ((ESI, Fig. 9 †)), from which it is evident that the addition of 10 mol% SnI 2 greatly reduces the rate of oxidation of the CsSnI 3 . Since this process would be expected to begin at the interface with the ETL any signicant change in the energetics at this interface for electron extraction as a result of this transformation would manifest as a change in device series resistance and FF, the effect of which would be most pronounced in the early stages of oxidation process when the potential energy step forms. Cs 2 SnI 6 has been shown to have a relatively high electron and hole mobility 48 although its conduction band edge is $200 meV deeper lying than that of CsSnI 3 41 and so the formation of a Cs 2 SnI 6 layer between the remaining CsSnI 3 and C 60 would most likely impede the ow of electrons from the conduction band of the CsSnI 3 into the LUMO of the C 60 resulting in an increase in device series resistance and reduction in FF, as is observed to be the case. To test the generality of this result for another fullerene ETL and to remove the complexity associated with possible doping of the fullerene layer by oxygen upon exposure to air we have fabricated PPV devices using PCBM as the ETL with and without exposure of the CsSnI 3 layer to air for 1 minute, before transferring back into the glove box to complete device fabrication. Typical J-V characteristics for PPV devices with and without exposure of the perovskite layer to air are given in Fig. 7 from Fig. 7 Representative J-V characteristics of PPV devices with the structure: ITO|CuI 70 nm|CsSnI 3 (with and without 10 mol% excess SnI 2 ) |PCBM|BCP|Al: (a) CsSnI 3 with no excess SnI 2 ; (b) CsSnI 3 with 10 mol% excess SnI 2 . Graphs show devices made entirely in glove box (black) and those made using a CsSnI 3 layer exposed to the laboratory environment for 1 minute (red) before being returned to the glove box to complete fabrication and testing. which it is clear that devices incorporating an excess of SnI 2 are much more stable than those without and that the changes in J sc , V oc and FF aer 1 minute air exposure are consistent with that in Fig. 6.
Finally, we have tested the stability of devices, with and without 10 mol% excess SnI 2 in the CsSnI 3 layer, over a period of four hours continuous illumination in a nitrogen atmosphere. Representative results are given in Fig. 8, from which it is clear that devices incorporating excess SnI 2 exhibit greatly improved stability including a stable V oc and FF aer the rst 20 minutes. Whilst J sc deteriorates rapidly by $10% within the rst 50 minutes the rate of reduction in J sc then slows markedly. Conversely devices with no excess of CsSnI 3 exhibit a continuous rapid decline in FF and J sc .
Collectively the aforementioned stability studies show that the use of a 10 mol% excess of SnI 2 in CsSnI 3 preparation imparts dramatically improved device stability, which indicates that the stability of CsSnI 3 towards oxidation is very sensitive to the density of Sn vacancy defects. Since the transformation of CsSnI 3 into Cs 2 CsI 6 requires the loss of Sn it seems plausible that at the site of a Sn vacancy in CsSnI 3 the barrier to transformation into Cs 2 SnI 6 is reduced. It is also possible that the excess SnI 2 occupies some of the space between adjacent CsSnI 3 crystallites thereby impeding the ingress of the oxygen and water that gives rise to oxidation.

Conclusion
In summary we have demonstrated the preparation of orthorhombic CsSnI 3 lms from solution at room temperature with defect densities low enough for use as the light harvesting semiconductor in photovoltaic devices, even without the addition of excess Sn. We have shown how the V oc in a model planar layer device based on a CuI|CsSnI 3 |fullerene p-i-n structure can be doubled from $0.28 V to $0.55 V by engineering the energetics at the CsSnI 3 /fullerene interface. To our knowledge, the champion device V oc , FF and h are the highest that have been reported using CsSnI 3 as the light harvesting semiconductor in a PPV by a considerable margin. We have also demonstrated a simple and effective strategy for improving the stability of PPV devices based on CsSnI 3 and have rationalised the reason for this improvement using a combination of photoelectron spectroscopy, contact potential measurements and device based studies. This work shows that if tin-halide perovskites are prepared in a tin rich environment, in order to minimise the density of Sn vacancy defects, then consideration must be made as to the charge transport properties of the source of the excess tin.

Synthesis of caesium tin(II) iodide
CsI was added to SnI 2 in a 1 : 1.1 molar ratio under nitrogen. To this DMF was added to make an 8 wt% solution which is bright yellow in colour. The solution was shaken for a few minutes until all CsI and SnI 2 had dissolved and was le overnight before use. The perovskite material precipitated on spin casting to give dark red/brown coloured lms.

Device fabrication
Solar cells were fabricated on indium tin oxide (ITO) coated glass slides (Thin Film Devices Inc., 15 AE 3 U sq À1 ). The slides were prepared by cleaning the ITO using acetone, water and isopropanol in an ultrasound bath for 10 minutes each followed by UV/O 3 treatment for 15 minutes. The slides were then loaded into a dry nitrogen lled glove box for device fabrication. 70-100 nm thick CuI lms were deposited onto the ITO glass substrates by thermal evaporation. CuI can also be deposited by spin casting a 50 mg ml À1 in chlorobenzene : di-n-propyl sulde (5.5 : 1) at 1000 rpm. Onto this 8 wt% CsSnI 3 with a 10 mol% excess of SnI 2 was spin cast at 4000 rpm from a DMF solution for 60 s. 40 nm of C 60 was the deposited by thermal evaporation at a rate of 0.1-0.4 nm s À1 . PC 61 BM was spin cast at 1000 rpm from 15 mg ml À1 chlorobenzene solution, and IC 60 BA from 3, 5 or 7 mg ml À1 chlorobenzene solutions for 60 s. The PCBM and ICBA solutions were ltered with a 0.2 mm PTFE syringe lter before use. Finally 8 nm of bathocuproine (BCP) followed by 50 nm of Al were deposited by thermal evaporation at a rate of 0.5-1Å s À1 . The Al was evaporated through a mask to produce 0.06 cm 2 pixels. For devices without CuI, ITO was cleaned in the same way, but was put under the UV/O 3 lamp for 5 minutes longer to help the perovskite lm wet the surface.

Device testing
Device testing was performed in the same glove box as used for device fabrication unless otherwise stated. Current-voltage (J-V) curves were measured using a Keithley 2400 source-meter under AM1.5G solar illumination at 100 mW cm À2 (1 sun) by sweeping from À1 V to +1 V in 0.01 V increments at a rate of 0.1 V s À1 . There was no signicant current-voltage hysteresis, as shown in Fig. S10. † IPCE measurements were carried out using a Sciencetech SF150 xenon arc lamp and a PTI monochromator, with the monochromatic light intensity calibrated using a Si photodiode (Newport 818-UV). The incoming monochromatic light was chopped at 500 Hz. For signal measurement a Stanford Research Systems SR 830 lock-in amplier was used.

X-ray diffraction
CsSnI 3 lms were spin cast from 16 wt% DMF solutions at 2000 rpm onto glass slides coated with 10 nm of CuI. The X-ray diffraction patterns were collected under a ow of nitrogen using a Cu Ka 1/2 source in q-q mode on a Bruker D8 Advance powder diffractometer equipped with an Anton-Paar HTK900 gas chamber. Patterns were corrected for the height of the sample by calibrating the 2q axis against reference patterns. Simulated diffraction patterns were calculated using the program Mercury 3.1 with CIFs from the Inorganic Crystal Structure Database (ICSD).

UPS and XPS
UPS and XPS were performed on freshly prepared CsSnI 3 lms spin cast onto Au substrates. The samples were exposed to the laboratory environment for $1 min prior to loading into the ultra-high vacuum system. UPS spectra were recorded using UV photons with energy 21.22 eV from a He I plasma source. The Xray photoelectron spectroscopy (XPS) measurements were carried out in a UHV system with a base pressure of 5 Â 10 À11 mbar. The sample was excited with X-rays from a mono-chromated Al Ka source (hn ¼ 1486.6 eV) with the photoelectrons being detected at a 45 take-off angle using an Omicron Sphera electron analyser.