Ambient-air-stable inorganic Cs2SnI6 double perovskite thin films via aerosol-assisted chemical vapour deposition

Air-stable caesium tin iodide double perovskite (Cs2SnI6) thin films have been fabricated via aerosol-assisted chemical vapour deposition (AACVD). We compare the properties of the double perovskite filmsmade using AACVD with those made by the widely used spin-coating method. Films with purer crystalline phase (less CsI impurity) and far better stability in ambient air can be obtained by AACVD compared with spin coating. The AACVD-grown Cs2SnI6 films retain high phase purity for at least 100 days aging in air with negligible CsI impurities detected over this time, as determined by X-ray diffraction. The films exhibit an optical band gap energy (Eg) of ca. 1.3 eV and a homogeneous morphology with the expected nominal stoichiometry within error, as probed by energy-dispersive X-ray spectroscopy. Overall, the characteristics of the Cs2SnI6 films are highly process-dependent, e.g. they are influenced by the presence of hydroiodic acid (HI) in the precursor solution. Without HI addition, an iodine-deficient film with more CsI is produced, which also exhibits a larger Eg of ca. 1.6 eV. In addition to bulk properties, we utilise X-ray photoelectron spectroscopy (XPS) to scrutinise the surface characteristics in detail. We find excess Sn and I located at the surfaces. This can be attributed to the presence of SnI4 from the deposition precursor vapour. Furthermore, following aging in air, an increase in CsI impurity for the AACVD (+HI)-grown film is observed, along with a reduction in SnI4 at the surfaces. Near-ambient pressure XPS (NAP-XPS) is used to examine the surface stability of AACVD (+HI)-grown films on exposure to O2 and H2O. No enhancement in the amount of CsI impurity is observed after both H2O vapour (9 mbar) and O2 (5 mbar) exposure. Nevertheless, the concentrations of tin and iodine change after exposure, suggesting that SnI4 protects Cs2SnI6 from degradation. This passivation effect of SnI4 on Cs2SnI6 surfaces is proposed to explain the additional stability of Cs2SnI6 fabricated via AACVD.


Introduction
In recent years, perovskite solar cells (PSCs) have attracted tremendous attention, 1-8 since the rst description by Kojima et al. in 2009. 9 Within a decade, this type of photovoltaic technology has undergone unprecedented advances in certicated power conversion efficiency (PCE), which has reached more than 20%. [4][5][6][7] As a result, the PSCs are considered to be highly promising candidates for reduction in the cost per watt of commercial solar-energy-conversion devices. Nevertheless, there are two dominant concerns still constraining the development of PSC commercialisation: toxicity and stability. To date, all of the most efficient PSCs have been composed of lead-containing light absorbers, potentially resulting in serious human-body and environmental damage. Consequently, developing lead-free perovskite materials for use in PSCs is preferable to reduce the toxicity of photovoltaic devices. In addition, the instability/degradation of typical perovskite materials, particularly when placed in a humid environment is extremely problematic; 10 upon exposure to moisture, the most widely used organolead perovskite, methylammonium lead (tri-)iodide (MAPI), degrades into lead iodide with a complete loss of the nitrogen moiety. 11 Many attempts to improve the stability of halide perovskites have been investigated by a variety of approaches, including the use of alternative lead precursors, 12 mixed cations, 13,14 and fabrication of two-dimensional perovskites. 15 Nonetheless, the stability still lags behind industrial photovoltaic standards which typically require a <10% loss in power generation aer utilisation of 20 years. 16 Therefore, the synthesis of stable halide perovskite lms for use in photovoltaics is essential for the future development of PSCs.
Among lead-free halide perovskite materials, tin-based perovskites are widely investigated as tin is a group IV metal and can be isoelectronic with lead. The rst pure organo-Sn PSC was reported in 2014 with an initial PCE of approximately 6%. 17 However, compared to organo-Pb PSCs, organo-Sn PSCs show not only lower PCEs but even poorer atmospheric stability (typically showing degradation within an hour). 17 To stabilise Sn perovskites, it is possible to replace the unstable organic cation (e.g. methylammonium, MA) with inorganic metal ions such as caesium (Cs). However, the PCEs of CsSnI 3 PSCs are typically below 4%, [18][19][20] with a current champion PCE of 4.8% reported recently. 21 The stability with respect to oxidation is poor owing to the fact that Sn 2+ in CsSnI 3 (or MASnI 3 etc.) is readily oxidised to Sn 4+ . 20 Therefore, the Sn-decient derivative perovskite, Cs 2 SnI 6 , has been investigated as one of the most promising Pbfree light absorbers in very recent years. This is due to the tetravalent Sn in the Cs 2 SnI 6 structure, which cannot be further oxidised. As a result, the Cs 2 SnI 6 double perovskite can offer far better stability under atmospheric conditions containing oxygen and water vapour. [22][23][24][25][26][27][28][29][30] The Cs 2 SnI 6 double perovskite which is ambipolar (i.e. it can be doped as n-type or p-type) exhibits excellent carrier mobility when doped as an n-type semiconducting material. In addition, it has a relatively low optimised energy band gap (E g ) of ca. 1.3 eV and high absorption coefficient (of over 10 5 cm À1 for energies above 1.7 eV), which shows great potential for use as a light absorber in photovoltaic devices. [22][23][24] Very recently, using a special architecture, Cs 2 SnI 6 -based solar cells with PCEs of ca. 2.0% and acceptable stability in air have been reported. 23 By doping Br into the structure, a higher PCE of ca. 5.2% has been reported for Cs 2 SnBr x I 6Àx (x ¼ $2). 31 Nevertheless, many discrepancies in the properties of Cs 2 SnI 6 have arisen, such as variations in E g between 1.3 eV and 1.6 eV and carrier mobility (varying over the range 1-310 cm 2 V À1 s À1 ), which are both found to be highly process-dependent. [22][23][24][25] Lee et al. demonstrated that different CsI-impurity levels lead to variations in the band gap energies, carrier concentrations and mobilities, which can be controlled by preparation methods. 23 Given the improvement in Pb-based perovskites, one may envisage the potential to obtain high quality Cs 2 SnI 6 thin lms with favourable characteristics for use in photovoltaics using novel preparation approaches.
Aerosol-assisted chemical vapour deposition (AACVD) is a promising process for fabrication of perovskite thin lms (one-step 32-34 and two-step processes 35,36 ), since it merges the advantages from solution-based methods (low cost) and chemical vapour deposition (high quality, controlled thickness). [37][38][39][40] Nonetheless, until now, AACVD studies of halide perovskites are still limited since Lewis et al. rst introduced AACVD-grown organo-Pb perovskite thin lms. [32][33][34][35] A detailed study of AACVD-grown Sn-based perovskites, which includes characterisation of their bulk and surface properties and surface stability/degradation has not been reported thus far. In this work, we utilise a one-step AACVD process to fabricate high quality Cs 2 SnI 6 thin lms for the rst time. A thorough study of the fabrication and characterisation of Cs 2 SnI 6 material is also provided. By applying this facile, scalable process, lms with high purity and uniformity can be obtained. The growth of lms at a relatively low temperature of 150 C is well adapted to exible device applications, 41-45 when compared to vacuum deposition (190 C). 25 We also demonstrate the importance of addition of hydroiodic (HI) acid to the precursor solution to afford Cs 2 SnI 6 lms with a reduced number of macroscopic defects such as pinholes. Iodine vacancies (V I ) are easily formed as well as tin interstitials (Sn i ) in this material. 29 HI is selected for two reasons: rstly, it is an ideal iodine source to afford Irich Cs 2 SnI 6 material. Secondly, using excess SnI 4 to supply iodine could potentially lead to deleterious formation of Sn i . We also employ near-ambient pressure X-ray photoelectron spectroscopy (NAP-XPS) to measure the in situ degradation of the AACVD-grown lms prepared with the addition of HI. By combining XPS data from a number of experiments, we show that surface passivation through a layer containing excess Sn and I is one factor imparting stability of the Cs 2 SnI 6 lms. These insights will help the future development to Pb-free halide perovskites for use in photovoltaic devices.

Sample preparation
Firstly, 0.5196 g of caesium iodide (CsI, 99.9%, Aldrich) was added into 10 mL of warm ($70 C) anhydrous N,N-dimethylformamide (DMF, 99.8%, Sigma-Aldrich) solvent with gentle stirring to generate a clear pale-yellow solution (0.2 M). A stoichiometric amount (0.6263 g, 0.1 M) of tin(IV) iodide (SnI 4 , 99.999%, Aldrich) was poured into the solution. The colour of the precursor solution immediately became dark brown and then black aer a few minutes. A small amount (300 mL, 0.2 M) of hydroiodic acid (HI, 57 wt%, Aldrich) was added to produce an I-rich solution (denoted as +HI). Another precursor solution was prepared without the addition of HI (denoted as ÀHI) for comparison. The resulting solutions were used for the AACVD process or spin-coating.
The prepared solution was directly used as the feed for AACVD without further procedures. The apparatus used for AACVD has been previously described in detail by Ramasamy et al. 46 Firstly, 10 mL of the DMF solution was poured into a twonecked 100 mL round-bottom ask with a gas inlet. This allows Ar carrier gas (ow rate $300 sccm) to pass into the solution to support the transport of aerosol generated by a humidier. This ask was connected to a tube in a furnace where the temperature during deposition was set to 130 C. Indium tin oxide (ITO)coated glass (Ossila) with a size of 2.0 Â 1.5 cm or gold-coated silicon (Au-Si) were utilised as substrates. The deposition rate was approximately 20 nm per minute as measured by a Veeco Dektak 8 Surface Prolometer and the deposition time was set for an hour. As a result, we obtained lms with a thickness of around 1.2 mm on the substrates. Following the deposition, the samples were annealed at 150 C in ambient air to remove residual solvent from the samples and tube for 30 minutes. The sample was then allowed to cool to below 100 C before removal from the AACVD tube. The prepared samples were placed into (sample) tubes lled with Ar and then stored in a small desiccator. For the aging study, the lms were aged in ambient air at an average relative humidity (RH) of >70% for periods up to $100 days. All fresh samples were exposed to ambient air for less than 10 minutes when transferring from the growth tubes to a characterisation chamber.
A small volume (100 mL) of the same precursor solutions (+HI) was also used to prepare spin-coated lms of similar ($1.2 mm) thickness for comparison. The lms were deposited at 4000 rpm for 30 seconds and subsequently transferred to a hot plate for annealing at 150 C in air for 30 minutes. The resultant samples were stored or aged in the same way as the AACVDprocessed samples.

Instrumentation
X-ray diffraction (XRD) patterns were recorded using a Bruker D8 Advance. A grazing incidence (GI) angle of 3 was used to detect the signals from the thin lms without the substrates. The scanning range (2q) was recorded from 5 to 80 with a step size of 0.05 and dwell time of 4.5 s.
The morphology of the lms was measured using scanning electron microscopy (SEM) and bulk elemental analysis was carried out using an energy-dispersive X-ray spectroscopy (EDX, Philips XL30 equipped with a DX4 EDX spectrometer). EDX was conducted in the SEM chamber and the error in bulk stoichiometry measurements was AE2%.
Optical diffuse reectance measurements were performed using an ultra-violet-visible-near-infra-red (UV-VIS-NIR) spectrometer (PerkinElmer Lambda-1050) with an integrated sphere (IS) module to collect the scattered light. Spectra were recorded over a wavelength range of 300 to 1500 nm from samples deposited on soda-lime glass substrates. Reectance (R) was automatically converted to absorption (a) data in accordance with the Kubelka-Munk equation. The optical energy band gap (E g ) was acquired using a Tauc plot to nd the onset point of the absorption spectra.
X-ray photoelectron spectroscopy (XPS) measurements were performed with either a Kratos Axis Ultra or SPECS XPS instrument. Both facilities are equipped with monochromated Al Ka X-ray sources with a photon energy of 1486.6 eV. Emitted photoelectrons were collected using either a 165 mm hemispherical energy analyser (Kratos) or a 150 mm hemispherical energy analyser (Phoibos 150 SPECS), respectively. All measurements conducted in this section were carried out in ultra-high vacuum (UHV) conditions. Binding energies (BEs) were calibrated to C 1s from adventitious carbon at 284.8 eV for ITO-glass substrates or Au 4f 7/2 at 84.0 eV for Au-Si substrates.
The SPECS instrument also allows for NAP-XPS measurements to be carried out and the details can be found elsewhere. 11 The NAP cell is equipped with differential pumping to enable gas pressures up to $15 mbar to be applied during measurement. In this work, 9 mbar water vapour (RH ¼ $30%) and 5 mbar oxygen were separately applied. All XPS data were analysed using CasaXPS soware, in which a Shirley background and pseudo-Voigt peaks (30% Lorentzian and 70% Gaussian) were tted to the photoelectron peaks acquired. 47 BE values are quoted to an accuracy of AE0.1 and AE0.05 eV for typical and in situ XPS study, respectively, due to instrumental precision. To obtain quantied surface compositional information, the built-in CasaXPS sensitivity factors (Kratos, relative to F 1s as 1.0) are utilised for calculating the available stoichiometry. Note that this does not allow the quantitative analysis of spectra acquired at near-ambient pressure (NAP) conditions due to the absence of adapted transmission-function data. All surface elemental concentrations are normalised to Cs (Cs 3d 5/2 spectra) as 2 in an ideal stoichiometric Cs 2 SnI 6 lm. Excess concentrations of Sn and I above 1 and 6, respectively are likely to indicate the presence of non-Cs 2 SnI 6 species.

Results and discussion
Following annealing under ambient air in the AACVD reactor, highly uniform and black AACVD (+HI)-processed Cs 2 SnI 6 lms were formed, as shown in Fig. 1A. This indicates strong visiblelight absorption by the as-prepared AACVD (+HI)-grown Cs 2 SnI 6 lm as the part of the logo covered by the lm is barely visible. Following $100 days storage under ambient air (average RH > 70%), the appearance of the lm (Fig. 1B) is similar to the fresh sample ( Fig. 1A). This suggests that there is no signicant change in the visible light absorption of the lm aer being aged in a humid environment. Peeling and scratching tests suggest that the AACVD lms produced here show signicantly better adhesion to the substrates than the spin-coated lms prepared under similar conditions (see ESI, Fig. S1 †).
In order to determine the crystalline phase, grazing incidence X-ray diffraction (GIXRD) patterns of various lms (Fig. 2) were examined. Fig. 2A shows XRD reections of a typical AACVD (+HI)-grown lm as a function of aging time (up to $100 days). For the pristine sample, the XRD peaks indicate that the crystal structure is in excellent agreement with the Cs 2 SnI 6 double perovskite structure (JCPDS no. 51-0466, cubic, space group Fm3m, a ¼ 11.65Å). 26 The dominant (strongest) diffraction peak from the (222) crystal plane at 26.55 and other primary peaks such as (400) are also annotated in the gure. The face-centred-cubic double perovskite structure (with Sn deciency at the centre) increases the intensity of the (222) reection relative to the (111) reection. 22 A more detailed examination reveals a diffraction peak due to the (110) reection from a trace amount of the cubic CsI impurity at 27.60 (JCPDS no. 89-4257). 48 The amount of this contaminant is very low in fresh samples ( Fig. 2A), suggesting that the AACVD process allows formation of highly pure Cs 2 SnI 6 crystalline lms. In addition, we observe that the AACVD (+HI)-processed lm appears to be highly stable following $10 and $100 days exposure to air (Fig. 2A). The amount of CsI impurity remains low and there are no other new crystalline phases generated. The ratios of the intensity of the CsI (110) reection to that of the Cs 2 SnI 6 (220) reection are summarised in Table 1. This highlights the relatively small change in the amount of CsI impurity when aged in ambient air. As shown, the ratio is in the vicinity of $0.020 aer 100 days' aging, showing that AACVD (+HI)-processed Cs 2 SnI 6 is signicantly more stable than the widely used methylammonium lead iodide perovskite. MAPI is easily transformed to lead iodide (PbI 2 ) under ambient air within a week due to reaction with moisture in the environment. 49 Fig . 2B shows the effect on crystal structure produced by the addition of HI in the precursor solution. In the absence of HI, a noticeable amount of CsI impurity is present (I CsI(110) /I Cs 2 SnI 6 (222) ¼ $0.160, 10 times more than that observed in the AACVD (+HI)-grown lms, as shown in Table 1). This indicates that the synthesis process (+HI or ÀHI) can signicantly inuence the crystalline phase composition. In the relatively I-poor environment (ÀHI), iodine vacancies (V I ) can form more easily. 29 A density functional theory (DFT) study suggests V I and interstitial tin (Sn i ) are the dominant defect types in Cs 2 SnI 6 . 29 The optimised molar concentration of HI in the precursor solution to minimise the CsI impurity phase was found to be 0.2 M, as described in the Experimental section. The concentration is roughly equivalent to twice the SnI 4 precursor molar concentration, in good agreement with the following reaction for formation of [SnI 6 ] 2À octahedra in Cs 2 SnI 6 : 50 Thus it appears that excess iodine (in an I-rich environment) facilitates the formation of the [SnI 6 ] 2À sub-lattice of the Cs 2 SnI 6 double perovskite material. Not only do the different preparation methods lead to distinct phase compositions but the stability can also be affected by the thin lm deposition method. The preparation route used to synthesise Cs 2 SnI 6 has been shown to be a critical factor governing its stability. 25 Samples prepared by spin coating were found to contain a signicant CsI impurity phase aer only 5 days under ambient air. 22,25 Fig. 2C shows the effect of aging in ambient air on samples prepared by spin coating. There is a signicant increase in CsI content with aging compared to the AACVD (+HI)-processed sample. The I CsI(110) /I Cs 2 SnI 6 (222) ratio of the pristine spin-coated (+HI) lm is attributed to a slightly higher initial impurity level ($0.034) relative to that of the pristine AACVD (+HI)-grown lm. Aer 100 days, a substantial amount ($0.600) of CsI is observed in the spin-coated lm. As described, the precursor solution (+HI) and the annealing process are identical between the AACVD and spin-coating processes. Therefore, the difference in stability can be attributed to the method used to prepare the thin lms. The amount of CsI appearing with time in the AACVD (+HI)-processed samples is compared with the results of Saparov et al. for lms deposited by vacuum deposition in Fig. S2 (ESI †). 25 This suggests that the stability of the Cs 2 SnI 6 lms fabricated by AACVD is comparable to or even better than that of lms made via vacuum vapour deposition. 25 We note that the AACVD (+HI)-grown lms were placed in an ambient environment with higher relative humidity (R.H. > 70%) when compared to the environment in the work of Saparov and others. 25 These results suggest that AACVD (+HI) can produce Cs 2 SnI 6 double perovskite lms that are stable in humid air.
Nevertheless, lms prepared by both AACVD and spincoating were degraded (to a lesser or greater extent) producing CsI when aged in ambient air. This may imply a bulk decomposition mechanism: Cs 2 SnI 6 / 2CsI + SnI 4 . ( Clearly, to generate CsI by decomposition of Cs 2 SnI 6 , SnI 4 should be formed in accordance with this reaction. However, no peaks which can be assigned to SnI 4 are observed in the XRD patterns. This may be because SnI 4 is somehow absent, which we discuss in later sections. In order to determine whether the lm quality could be inuencing the stability of the lms, the morphology of the as-prepared lms was examined using SEM, as shown in Fig. 3. The crystal structures at high magnication for samples produced by AACVD (+HI), AACVD (ÀHI), and spincoating (+HI) are shown in Fig. 3A, C, and E, respectively. All of the samples show micrometre-sized octahedral grains; however, the average size of the grains is larger for the AACVD (+HI)-Cs 2 SnI 6 lm. The shape of the crystals in lms made without HI is less well-dened, with what appears to be melted material at the surface.
SEM shows that the different processing techniques lead to differences in grain sizes. Larger grain sizes for the AACVD (+HI)-Cs 2 SnI 6 lm have a lower overall surface-to-volume ratio, and are thus more resilient to atmospheric degradation partially contributing to an enhanced stability. By measuring grain sizes for the different lms, we estimate that the surface area of the AACVD (+HI)-grown lm (Fig. 3A) is approximately 16Â that of the lm prepared by spin coating (Fig. 3E). However, the rate of degradation of the spin-coated lm is ca. 70Â that of the AACVD (+HI)-prepared lm, as estimated from XRD (Fig. S2 †). This suggests that additional factors are important in determining the stability of the AACVD-prepared lms.
At a lower magnication, the homogeneity of the lms is shown in Fig. 3B, D, and F. The AACVD (+HI)-processed lm (Fig. 3B) is very smooth, uniform, and (relatively) pin-hole free, which is important to its application as a light absorber in photovoltaic devices. 51 The other two samples ( Fig. 3D and F, AACVD (ÀHI) and spin-coated (+HI) respectively) show higher roughness, less homogeneity, and more pinholes. This could also contribute to differences in the stability between AACVD and spin-coated samples. Degradation sources such as water vapour or oxygen are able to enter the bulk via pinholes or pores more easily. This is consistent with the rapid and more obvious degradation of the spin-coated lm as measured by XRD compared to the AACVD-grown lms.
The bulk stoichiometries of the lms, as determined by energy-dispersive X-ray (EDX) spectroscopy, are also shown in Table 1. The AACVD (+HI)-prepared sample has the expected elemental stoichiometry of ideal Cs 2 SnI 6 within experimental error. The Cs : Sn ratios for all samples are very similar, and lie in the range 1.9-2.0. This suggests that if Cs 2 SnI 6 is converted Table 1 Summary of the results from characterisation of the different samples: I Cs(110) /I Cs 2 SnI 6 (222) indicates the ratios of the intensity of CsI (110) to Cs 2 SnI 6 (222) reflections from the XRD patterns. Elemental bulk stoichiometries (atomic) Cs/Sn and I/Sn are obtained from EDX. The optical energy bandgaps (E g ) are acquired from Tauc plots of the absorption spectra. The ratios of CsI to Cs 2 SnI 6 at the surface, acquired from XPS of the Cs 3d signal (Kratos) are also shown. Note '-' means no data available  into CsI, the CsI impurity and Sn degradation products remain in the lm since it is unlikely that both are lost at the same rate. Nevertheless, the ratios of iodine to tin do show some variations, indicating a clear loss of iodine with aging for the spincoated sample. Moreover, the slight lack of iodine in the fresh AACVD (ÀHI)-grown sample indicates that in an I-poor environment, V I defects tend to be generated. The optical band gap energies (E g ) of the lms were determined using ultraviolet-visible-near-infrared (UV-VIS-NIR) spectroscopy. The Tauc plots (Fig. 4) were obtained from the absorption spectra transformed from the diffuse reection spectra via the Kubelka-Munk function. Cs 2 SnI 6 is typically referred to as a direct bandgap material, which has been conrmed by photo-luminescence results. 22,23,52 By determining the onset of the plots, the value of E g was obtained for the different samples. In Fig. 4A, we observe that the AACVD (+HI)grown Cs 2 SnI 6 lm has a bandgap of ca. 1.30 eV, which is consistent with the smallest value in the range previously reported (1.3-1.6 eV). [22][23][24][25]52,53 Following aging for $100 days, the band gap energy increases slightly to ca. 1.35 eV. More signicant differences are observed in comparing lms prepared with and without the addition of HI. Fig. 4B shows that the AACVD (ÀHI)-prepared sample has a relatively high E g of ca. 1.60 eV. This again suggests that iodine deciency or the presence of CsI impurities can signicantly inuence the properties of the Cs 2 SnI 6 thin lms. It can be observed in Table 1 that an increase in CsI impurity level leads to higher E g , in good agreement with a recent nding. 23 Lee et al. found that, by using spin-coating with different parameters, varying levels of CsI impurity were formed in Cs 2 SnI 6 lms. 23 This results in a range of E g values from 1.3 to 1.4 eV as well as differences in carrier concentration and mobility. 23 Our study further demonstrates that the characteristics of Cs 2 SnI 6 lms are highly process and defect dependent. Therefore, innovation in the preparation of Cs 2 SnI 6 lms is of importance in optimising these characteristics.
These results illustrate the effects of degradation and the intrinsic differences between the bulk properties of the lms prepared by various methods. The surface characteristics are equally important since degradation occurs at the solid-vapour interfaces rst, where the surfaces are in contact with degradation-inducing reactants. The X-ray photoelectron spectra (XPS) shown in Fig. 5 depict the surface compositions of various fresh and aged lms obtained by recording the Cs 3d core levels. Two components, giving rise to two 3d 5/2,3/2 doublets, are observed for all of the double perovskite samples, both fresh and aged. The higher and lower binding energy (BE) Cs 3d 5/2 components, located at 725.1 AE 0.1 eV and 724.2 AE 0.1 eV, can be attributed to CsI and Cs 2 SnI 6 , respectively. 54 The assignment of the CsI features is also conrmed by comparison with spectra of a spin-coated pristine CsI lm, also shown in Fig. 5. Moreover, we observed that each sample showed different surface CsI-impurity levels; spectra of lms with a signicant amount of CsI show pronounced high-bindingenergy shoulders in the Cs 3d peaks (3d 5/2 and 3d 3/2 ). In order to easily compare, the area ratios of the CsI to Cs 2 SnI 6 Cs 3d 5/2 components tted are summarised in Table 1. The amount of CsI surface impurity in the AACVD (+HI)-prepared sample increases only slightly aer aging for $100 days (CsI : Cs 2 SnI 6 increases from 0.14 AE 0.01 to 0.20 AE 0.02). By contrast, that of the spin-coated lm signicantly increases (corresponding ratios 0.21 AE 0.02 to 0.35 AE 0.04) as a result of aging for a much shorter period (10 days).
This indicates the surface stability of the AACVD (+HI)-grown lm is better than the spin-coated one, and this is consistent with the ndings from the bulk properties. Interestingly, we observe that the CsI : Cs 2 SnI 6 ratio in the fresh sample fabricated without HI (0.34 AE 0.03) is similar to that of the aged spincoated sample. These observations suggest that the presence of CsI impurity at the surfaces is strongly inuenced by the preparation process. Table 1 shows that the variations between lms in the relative amount of CsI impurity found at the surface are similar to the trends observed in the bulk. While we cannot directly compare the concentrations determined from XPS with  the ratio I CsI(110) /I Cs 2 SnI 6 (222) obtained from XRD, we note that for the AACVD (+HI)-grown lms, the surface CsI : Cs 2 SnI 6 ratio is approximately ten times that of I CsI(110) /I Cs 2 SnI 6 (222) , whereas for the more degraded lms this factor is lower (approximately [2][3][4][5]. This suggests that in the initial stages of degradation, CsI is localised at the lm surfaces. In addition to the amounts of CsI impurity, the stoichiometry of the other elements is of importance in understanding the surface compositions of the various lms, and is shown in Table  S1. † In order to easily observe deviations from stoichiometry at the surface of the lms, we normalise the calculated ratios of the elements to [Cs] ¼ 2.0 (i.e. Cs 2 Sn x I y ), as for an ideal stoichiometric Cs 2 SnI 6 lm. We found that all samples, whether fresh or aged, and regardless of preparation method contain excess Sn ([Sn] > 1). These Sn atoms could originate from the SnI 4 precursor as we also observe excess I at the surface in all samples. For the fresh AACVD (+HI)-grown sample, the excess I is close to 4 times the excess Sn. This suggests the presence of SnI 4 in signicant excess, as the excess values for I and Sn are high. In addition, the amount of SnI 4 is far more than that of CsI, suggesting that the majority of the SnI 4 does not originate from the decomposition process in eqn (2). As fresh AACVDgrown lms have a larger excess of I and Sn, we speculate that the residual SnI 4 vapour in the AACVD tube attaches to the outermost surfaces of the lm. This may occur particularly during the cooling process at temperatures of $100-150 C, when the reaction between SnI 4 and CsI cannot happen. The more volatile component, SnI 4 , 50 continues to be transported along the tube and condenses on the surface of the lm during the cooling cycle. The relative changes in Sn concentration can also be observed in Fig. S3 † by comparing the intensities of the Cs 3d 5/2 and Sn 3p 3/2 peaks.
Aer aging the AACVD-grown lm in ambient air, the excess Sn decreases somewhat while the concentration of excess I is very signicantly reduced. This suggests the SnI 4 at the surfaces undergoes decomposition (loss of iodine) with I 2 or HI release from the lm by some mechanism. SnI 4 is known to be easily hydrolysed in the presence of water, so the SnI 4 at the surfaces is expected to react with moisture, forming hydrated SnO 2 (SnO 2 $nH 2 O). [55][56][57][58] By comparison, the changes in the surface composition of the spin-coated lm on aging are much smaller, starting from a smaller initial excess of Sn and I. This reinforces the suggestion that the excess SnI 4 on the surfaces of the AACVD-prepared samples is due to the nature of the process. Interestingly, the fresh AACVD (ÀHI)-grown sample contains a very signicant amount of excess Sn but a lower excess of I compared to the fresh AACVD (+HI)-prepared lm, consistent with the EDX nding. The large excess of Sn could be due to a signicant amount of I-decient tin iodide, and I-decient Sn sites within the perovskite phase. This also suggests again that without addition of HI, the formation of Cs 2 SnI 6 cannot be facilitated as per eqn (1).
It is clear that changes in surface composition of the lms are observed aer aging in air even for the relatively stable AACVD (+HI)-grown sample. It is therefore of interest to identify which gases in air (e.g. H 2 O vapour or O 2 ) are responsible for the surface degradation. NAP-XPS is a novel technique which allows investigation of the surface stability and degradation mechanisms of halide perovskite materials. 11 This is because this technique enables the surface composition of the lms to be measured under near-ambient pressure (NAP) conditions in a chamber lled with specic gases, and changes during gas exposure can be determined. In this study, the effects of water vapour and oxygen were explored. The Cs 3d NAP-XPS spectra recorded from fresh AACVD (+HI)-grown Cs 2 SnI 6 lms during and aer exposure to O 2 (5 mbar) and water vapour (9 mbar, corresponding to $30% RH at 25 C) are shown in Fig. 6. The two samples were prepared from the same initial sample by cutting it into two pieces. Following water vapour exposure (Fig. 6A), we observe that the CsI component does not increase relative to Cs 2 SnI 6 . Similarly, no obvious change was found in the Cs 3d spectra before and aer the lm was exposed to O 2 . These results suggest under those environments, the lms do not degrade signicantly via the mechanism in eqn (2), which should form CsI and SnI 4 . This is possibly as a result of insufficient exposure time as well as the limited gas pressure in the chamber. However, as compared to similar studies of MAPI, it is clear that Cs 2 SnI 6 shows higher resistance to moisture when both lms are exposed to environments with a RH of $30%. 11 Cs remains at the surfaces of the Cs 2 SnI 6 lm, whereas the equivalent 'A' cation in the ABX 3 perovskite structure of MAPI, the methylammonium ion (CH 3 NH 3 + ), decomposes and nitrogen is lost from the surfaces of the lm in the form of ammonia gas. Other core level spectra (I 3d, Sn 3d, C 1s, and O 1s) recorded from the lms are shown in Fig. S4 † for H 2 O vapour and Fig. S5 † for O 2 exposure, respectively. In both I and Sn 3d spectra, we do not observe BE changes or new components aer exposure. We note, however, that the chemical shis between Cs 2 SnI 6 , SnI 4 and CsI cannot be resolved in either I 3d or Sn 3d spectra. 59 In addition, the potential by-product formed by hydrolysis (SnO 2 -$nH 2 O), also has a similar Sn 3d BE to SnI 4 (Sn 3d 5/2 at BEs of $487.0 eV), 60-63 and therefore cannot be directly observed by XPS. Nevertheless, we can observe relative changes in quantied atomic concentrations aer exposure, which are summarised in Table 2. Again, we observe excess Sn and I in both fresh lms although the initial stoichiometries are different. This suggests that the amount of SnI 4 attaching on the lm surfaces is inhomogeneous and difficult to control in the AACVD process. Nevertheless, the presence of excess Sn and I at the lm surfaces is clearly revealed. For both exposure cases, the amount of Sn at the lm surface increases during exposure (relative to the amount of Cs). The I concentration increases aer the lm is exposed to moisture, by an amount consistent with an increase in the concentration of SnI 4 at the surface, whereas I is lost from the surface aer exposure to O 2 . There are no signicant changes in the amount of CsI relative to Cs 2 SnI 6 during exposure to H 2 O or O 2 ( Table 2). Changes are observed in the O content of the surfaces, as shown in Fig. S4, S5 and Table S2; † however, simultaneous accumulation of the C 1s spectra shows that these are largely associated with surface contamination, in particular with hydroxide attached to carbon. We found high levels of hydrocarbon contamination in the fresh lms (Table  S2 †), which can be attributed to ex situ preparation and utilisation of an organic solvent (DMF). We note that it is possible for trace carbon-containing and water contaminants to be introduced during exposure to even very pure gases in NAP-XPS, 64 as is suggested by Table S2. † Thus we cannot rule out the introduction of some H 2 O during the O 2 -exposure experiment. The picture that emerges is therefore a complex one, but it is clear that the surface is unstable with respect to Sn and I content during exposure to air. All the fresh AACVD (+HI)-grown lms contain an amount of CsI at the surface in excess of that observed in the bulk; however the amount of SnI 4 is considerably in excess of that expected by the simple decomposition reaction in eqn (2). We therefore suggest that the initial 'fresh' surface concentrations are determined by some decomposition, beginning at grain boundaries, to give CsI and SnI 4 , supplemented by an excess of Sn and I (largely as SnI 4 ) from residual precursor attaching during the AACVD growth. In our experiments on surface exposure and aging, we take the results from NAP-XPS to indicate the initial stages of degradation, and those from XPS aer 100 days to give information on longer-term effects. The NAP-XPS data suggest that the next stages of degradation involve an increase in surface Sn content, with some indication of further SnI 4 formation or migration from the bulk, while the longer term aging experiments suggest this is followed by hydrolysis of this SnI 4 over the longer term, accompanied by loss of I from the surface. We believe that the large excess of SnI 4 at the surface of the AACVD (+HI)-prepared lms provides a passivation functionality that protects Cs 2 SnI 6 from degradation when exposed to air. This is reinforced by the absence of any increase (within error) in CsI concentration at the surfaces aer exposure to H 2 O or O 2 in NAP-XPS, and by the rather small increase in surface and bulk concentrations aer 100 days aging in air. This suggests why AACVD-grown lms have better stability when they are placed in ambient air compared to those fabricated through spin coating.
To summarise, we propose the decomposition of Cs 2 SnI 6 occurs by the following steps: (1) initial decomposition at grain boundaries via eqn (2) to produce CsI and SnI 4 ; (2) some further migration of Sn, possibly as SnI 4 to the outermost surfaces; (3) SnI 4 is consumed by hydrolysis with loss of I 2 or HI; (4) Cs 2 SnI 6 is re-exposed due to removal of SnI 4 and therefore decomposed to CsI and SnI 4 in air; (5) excess SnI 4 is consumed again via (3).

Conclusions
To conclude, we have successfully fabricated ambient-air-stable Cs 2 SnI 6 double perovskite lms resistive to decomposition in ambient air via a novel deposition method, AACVD. Compared to spin-coating, a reduction in the amounts of CsI impurity phase in the lms made by AACVD has been demonstrated. We also nd that addition of HI to the precursor solution reduces not only the amount of CsI in the lm but the overall iodine deciency. Moreover, the AACVD (HI+)-grown lms show superior stability in humid ambient air. No further CsI is formed at the lm surface on exposure to up to 9 mbar of H 2 O vapour or 5 mbar O 2 . We nd that excess SnI 4 present at the surfaces of AACVD (+HI)-grown lms (most probably from precursor adsorption) may act as a protective layer to prevent Cs 2 SnI 6 from degradation. We propose the degradation proceeds by decomposition into CsI and SnI 4 , followed by hydrolysis and loss of the latter. Ultimately, this study provides Table 2 Ratios of the concentrations of different components in a fresh AACVD (+HI)-prepared film determined from NAP-XPS before and after exposure to 9 mbar H 2 O or 5 mbar O 2 . All elements are normalised to [Cs] ¼ 2.0, consistent with Table 1. Excess Sn (excess I) refers to the amount of SnI that cannot be accounted for in the Cs 2 SnI 6 (Cs 2 SnI 6 + CsI) phase(s) an insight into novel stable-halide-perovskite fabrication and an understanding of the enhanced stability of Cs 2 SnI 6 lms made by AACVD.

Conflicts of interest
There are no conicts to declare.