Unreacted PbI 2 as a Double-Edged Sword for Enhancing the Performance of Perovskite Solar Cells

Transcription

1 Article pubs.acs.org/jacs Unreacted PbI 2 as a Double-Edged Sword for Enhancing the Performance of Perovskite Solar Cells T. Jesper Jacobsson,*,, Juan-Pablo Correa-Baena, Elham Halvani Anaraki,, Bertrand Philippe, Samuel D. Stranks,,# Marine E. F. Bouduban, Wolfgang Tress, Kurt Schenk, Joe l Teuscher, Jacques-E. Moser, Ha kan Rensmo, and Anders Hagfeldt*,, University of Cambridge, Department of Chemistry, Lensfield Road, Cambridge CB2 1EW, U.K. Laboratory for Photomolecular Science, Institute of Chemical Sciences and Engineering, Ećole Polytechnique Fe deŕale de Lausanne, CH-1015-Lausanne, Switzerland Department of Materials Engineering, Isfahan university of Technology, Isfahan, , Iran Department of Physics and Astronomy, Uppsala University, Box 516, Uppsala, Sweden Research Laboratory of Electronics, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, Massachusetts 02139, United States # Cavendish Laboratory, JJ Thomson Avenue, Cambridge CB3 0HE, U.K. Photochemical Dynamics Group, Institute of Chemical Sciences and Engineering, Ećole Polytechnique Fe deŕale de Lausanne, CH-1015-Lausanne, Switzerland Laboratory of Photonics and Interfaces, Institute of Chemical Sciences and Engineering, Ećole Polytechnique Fe deŕale de Lausanne, CH-1015-Lausanne, Switzerland Ećole Polytechnique Fe deŕale de Lausanne, CH-1015-Lausanne, Switzerland Department of Chemistry Ångstro m Laboratory, Uppsala University, Box 538, Uppsala, Sweden *S Supporting Information ABSTRACT: Lead halide perovskites have over the past few years attracted considerable interest as photo absorbers in PV applications with record efficiencies now reaching 22%. It has recently been found that not only the composition but also the precise stoichiometry is important for the device performance. Recent reports have, for example, demonstrated small amount of PbI 2 in the perovskite films to be beneficial for the overall performance of both the standard perovskite, CH 3 NH 3 PbI 3, as well as for the mixed perovskites (CH 3 NH 3 ) x - (CH(NH 2 ) 2 ) (1 x) PbBr y I (3 y). In this work a broad range of characterization techniques including X-ray diffraction (XRD), scanning electron microscopy (SEM), transmission electron microscopy (TEM), photo electron spectroscopy (PES), transient absorption spectroscopy (TAS), UV vis, electroluminescence (EL), photoluminescence (PL), and confocal PL mapping have been used to further understand the importance of remnant PbI 2 in perovskite solar cells. Our best devices were over 18% efficient, and had in line with previous results a small amount of excess PbI 2. For the PbI 2 -deficient samples, the photocurrent dropped, which could be attributed to accumulation of organic species at the grain boundaries, low charge carrier mobility, and decreased electron injection into the TiO 2. The PbI 2 -deficient compositions did, however, also have advantages. The record V oc wasashighas1.20v and was found in PbI 2 -deficient samples. This was correlated with high crystal quality, longer charge carrier lifetimes, and high PL yields and was rationalized as a consequence of the dynamics of the perovskite formation. We further found the ion migration to be obstructed in the PbI 2 -deficient samples, which decreased the JV hysteresis and increased the photostability. PbI 2 -deficient synthesis conditions can thus be used to deposit perovskites with excellent crystal quality but with the downside of grain boundaries enriched in organic species, which act as a barrier toward current transport. Exploring ways to tune the synthesis conditions to give the high crystal quality obtained under PbI 2 -poor condition while maintaining the favorable grain boundary characteristics obtained under PbI 2 -rich conditions would thus be a strategy toward more efficiency devices. INTRODUCTION Lead halide perovskites have during the past few years attracted considerable interest as light absorbers in solar cells. The first Received: June 19, 2016 Published: July 20, American Chemical Society DOI: /jacs.6b06320 J. Am. Chem. Soc. 2016, 138,

2 Journal of the American Chemical Society paper on the matter was published in After some key advances in the following years 2 6 the field has expanded rapidly with top efficiencies now reaching beyond 22%. 7 There is also a realistic hope of perovskites reaching the market, either as standalone cells or as top cells in tandem configurations with conventional solar cells The standard perovskite in the community has so far been methylammonium lead iodide, MAPbI 3, but both the organic ion and the halide can be substituted. MA has for example been replaced with formamidinium (FA), and iodine by chlorine 15 and bromine. 16 The MA/FA and Br/I ratios can also be changed simultaneously, 14,17,18 and the best cells at the moment are based on mixed perovskites with compositions around MA 1/3 FA 2/3 Pb- (Br 1/6 I 5/6 ) 3. 19,20 For a given perovskite, the precise stoichiometry is important as well. Most notably, a slight excess of PbI 2 in the perovskite film has been observed to be beneficial. 19,21,22 The best performing cells have a few percent excess PbI 2 in them but exactly why is still unclear. This bring us to the core of this paper, which aims at investigating how an excess, or a deficiency, of PbI 2 affect the physics and the performance of mixed perovskite devices. The possible benefit of excess PbI 2 was first reported in studies using two-step synthesis methods. 22,23 A PbI 2 film is then being deposited and subsequently exposed to solutions of the organic salts whereupon the perovskite forms. 23,24 The overall stoichiometry depends on the length of the second step. Short times results in uncompleted conversion and some PbI 2 will consequently remain in the final film. 23,25,26 Due to trivial geometry, PbI 2 will to a greater extent be found close to the back contact. A thick PbI 2 layer between the back contact and the perovskite results in electronic insulation and poor cell performance The highest cell efficiencies were, however, obtained when most but not all of the PbI 2 was converted into the perovskite. 26 For exposure and subsequent annealing times slightly longer than required for complete conversion, the perovskite begins to Article degrade into PbI 2 and a volatile organic component. 28 This PbI 2, formed as a consequence of perovskite degradation, could also be beneficial. 29,30 Even longer reaction times result in severe and detrimental degradation and a fine balance is to be found. One-step methods, which we use in this study, provide a higher degree of control of the stoichiometry and give consistent results indicating that small amounts of PbI 2 in the deposited films are beneficial for device performance. 19,21 Not everyone has, however, found the excess PbI 2 to be beneficial, 25,31,32 though those results appear to be exceptions. The reason behind this observation, the mechanistic details, and whether or not PbI 2 is truly essential for high efficiency devices or merely the cause of favorable secondary effects, achievable by other approaches, are still open questions. A few hypotheses focus on energy level alignment. PbI 2 may for example passivate the TiO 2 interface 19,26,29,33 and thereby decrease hole recombination, either due to surface passivation or band edge matching between TiO 2, PbI 2, and the perovskite (Figure 1a). Another hypothesis is that PbI 2 could facilitate electron injection into TiO 2. 21,34 As an excess of PbI 2 are beneficial also when blocking layers other than TiO 2 are used, i.e., SnO 2, 17 those appear to be less likely explanations. At the perovskite/htl interface PbI 2 could, according to a similar reasoning, act as an electron blocking layer, facilitate hole injection, and thereby decrease recombination 29,35 (Figure 1b). If the PbI 2 layers are too thick, or if the energy matching is wrong, they may instead insulate individual grain and block charge transfer (Figure 1c,d). Another set of hypotheses focuses on grain boundaries. The composition of the center of the perovskite grains could possibly be fairly insensitive toward small changes in overall stoichiometry. The grain boundaries and the regions between the grains would thus by necessity be highly sensitive to the overall composition (Figure 1e g). This could affect, for example, defect states, dangling bonds, conductivity, doping, and ion migration. Figure 1. Consequences of PbI 2 in the perovskite film related to possible energy alinements suggested in the literature and an artistic illustration of difference grain boundary character as a function of overall stoichiometry. (a) PbI 2 as a passivating layer at the back contact. (b) PbI 2 as a passivation layer next to the hole-selective layer. (c) PbI 2 as an electron blocking layer next to the back contact. (d) PbI 2 as charge carrier barrier between perovskite grains. (e) Grain boundary with a large surplus of PbI 2. (f) Grain boundary with a small deficiency of organic species. (g) Grain boundary with a large surplus of organic species DOI: /jacs.6b06320 J. Am. Chem. Soc. 2016, 138,

3 Journal of the American Chemical Society Some results indicate recombination to be faster within grain boundaries deficient in PbI 2. 29,36 PbI 2 would thus act as a passivation layer between the grains. 36 This is supported by computations that indicate that PbI 2 -rich conditions results in good termination with fewer intra band gap states as possible recombination centers. 37 Other computations show that the conditions under which the perovskite is formed influences the type and the number of defects within the perovskite 38 and that synthesis under iodine poor conditions leads to less bulk defects related to deep traps. Those conclusions are not directly generalizable to the role of PbI 2 in perovskite films, but as they show a correlation between formation of bulk defect and synthesis conditions, this may be important. The general view is that the perovskite grain boundaries are defect tolerant and intrinsically benign, which is supported by theoretical simulations, 39 and there are claims that charge transport is especially efficient in the grain boundaries. 40 This may, however, change as a function of the composition of the grain boundaries. Finally, there are hypotheses stating that PbI 2 is not beneficial in itself, but rather are correlated with other secondary advantageous effects. That could for example be larger perovskite grains 19,21 or improved crystallinity. 21 Remnant PbI 2 may thus be a redundant component as those benefits may be engineered along other paths as well. A deeper understanding of the interplay between PbI 2, overall stoichiometry, and device performance would be valuable for further solar cell development. In this paper, we do not reach complete understanding, but we present the most comprehensive investigation up to date on the topic, which takes us closer. Here we discuss the aggregated results from a wide range of techniques used to investigate perovskites with compositions that stretches from a large deficiency of PbI 2 (henceforth referred Article to as understoichiometric) to a large surplus of PbI 2 (henceforth referred to as overstoichiometric). The results illuminate connections between device performance, stoichiometry, ion movement, recombination behavior, and trap formation valuable for future work toward improving perovskite devices. EXPERIMENTAL METHODS The perovskite solar cells were prepared in line with previous reports. 17,19,20,41 49 The solar cell stack is composed of FTO, a compact TiO 2 -layer, mesoporous TiO 2, a mixed perovskite deposited by spin coating using a one-step antisolvent method, doped Spiro-MeOTAD as a hole conductor, and an evaporated gold contact. A detailed description of the synthesis is found in the Supporting Information. Unique for this sample series is the variation in stoichiometry. All the final precursor solutions had the following concentrations: [Pb2+] = 1.25 M, [PbBr 2 ]= 0.22 M, [PbI 2 ] = 1.04 M, [MABr] = [PbBr 2 ] = 0.22 M. To probe the importance of PbI 2, [FAI] was varied. When we henceforth state a composition as +10% with respect to PbI 2 we mean that [FAI] = [PbI 2 ] 0.9. Analogously, we by 10% mean that [FAI] = [PbI 2 ] 1.1. Plus 10% thus mean a surplus of PbI 2 in the solution (or a deficiency of FA), which result in excess PbI 2 in the perovskite films. Perovskites with different stoichiometry where investigated by XRD, SEM, TEM, UV vis spectroscopy, steady state photoluminescence (PL), transient absorption spectroscopy (TAS), confocal PL mapping, Photoelectron spectroscopy (PES), Hard X-ray PES (HAXPES), Soft X-ray PES (SOXPES). Complete devices were characterized in terms of JV characteristics, external photocurrent efficiency (EQE) and electroluminescence. The details concerning the equipment and the experimental parameters are found in the Supporting Information. RESULTS Basic Optical and Crystallographic Characterization. X-ray diffraction was measured on a subset of the samples with varying stoichiometry (Figure 2a) stretching from Figure 2. (a) XRD data normalized with respect to the strongest perovskite reflex (001) for seven different stoichiometries. Data are shifted in height to be easier to compare. The full set of diffraction figures are given in the SI (b) The (001) peak of the normalized data in (a). (c) UV vis for the same samples as in (a). (d) PL data for the same samples measured at the day of deposition. (e) and (f) Top view SEM images for a PbI 2 -deficient and a PbI 2 -rich sample. The scale bars are 200 nm. The full set of SEM images are found in the SI DOI: /jacs.6b06320 J. Am. Chem. Soc. 2016, 138,

4 Journal of the American Chemical Society 10% understoichiometric (PbI 2 -deficient) to 20% overstoichiometric (PbI 2 -rich). One single crystalline perovskite phase formed in all samples. By comparing normalized diffraction data for the (001)-reflection (Figure 2b), no systematic peak shifts were observed, which show the composition of the crystalline perovskite to be unaffected by the overall stoichiometry. A small random shift is seen, which indicates a slight shift in the I/Br ratio. 20 On the basis of previous quantifications, the observed shifts are too small to be of significant importance for the device performance. 20 The symmetry of the perovskite phase was found to be cubic, as expected based on previous measurements 20,45 and theoretical considerations. 50 The diffraction peak in Figure 2b is wider for the two understoichiometric samples, which indicates smaller crystallites. By Scherrer s equation we estimate those to be around 30 nm for the understoichiometric samples and 100 nm or more for the rest. A feature size of 30 nm does, however, not correlate with the SEM data (Figure 2e,f) discussed below where the grains appears to be a few hundred nm, and thus might consist of several crystallites. Apart from the perovskite, the only crystalline phase observed is PbI 2, which, as expected, increases in amount for the more overstoichiometric samples. The PbI 2 -peaks are rather narrow. The PbI 2 in the films are thus either amorphous or in the form of rather large crystallites, i.e., more than 100 nm. UV vis absorption was measured on the same subset of samples (Figure 2c). The overall absorption behavior was rather unaffected by stoichiometry and had a low background signal indicating uniform films of high quality with respect to macroscopic inhomogeneity. There appears to be a difference in absolute absorption between the samples. However, this was caused by a measurement artifact and the effect disappeared in control measurements using an integrating sphere (SI). The band gaps were extracted from the absorption data, as elaborated in the Supporting Information, and were found to be between 1.62 and 1.65 ev. Valence band spectroscopy was also used and reveled no dependence in the valence band edge position with respect to the stoichiometry (SI). PL was measured both on the day of deposition and after 30 days of storage in dry air. Directly after deposition, one single strong peak was observed in all samples (Figure 2d). The peaks were centered close to the band gap energy, with the exception of a slight blue shift in the +5% sample. The PL and the absorption data thus supports the conclusion form the XRD data that the perovskite phase essentially is the same in the different samples. The PL intensities were higher for the stoichiometric and the understoichiometric samples. At the day of synthesis, the difference was up to 1 order of magnitude (Figure 2d). The difference hade some batch dependence, indicating that competing mechanisms are involved in determining the emissivity. After a month in dry air, the difference in intensity decreased moderately (SI). Large difference in PL intensity have been reported in the literature and have been reported to different defects and there quenching, 54 suggesting either less defects or efficient blocking of them in the PbI 2 -poor samples. Under strong illumination (10 sun), the dominant peak in the overstoichiometric films began to broaden and split into two peaks (SI). The new peaks were located at lower energies, consistent with formation of an iodine-enriched phase. Such a secondary phase can already in small quantities act as efficient recombination centers detrimental for device performance. 20 This was not observed in the understoichiometric samples, indicating higher photo stability. On the basis of JV hysteresis for Article the corresponding devices, we relate this effect to differences in ion migration, which appears to be slower over PbI 2 -deficient grain boundaries. This photoinduced ion migration is discussed in more detail in a parallel paper. 55 SEM was used to further investigate the grain size and the surface morphology, which are known to influence the device performance. 19,43,56 Cross section images showed the film thickness to be rather unaffected by the stoichiometry (SI). Examples of top view SEM images are found in Figure 2e,f. A complete set of images are found in the Supporting Information. The surface morphology was rather smooth and the pin holes were few for all samples. The overstoichiometric films were slightly rougher, but we have not quantified this. In terms of grain size, the cross section images are consistent with the top views. A typical grain size is in the order of 200 nm for all samples, but possibly slightly smaller around the stoichiometric composition. The grain size is smaller than the cross section of the perovskite film. A large fraction of the photo generated charge carriers thus has to pass one or more grain boundaries before collection. A grain size of 200 nm is inconsistent with the XRD data for the understoichiometric samples. This possibly means that the grains observed in the SEM images are monocrystalline for the overstoichiometric samples but polycrystalline in the understoichiometric ones. This could possibly correlate with the striations observed on the grains in the understoichiometric samples (Figure 2e), which could be a proxy for planar defects and a change in the stacking direction within the grains. This indicates that a difference in device performance more likely are an effect of crystal quality, stacking within individual grains, polycrystallinity of grains, and the property of the grain boundaries rather than due to the average size of the perovskite grains per se. Photoelectron Spectroscopy. If the composition of the perovskite phase is essentially unaffected by overall stoichiometry, the grain boundaries will according to the mass balance be more dependent. The most accessible grain boundaries are located at the surface. Although possibly different from the internal grain boundaries, they provide an indication of what to expect based on the overall stoichiometry. As a baseline for further comparison, reference spectra were measured for FAPbI 3, MAPbBr 3, and FA 0.85 MA 0.15 PbBr 0.45 I 2 (Figure 3a). The I 4d core levels (53 48 ev), and the Pb 5d core level (25 15 ev) are all in line with previous results found in the 80 0 evrange. 57 So is also the Br 3d core level (72 67 ev) (Figure 3b). Those signals thus originate from a similar probing depth, which means that the peak intensities easily can be linked to the composition via atomic cross sections. Since the analysis only includes d-orbitals, the quantification is also rather insensitive to factors linked to the sample, the detector, and the X-ray polarization configuration. The spectra were intensity normalized with respect to the Pb 5d5/2 peak. For FA 0.85 MA 0.15 PbBr 0.45 I 2.55, the experimental Br/Pb and I/Pb ratio was 0.45 and 2.36 respectively. The theoretical values based on the bulk stoichiometry are 0.45 and The I/Pb ratio for FAPbI 3 was estimated to be 2.4 and the Br/Pb ratio in MAPbBr 3 to 2.93 (instead of 3 theoretically). The presence of PbI 2 at the probing depth decreases the I/Pb ratio and explains the difference between the experimental and theoretical values. Worth to mention is that those values indicate that pure stoichiometric MAPbBr 3 more easily is formed than FAPbI 3, which contains more PbI 2 at the surface. The MA and FA features can easily be distinguished from each other by the N 1s spectra (Figure 3c). FAPbI 3 and MAPbBr 3 give DOI: /jacs.6b06320 J. Am. Chem. Soc. 2016, 138,

5 Journal of the American Chemical Society Article Figure 3. (a) Overview spectra of the mixed perovskite (stoichiometric, i.e., 0), FAPbI 3 and MAPbBr 3 deposited on amorphous SnO 2 /FTO and of the bare amorphous SnO 2 /FTO substrate recorded with a photon energy of 4000 ev (in red, brown, orange, and black respectively). The SnO 2 substrate was normalized with respect to intensity vs Sn 3d 5/2. No substrate peaks were observed in the perovskite spectra, demonstrating full coverage. (b) Br 3d/I 4d/Pb 5d (c) N 1s and (d) C 1s core level spectra of the mixed perovskite, FAPbI 3 and MAPbBr 3. (e) and (f) Br 3d/I 4d/Pb 5d core level peaks of the perovskite materials with three different stoichiometries: 10%, 0, +10% in light gray, gray, and dark gray respectively recorded at 4000 ev (e) and 758 ev (f). All the spectra were intensity normalized to the Pb 5d 5/2 core level peaks. A zoom on the I 4d peak is shown on the right. The spectra of the PbI 2 precursors are shown as reference (in red). a single peak located at and ev respectively. Both peaks are found and are separable in the mixed perovskite, although at slightly higher energy ( and ev). The ratio between the metal and the halides can be determined based on Figure 3b and the N 1s signal can be used to estimate the cation ratio. For the mixed perovskite, MA:FA was estimated to 12:88, which is relatively close to the overall experimental composition of 15:85. Quite similar information is given by the C 1s spectra where FAPbI 3 and MAPbBr 3 has a specific peak at and ev (Figure 3d). A third peak is observed at ev, which is attributed to common surface contaminations usually found on ex situ prepared samples. This peak can overlap DOI: /jacs.6b06320 J. Am. Chem. Soc. 2016, 138,

6 Journal of the American Chemical Society with the MA and FA signal (see SI) and the Cs1 signal was therefore not used for quantifications. PES-spectra in the energy range of 80 0 ev for three different stoichiometries ( 10%, 0, + 10%) of the mixed perovskite, FA 0.85 MA 0.15 PbBr 0.45 I 2.55, are found in Figure 3e,f. Three different photon energies were used in order to vary the probing depth, 758, 2100, and 4000 ev. Using values from Tanuma et al., 58 a probing depth of approximately 18 nm are explored with photon energies of 4000 ev (HAXPES). A higher surface sensitivity with a probing depth of around 5 nm is achieved with a photon energy of 758 ev (SOXPES). At the intermediate energy of 2100 ev, the probing depth was estimated to 11 nm. The probing depths are reasonable estimates but the exact values depend on details in the material structure. By normalizing the spectra with respect to the lead component, the variation and difference in the iodine peak can be compared. At the highest probing depth (4000 ev, 18 nm), the peak intensity of I 4d clearly is, as expected, different between the perovskites and PbI 2.Adifference between the three samples is visible indicating some stoichiometric difference between them. At hν = 758 ev, which represent a smaller probing depth, the same differences can be observed between the samples and a larger change can be observed when compared to PbI 2. On the basis of these spectra, Table 1 presents the calculated I/Pb (a), Table 1. Intensity Ratios between Different Core Levels Calculated from Experimental Results a (a) I/Pb intensity ratio hν [ev] probing depth [nm] 10% 0 +10% PbI (b) Br/Pb intensity ratio (theo. 0.45) Article This indicate unreacted FAI in the surface layer. The I/Pb ratio is still decreasing from 10% to +10%, with the highest value for the PbI 2 -deficient sample, which consequently has a relative excess of FAI. Those data are consistent with a model of perovskite growth that starts with precipitation of PbI 2 followed by intercalation of the organic ions under the formation of the perovskite. This is opposed to a single step growth of the perovskite. The perovskite would then grow from the outside in, which preferably will leave unreacted organic species at the surface rather than PbI 2, which preferentially would be found deeper down in the grains if the conversion is not complete. The Br/Pb ratio (Table 1b) does not show any clear trend with respect to stoichiometry. This agrees with the observation that PbBr 2 more easily forms the perovskite, which should make the Br/Pb ratio less sensitive to the amount of PbI 2 in the system. When the probing depth decreases, an increase is observed but not as significant as for the I/Pb ratio. This either indicates a slight bromide enrichment in the surface perovskite, or the presence of unreacted bromide salts on the surface but at a lower amount than the iodide salts. The I/Br ratio (Table 1c) confirm the point raised earlier. The I/Br ratio decreases from 10% to +10% in line with the excess of FAI, especially when the synthesis is deficient in PbI 2. When the probing depth is decreased, the I/Br ratio decreases slightly in agreement with a surface with a more bromide rich perovskite phase. The FA/MA ratio for the different stoichiometries was determined based on the N 1s core level spectra at 4000 and 758 ev (Figure 4). No significant changes were observed between the hν [ev] probing depth [nm] 10% 0 +10% PbI (c) I/Br intensity ratio (theo. 5.66) hν [ev] probing depth [nm] 10% 0 +10% PbI a (a) I/Pb, (b) Br/Pb, and (c) I/Br of the mixed perovskites with three different stoichiometry as function of the excitation energy (4000, 2100, and 758 ev). The I/Pb ratio of PbI 2 is also given. The relative atomic percentage of Pb, I, and Br are presented in the SI. Br/Pb (b), and I/Br (c) ratio estimated with three different photon energies for the three different mixed perovskite materials. PbI 2 is included as a reference and to validate our quantifications. The estimated I/Pb ratio is given in Table 1a. The I/Pb ratio for PbI 2 is close to 2 in agreement with its chemical formula. For the deepest probing depth (4000 ev, 18 nm), the experimental values for the mixed perovskite are below 2.55, which is the theoretical value, and decreases from the 10% to the +10% stoichiometry. This is in line with more unreacted PbI 2 in the films as is expected based on the stoichiometry of the precursor solutions. Also the PbI 2 -deficient sample ( 10%) has a ratio below 2.55 suggesting noncomplete conversion into the perovskite, in line with the XRD data. For the more surface sensitive technique (758 ev, 5 nm), the I/Pb ratios are higher than Figure 4. N 1s core level peaks of the perovskite materials with three different stoichiometries: 10%, 0, +10% in light gray, gray and dark gray respectively recorded at 4000 ev (left) and 758 ev (right). The proportion of FA is indicated on each spectrum. three samples or when the analysis depth was changed and FA was always representing between 85 and 90% of the total amount of cations. This stability may be surprising considering the fluctuation observed in Table 1. However, we have seen that at the surface, the slight diminution of FAPbI 3 is balanced by the presence of FAI, which will keep the FA/MA ratio relatively constant. The main difference observed at the surface of the perovskite samples with different stoichiometries are schematically summarized in Figure 5. The PES measurements confirm that the perovskite phase is rather unaffected by the stoichiometry, they demonstrate a surplus of organic species at the surface dependent on the DOI: /jacs.6b06320 J. Am. Chem. Soc. 2016, 138,

7 Journal of the American Chemical Society Article Figure 5. Schematic illustration summarizing the main difference observed by PES between the three different stoichiometries: 10%, 0, and +10%. Figure 6. (a,b). TEM images of a PbI 2 -deficient sample. (c f). TEM images of a PbI 2 -rich sample. (e) and (f) is an image of the same spot illustrating sample degradation under the electron beam. The samples were prepared by scratching of a perovskite film deposited on SLG with a razor blade, in which the TEM-grid then was rubbed. stoichiometry, and that unreacted PbI 2 is located within the perovskite film. TEM. Samples of over and understoichiometric perovskites were investigated by TEM to get more direct information about the grain boundaries. Much of the sample was too thick to be electron transparent but some grains of a suitable size could be found. For the high resolution images in Figure 6a d, lattice fringes, which indicate a high degree of crystallinity, are seen for both the under- and overstoichiometric sample. Those appear to stretch out to the surface of the grains to a larger degree in the samples with more PbI 2 (compare Figure 6b and 6d). That would support the grain boundary hypothesis presented based on the PES data. We must stress that this not is a solid conclusion but rather a weak indication in line with the other data. The reason for this ambiguity is sample instability. At high magnifications, the surface of the grains sometimes appears to melt and shrink, making it difficult to distinguish real structural effects from beam induced artifacts. An example of the beam damage is given by comparing Figure 6e,f where the same spot is photographed after different exposure times. The beam induced degradation are affected by the stoichiometry. In the overstoichiometric sample, dark grains a few nm across are seen (Figure 6c and 6e), which gets more pronounced with longer exposure times (Figure 6f). This was not observed to the same extent in the understoichiometric sample. Those grains are likely PbI 2. The higher frequency of these dots in the overstoichiometric sample and the uniform distribution indicates that not all unreacted PbI 2 is located at the grain boundary but it also is distributed, in amorphous form, within the particles. This is in line with the PES-results and points toward a crystallization process where the perovskite grains are grown from the outside of PbI 2 -grains by diffusion and intercalation of the organic ions. A surplus of organic ions would facilitate complete conversion whereas a deficiency of organic ion may leave small pockets of unreacted amorphous PbI 2 within the perovskite grains. This would further indicate a higher crystal DOI: /jacs.6b06320 J. Am. Chem. Soc. 2016, 138,

8 Journal of the American Chemical Society Article Figure 7. (a) Transient absorbance spectra (smoothed) of four different stoichiometries: -10%, 0%, 10%, and 20%. Measured at 2 ps after excitation with a 480 nm laser. Data is normalized with respect to the absorbance at 480 nm. (b) Transient absorbance dynamics after 480 nm excitation, measured at the maximum of the ground state bleaching in (a), which are 760, 750, 756 and 723 nm for the 10%, 0%, 10% and 20% sample, respectively. Thin lines are raw data and bold lines are multiexponential fits. (c) Modeled electro-absorption (first derivative of the absorption spectrum, blue line) for the samples with 10% excess PbI 2 superimposed on the smoothed transient absorbance spectrum at 2 ps (dashed black line). The inset is the absorbance as a function of wavelength for the same film. core quality in the PbI 2 -deficient samples but with less welldefined grain boundaries. This is in line with a high V oc and poor transport properties discussed below. Within the bigger perovskite grains, these small inclusions could act as seeds that under stress accelerate the decomposition of the perovskite into PbI 2 and organic salts, explaining the higher stability observed in the understoichiometric sample. Transient Absorption Measurements. To access the time dynamics of the photoexcited states, transient absorption spectroscopy (TAS) was carried out on perovskite films of four different stoichiometries: 10%, 0%, +10%, and +20%. The dominant feature of the TAS spectra measured at 2 ps after a 480 nm excitation pulse (Figure 7a) was a negative signal with a maximum amplitude around 760 nm. This was assigned to a combination of ground-state bleaching (GSB) and stimulated emission (SE) The maximum was observed to shift from 760 nm for the understoichiometric sample to 723 nm for the most overstoichiometric sample. Such a blue shift is consistent with a slightly more bromide rich composition but could also be related to differences in the strain or in the tilt of the PbI 2 -octahedra, which is known to influence the optical properties. 50,63 65 The spectral features also broaden as one moves toward the most overstoichiometric sample. Both the spectral shift and the broadening indicate a more complex material with potentially more strain and compositional inhomogeneity. That does not correlate with lower device performance as seen below, but a narrower bleach peak could possibly be associated with a higher crystal core quality and a higher V oc. We finally wish to stress the appearance of a distinct spectral signature (negative at 500 nm) in the overstoichiometric samples (+20%), which we attribute to PbI 2. The time-dependent behavior of the GSB+SE signal was obtained by extracting the time dynamics at its respective maxima (Figure 7b). The time traces were fitted with a triexponential model convoluted with a Gaussian instrument response function. The resulting time constants: τ 1, τ 2,andτ 3 are provided in Table 2. τ 1 describes the rise of the transient, τ 2 denotes the recovery of the GSB+SE signal, and τ 3 is associated with long-lived species. On the contrary to τ 1 and τ 3, the second time constant, τ 2, shows a clear trend with the stoichiometry and increases with increasing PbI 2 -content and range from 0.91 ps for the most understoichiometric sample to 24 ps for the PbI 2 -richest sample. The exact mechanism for the difference in the GSB+SE signal Table 2. Fitting Constants Extracted from a Tri-exponential Function Convoluted with a Gaussian IRF with a FWHM = 80 fs sample Λ max [nm] τ 1 [ps] τ 2 [ps] τ 3 [ps] 10% % % % recovery is at this stage hard to pinpoint, and those decay times should not be confused with the charge carrier lifetimes under solar illumination, which are substantially longer. Another interesting feature is the presence of oscillations between 500 and 700 nm in the transient absorption spectra (Figure 7a). As emerges from Figure 7c, featuring the 2 ps TA-spectra for one of the overstoichiometric samples (10%) superimposed to a modeled electro-absorption signal, those oscillations match the spectrum of a short-lived photoinduced electro-absorption signal. The presence of such a signal is interesting as it is indicative of the presence of photoinduced carriers exhibiting a permanent dipole and yielding a sensible electric field. At the moment we attribute this to the presence of PbI 2 - domains within the materials. Time-Resolved PL Mapping. To further understand the recombination kinetics, confocal PL maps of samples with different stoichiometries were sampled. PL measurements give a relative measure of the fraction of radiative and nonradiative decay and to achieve the highest device performance, nonradiative decay should be minimized or eliminated. 66 To ensure morphologies comparable to other studies while avoiding significant quenching of the PL signal from electron injection, the mesoporous TiO 2 scaffold was coated with a conformal 5 nm thick layer of insulating Al 2 O 3 by ALD, on top of which the perovskites were deposited. Confocal PL intensity maps with excitation through the glass are given in Figure 8a d and the corresponding intensity histograms in Figure 8e. The understoichiometric sample ( 10%) shows the most uniform emission distribution and the absolute intensity is almost an order of magnitude higher than for the other compositions, consistent with the PL data in Figure 2d. The other compositions have similar net intensity magnitudes, though we note that the +10% composition appears to have two different distributions of intensity of emitting species and a lot of heterogeneity in the emission. The time-resolved PL decay (Figure 8f) when exited through the DOI: /jacs.6b06320 J. Am. Chem. Soc. 2016, 138,

9 Journal of the American Chemical Society Article Figure 8. Confocal photoluminescence (PL) intensity maps of the (a) 10%, (b) 0%, (c) +10%, (d) +20% configurations infiltrated into a nonquenching mesoporous scaffold and measured through the glass side. The maps were normalized to their peak value. (e) Histograms of the absolute intensities extracted from the PL maps. (f) Time-resolved PL decays of the samples when excited through the top side with the same laser settings as above. (g,h). Time-resolved PL measurements for the (g) 10% and (h) +10% compositions when exciting the noninjecting (TiO 2 /Al 2 O 3 ) sample from the top-side (black) or the injecting sample (TiO 2 ) from either the top-side (blue) or the glass-side (green). Samples were photoexcited with a 405 nm laser with a repetition rate of 0.5 MHz and a fluence of 2 μj/cm 2 /pulse. Figure 9. (a) An illustration of the device architecture. (b) A cross section SEM image of a typical device (+15%). The scale bar is 300 nm. (c) A photo of a devices. (d) JV data for one of the best devices. The full set of JV curves are given in the Supporting Information. top-side gives an idea of the average lifetimes. These reflect the intensity distributions where the longest lifetime is seen for the understoichiometric sample ( 10%). This suggests that the highest V oc may be found for the understoichiometric compositions. To investigate how well the PL is quenched from electron injection into the mesoporous TiO 2, measurements were performed on samples without the insulating Al 2 O 3 -interlayer. For simplicity, only the 10% and +10% compositions are compared (Figure 8g,h). When excited through the top side, most of the charge carriers are generated far from the injection electrode. For the understoichiometric sample ( 10%), no substantial quenching was observed when compared to the analogous noninjecting sample with insulating Al 2 O 3. This suggest a short charge carrier diffusion length, 60,67 and that not all carriers are able to reach the quenching TiO 2. Even when exciting through the glass, which generates more of the photoexcited charge carriers closer to the perovskite/tio 2 interface, there is still a long-lived tail suggesting that the electron injection is suboptimal and a number of electrons are not injected into the TiO 2. In contrast, the relative quenching in the overstoichiometric sample (+10%) is much more substantial. The difference between exciting the injecting sample from the top-side and the glass-side is much less, suggesting longer charge carrier diffusion lengths and improved electron injection. These results collectively suggest that devices with the understoichiometric ( 10%) composition would have more limited J sc than the overstoichiometric configurations. Discussion and Device Performance. A large number of complete devices were made with compositions ranging from PbI 2 -deficient, or understoichiometric, ( 10%) to PbI 2 -rich, or overstoichiometric, (+30%). The device architecture and the physical appearance are given in Figure 9a c. The power conversion efficiencies vary over a broad range. The best cells had efficiencies above 18% (Figure 9d), whereas the worse compositions gave cells close to 10%. To focus on the specific effect on PbI 2 and to get a clearer parameter window, we were, as elaborated in the Experimental Methods section, using compositions slightly different from the optimized ones giving top efficiencies above 20%. 20 The compositions used here are close enough and 18% efficiency is high DOI: /jacs.6b06320 J. Am. Chem. Soc. 2016, 138,

10 Journal of the American Chemical Society Article Figure 10. (a) η against stoichiometry. (b) J sc (c) EQE against wavelength. (d) FF. (e) V oc. (f) Hysteresis index H. The complete set of JV curves for all individual samples and the corresponding data in tabular format are found in the SI. enough for the physics explored here to be generalizable to the highest performing devices. The cell efficiencies decreased for the more PbI 2 -deficient samples (Figure 10a), which is in line with previous reports. 19,21,26 The drop in efficiency was mainly a consequence of decreased photocurrents (Figure 10b). Those are robust observations reproduced in all batches made. On the basis of the PL data (Figure 8), the decrease in J sc is probably an effect of poor charge transport in the understoichiometric samples, possibly caused by an enrichment of organic species in the grain boundaries. The PES, XRD, UV vis and PL measurements together with mass balance constraints indicates that the grain boundaries get enriched in organic species in the understoichiometric samples. Band diagram computations predict that the organic ions not contribute with states close to the band edges. 50,68 Transport of photoexcited thermalized charge carriers is thus mainly occurring in bands originating from the lead halide framework. PbI 2 -deficient conditions result in a disruption of that framework by enriching the grain boundaries with organic ions. The grain boundaries could then act as barriers toward charge transport, explaining the lower photocurrents. The XRD measurements indicate that grains in the understoichiometric samples may be polycrystalline, which also would increase the number of internal grain boundaries the charge carriers would need to pass in order to be collected. External quantum efficiency (EQE) measurements show that for higher photon energies, the collection is independent of stoichiometry (Figure 10c). For the understoichiometric samples, lower values are observed for photon energies ranging from the absorption onset up to approximately 2.25 ev. That correlates with the drop in photocurrent and could potentially be a result of trapping of low energy photocarriers close to the band edges, or to the shorter diffusion length and the inferior electron injection indicated in Figure 8. The fill factor is essentially independent of stoichiometry (Figure 10d). The V oc does, however, tend to decrease when the amount of PbI 2 increases (Figure 10e). The highest measured V oc was as high as 1.20 V and found in one of the PbI 2 -deficient samples V is remarkably high and close to what could be hoped for given a band gap of 1.64 ev. PbI 2 may thus be involved in two counterbalancing effects. Even if the device performance is better for PbI 2 -rich devices, the excess PbI 2 may have a negative impact in decreasing the photovoltage. This balance may not always tip in the same direction, which would explain contradictory results in other reports, 18,19,21 as well as why this effect was not reproduced in every single batch. We should also point out that we have measured high, even though not equally high, V oc values for overstoichiometric samples (1.17 V with a mesoporpus TiO 2 architecture and 1.19 with a planar SnO 2 electron selective contact). Still, the higest V oc measured in the group for the mixed perovskite is for the PbI 2 -deficient samples here discussed, and high V oc values for PbI 2 -deficient cells were reproduced in most, even if not in all, batches. To verify whether the high open-circuit voltage correlates with high luminescence efficiencies, as expected from fundamental theory, electroluminescence was measured on devices with different stoichiometry. This was indeed the case, even though we observed measurement related variations, possibly due to a reversible instability related to hysteresis and degradation under long time storage before the measurements. The highest value of the external electroluminescence quantum efficiency (EQE EL ) in this series was measured for a 5% device and approached 0.8% at a current of 20 ma/cm 2 (SI). This is in good agreement with a V oc of 1.20 V 19,69 and is as far as we know a record value. The EQE EL increases approximately linearly with driving current as expected for a trap-recombination limited current 70 (SI). The high V oc for the PbI 2 -poor devices are also in good agreement with the longer lifetimes demonstrated in Figure 8. These high V oc values may be explained by higher crystal quality in the PbI 2 -poor devices. The higher PL intensities and the high EQE EL supports that claim. The TEM data (Figure 6), which DOI: /jacs.6b06320 J. Am. Chem. Soc. 2016, 138,

11 Journal of the American Chemical Society indicate more small PbI 2 -inclusions within the perovskite grains in the PbI 2 -rich samples support this as well. The dynamics of the crystallization and formation of the perovskite phase may be able to give a rationalization behind these observations. The perovskite formation is likely initiated by PbI 2 crystallization followed by perovskite formation by diffusion of the organic ions into the PbI 2 structure. For two-step protocols this is trivially true, but it seems to be the case also for the one-step protocol. This is supported by the PES data, which indicate a surplus of organic salts on the surface also for PbI 2 -deficient stoichiometries. In this mode of growth, the simple concept of chemical equilibria implies that PbI 2 -rich conditions will result in noncomplete conversion of the PbI 2 -grains, which thus will contain inclusions of unreacted PbI 2 (as seen in TEM section), which due to strain will be amorphous (as given by the peak width in XRD). As the highest performing devices are PbI 2 -rich, these PbI 2 inclusions are obviously not detrimental. A surplus of organic ions, as fond under PbI 2 -deficient conditions, could shift the equilibrium toward more complete conversion of PbI 2 and thus generate a more homogeneous and pure perovskite phase. Taken on its own, that would be intrinsically better, as indicated by the PL and EQE EL data. The downside is that the grain boundaries get enriched in organic species, as indicated by the PES data and a simple mass balance argument. That in turn correlates with shorter charge carrier diffusion length and slower injection into the back contact (Figure 8g,h) as well as a decrease in the EQE for lower photon energies (Figure 10c); both of which decreases the photocurrent (Figure 10b). There is a hysteresis in the IV-curves (Figure 9d and SI). This have been widely observed and discussed, and hypotheses for its origin involve ion migration, 75,76 interfacial charge transfer, 77 capacitive effects, 72 and ferroelectric effects. 78 A hysteresis index, H,isdefined in eq 1, where J b and J f is the current in the backward and forward scan, respectively. H = Voc oc J ( V)d V J( V)dV Voc J ( V)dV V 0 b 0 f 0 b The hysteresis was generally higher in the PbI 2 -rich samples (Figure 10f), whereas it was rather small in the stoichiometric and in most of the PbI 2 -deficient samples. Our data suggests that difference in ion migration of the grain boundaries are the cause behind this trend. The grain boundaries in the PbI 2 -deficient samples have, as discussed above, a higher organic content, that possible is more amorphous. The activation energy for ion migration is likely higher over those grain boundaries compared to more PbI 2 -rich grain boundaries. Those grain boundaries would not have the same layered structure as crystalline PbI 2 where ion migration could occur between the layers in much the same way as the organic ions intercalate the PbI 2 structure during perovskite formation. If ion migration is the main cause of the hysteresis, which goes toward being the dominating hypothesis, 76,79 it would reasonably be smaller if the ions to a larger degree would be confined within individual grains. This hypothesis is supported by PL measurements performed as a function of time under illumination. 54,55,80 We have performed a number of such measurements, which briefly are reproduced in the SI. We generally observe PL enhancements due to a reduction in nonradiative decay pathways and we recently reported that these PL enhancements are associated with a photoinduced migration of iodide-55. We find that this effect is most evident in the overstoichiometric samples, which are likely (1) Article to have the highest excess of iodide, and this is consistent with this configuration showing the most exaggerated hysteresis effects. In contrast, the effects are greatly reduced in the stoichiometric and understoichiometric samples, which have less excess iodide and hysteresis effects. A broadening of the emission was also observed, which indicate the onset of a phase separation into more bromide and iodide rich phases. This effect was considerably more pronounced in the PbI 2 -rich samples, consistent with lower barriers for ion migration and a higher hysteresis. From previous experiments, 20 we know that this type of phase separation can be detrimental for device performance. The stoichiometry is thus expected to affect also the stability, which will be further investigated. SUMMARY AND CONCLUSIONS We have used a broad range of techniques to investigate how the amount of remnant PbI 2 influences the properties of perovskite films and solar cells. We found the amount of PbI 2 to have a small or negligible effect on the phase composition and the surface morphology. The effect on the grain boundaries and the crystal quality was significant and had consequences for the charge carrier lifetimes, ion migration, photo luminescence, charge carrier injection, hysteresis, as well as on the device performance. The highest device performance was, in line with previous results, obtained with PbI 2 -rich samples and devices with efficiencies over 18% were produced. For PbI 2 -deficient perovskites, the photocurrent and consequently also the device performance dropped. This was attributed to an accumulation of organic species in the grain boundaries, which were found to hinder the charge carrier transport and decrease the electron injection into the back contact. The PbI 2 -deficient stoichiometries did, however, also have advantages. The highest V oc were found for understoichiometric samples, and values as high as 1.20 V were measured. This could be correlated with high crystal quality, longer charge carrier lifetimes, and high PL and EQE EL yields. On the basis of PES, TEM, XRD, SEM, and PL data, this could be rationalized as a consequence of the dynamics of the perovskite formation, which also for one-step protocols appears to be a two-step process where the perovskite is formed by intercalation of organic species in precipitated PbI 2 -grains. A surplus of organic salts, i.e., a PbI 2 -deficiency, leads to a more complete conversion of PbI 2 into the perovskite phase, which results in a more homogeneous perovskite phase of higher crystal quality. We also found the ion migration to be obstructed in the understoichiometric samples which decreased the JV hysteresis and increased the photostability. One of the core conclusions is that PbI 2 -deficient synthesis conditions can result in a perovskite phase with very high crystal quality, expressed in long lifetimes, and high PL and EQE EL yields. The downside is grain boundaries enriched in organic species that provide a barrier toward current transport and decreases the photocurrent. An interesting prospect worth further investigations is to explore if the synthesis conditions can be tuned in such a way that the crystal quality here obtained under PbI 2 -deficient conditions could be maintained while simultaneously generating grain boundaries with as favorable properties as under PbI 2 -rich conditions. That could potentially be a way toward new record devices, and some work have been started along this direction. 81, DOI: /jacs.6b06320 J. Am. Chem. Soc. 2016, 138,

12 Journal of the American Chemical Society ASSOCIATED CONTENT *S Supporting Information The Supporting Information is available free of charge on the ACS Publications website at DOI: /jacs.6b Details concerning synthesis, device production and characterization. Additional XRD, UV vis, PL, SEM, PES, and valence band spectroscopy data. Band gap determinations. Full set of JV-figures and corresponding device data in tabular form. (PDF) AUTHOR INFORMATION Corresponding Authors Notes The authors declare no competing financial interest. ACKNOWLEDGMENTS GRAPHENE project supported by the European Commission Seventh Framework Program under contract is gratefully acknowledged. BP and HR thank HZB for the allocation of synchrotron radiation beam time. SDS has received funding from the People Programme (Marie Curie Actions) of the European Union s Seventh Framework Programme (FP7/ ) under REA grant agreement number PIOF-GA REFERENCES (1) Kojima, A.; Teshima, K.; Shirai, Y.; Miyasaka, T. J. Am. Chem. Soc. 2009, 131, (2) Im, J. H.; Lee, C. R.; Lee, J. W.; Park, S. W.; Park, N. G. Nanoscale 2011, 3, (3) Lee, M. M.; Teuscher, J.; Miyasaka, T.; Murakami, T. N.; Snaith, H. J. Science 2012, 338, 643. (4) Kim, H. S.; Lee, C. R.; Im, J. H.; Lee, K. B.; Moehl, T.; Marchioro, A.; Moon, S. J.; Humphry-Baker, R.; Yum, J. H.; Moser, J. E.; Gratzel, M.; Park, N. G. Sci. Rep. 2012, 2, 1. (5) Heo, J. H.; Im, S. H.; Noh, J. H.; Mandal, T. N.; Lim, C. S.; Chang, J. A.; Lee, Y. H.; Kim, H. J.; Sarkar, A.; Nazeeruddin, M. K.; Gratzel, M.; Seok, S. I. Nat. Photonics 2013, 7, 487. (6) Burschka, J.; Pellet, N.; Moon, S. J.; Humphry-Baker, R.; Gao, P.; Nazeeruddin, M. K.; Gratzel, M. Nature 2013, 499, 316. (7) NREL efficiency chart. (8) Snaith, H. J. J. Phys. Chem. Lett. 2013, 4, (9) Bailie, C. D.; Christoforo, M. G.; Mailoa, J. P.; Bowring, A. R.; Unger, E. L.; Nguyen, W. H.; Burschka, J.; Pellet, N.; Lee, J. Z.; Graetzel, M.; Noufi, R.; Buonassisi, T.; Salleo, A.; McGehee, M. D. Energy Environ. Sci. 2015, 8, 956. (10) Albrecht, S.; Saliba, M.; Baena, J. P. C.; Lang, F.; Kegelmann, L.; Mews, M.; Steier, L.; Abate, A.; Rappich, J.; Korte, L.; Schlatmann, R.; Nazeeruddin, M. K.; Hagfeldt, A.; Graetzel, M.; Rech, B. Energy Environ. Sci. 2016, 9, 81. (11) Koh, T. M.; Fu, K.; Fang, Y.; Chen, S.; Sum, T. C.; Mathews, N.; Mhaisalkar, S. G.; Boix, P. P.; Baikie, T. J. Phys. Chem. C 2014, 118, (12) Pellet, N.; Gao, P.; Gregori, G.; Yang, T.-Y.; Nazeeruddin, M. K.; Maier, J.; Graetzel, M. Angew. Chem., Int. Ed. 2014, 53, (13) Lee, J.-W.; Seol, D.-J.; Cho, A.-N.; Park, N.-G. Adv. Mater. 2014, 26, (14) Yang, W. S.; Noh, J. H.; Jeon, N. J.; Kim, Y. C.; Ryu, S.; Seo, J.; Seok, S. I. Science 2015, 348, (15) Knop, O.; Wasylishen, R. E.; White, M. A.; Cameron, T. S.; Vanoort, M. J. M. Can. J. Chem. 1990, 68, 412. (16) Heo, J. H.; Song, D. H.; Im, S. H. Adv. Mater. 2014, 26, (17) Baena, J. P. C.; Steier, L.; Tress, W.; Saliba, M.; Neutzner, S.; Matsui, T.; Giordano, F.; Jacobsson, T. J.; Kandada, A. R. S.; Article Zakeeruddin, S. M.; Petrozza, A.; Abate, A.; Nazeeruddin, M. K.; Graetzel, M.; Hagfeldt, A. Energy Environ. Sci. 2015, 8, (18) Jeon, N. J.; Noh, J. H.; Yang, W. S.; Kim, Y. C.; Ryu, S.; Seo, J.; Seok, S. I. Nature 2015, 517, 476. (19) Bi, D.; Tress, W.; Dar, M. I.; Gao, P.; Luo, J.; Renevier, C.; Schenk, K.; Abate, A.; Giordano, F.; Correa Baena, J.-P.; Decoppet, J.-D.; Zakeeruddin, S. M.; Nazeeruddin, M. K.; Graẗzel, M.; Hagfeldt, A. Sci. Adv. 2016, 2, e (20) Jesper Jacobsson, T.; Correa-Baena, J.-P.; Pazoki, M.; Saliba, M.; Schenk, K.; Gratzel, M.; Hagfeldt, A. Energy Environ. Sci. 2016, 9, (21) Roldan-Carmona, C.; Gratia, P.; Zimmermann, I.; Grancini, G.; Gao, P.; Graetzel, M.; Nazeeruddin, M. K. Energy Environ. Sci. 2015, 8, (22) Kim, Y. C.; Jeon, N. J.; Noh, J. H.; Yang, W. S.; Seo, J.; Yun, J. S.; Ho-Baillie, A.; Huang, S.; Green, M. A.; Seidel, J.; Ahn, T. K.; Il Seok, S. Adv. Energ. Mater. 2016, 6, (23) Burschka, J.; Pellet, N.; Moon, S.-J.; Humphry-Baker, R.; Gao, P.; Nazeeruddin, M. K.; Graetzel, M. Nature 2013, 499, 316. (24) Chen, Q.; Zhou, H.; Hong, Z.; Luo, S.; Duan, H.-S.; Wang, H.-H.; Liu, Y.; Li, G.; Yang, Y. J. Am. Chem. Soc. 2014, 136, 622. (25) Lee, Y. H.; Luo, J.; Humphry-Baker, R.; Gao, P.; Graetzel, M.; Nazeeruddin, M. K. Adv. Funct. Mater. 2015, 25, (26) Cao, D. H.; Stoumpos, C. C.; Malliakas, C. D.; Katz, M. J.; Farha, O. K.; Hupp, J. T.; Kanatzidis, M. G. APL Mater. 2014, 2, (27) Bi, D.; El-Zohry, A. M.; Hagfeldt, A.; Boschloo, G. ACS Photonics 2015, 2, 589. (28) Supasai, T.; Rujisamphan, N.; Ullrich, K.; Chemseddine, A.; Dittrich, T. Appl. Phys. Lett. 2013, 103, (29) Chen, Q.; Zhou, H.; Song, T.-B.; Luo, S.; Hong, Z.; Duan, H.-S.; Dou, L.; Liu, Y.; Yang, Y. Nano Lett. 2014, 14, (30) Wang, S.; Dong, W.; Fang, X.; Zhang, Q.; Zhou, S.; Deng, Z.; Tao, R.; Shao, J.; Xia, R.; Song, C.; Hu, L.; Zhu, J. Nanoscale 2016, 8, (31) Salim, T.; Sun, S.; Abe, Y.; Krishna, A.; Grimsdale, A. C.; Lam, Y. M. J. Mater. Chem. A 2015, 3, (32) Shao, Y.; Xiao, Z.; Bi, C.; Yuan, Y.; Huang, J. Nat. Commun. 2014, 5, (33) Nakayashiki, S.; Daisuke, H.; Ogomi, Y.; Hayase, S. J. Photonics Energy 2015, 5, (34) Somsongkul, V.; Lang, F.; Jeong, A. R.; Rusu, M.; Arunchaiya, M.; Dittrich, T. Phys. Status Solidi RRL 2014, 8, 763. (35) Calloni, A.; Abate, A.; Bussetti, G.; Berti, G.; Yivlialin, R.; Ciccacci, F.; Duo, L. J. Phys. Chem. C 2015, 119, (36) Wang, L.; McCleese, C.; Kovalsky, A.; Zhao, Y.; Burda, C. J. Am. Chem. Soc. 2014, 136, (37) Haruyama, J.; Sodeyama, K.; Han, L.; Tateyama, Y. J. Phys. Chem. Lett. 2014, 5, (38) Buin, A.; Pietsch, P.; Xu, J.; Voznyy, O.; Ip, A. H.; Comin, R.; Sargent, E. H. Nano Lett. 2014, 14, (39) Yin, W.-J.; Shi, T.; Yan, Y. Adv. Mater. 2014, 26, (40) Yun, J. S.; Ho-Baillie, A.; Huang, S.; Woo, S. H.; Heo, Y.; Seidel, J.; Huang, F.; Cheng, Y.-B.; Green, M. A. J. Phys. Chem. Lett. 2015, 6, 875. (41) Jacobsson, T. J.; Tress, W.; Correa-Baena, J.-P.; Edvinsson, T.; Hagfeldt, A. J. Phys. Chem. C 2016, 120, (42) Guerrero, A.; Garcia-Belmonte, G.; Mora-Sero, I.; Bisquert, J.; Kang, Y. S.; Jacobsson, T. J.; Correa-Baena, J.-P.; Hagfeldt, A. J. Phys. Chem. C 2016, 120, (43) Correa-Baena, J.-P.; Anaya, M.; Lozano, G.; Tress, W.; Domanski, K.; Saliba, M.; Matsui, T.; Jacobsson, T. J.; Calvo, M. E.; Abate, A.; Graẗzel, M.; Míguez, H.; Hagfeldt, A. Adv. Mater. 2016, 28, (44) Giordano, F.; Abate, A.; Correa Baena, J. P.; Saliba, M.; Matsui, T.; Im, S. H.; Zakeeruddin, S. M.; Nazeeruddin, M. K.; Hagfeldt, A.; Graetzel, M. Nat. Commun. 2016, 7, (45) Jacobsson, T. J.; Schwan, L. J.; Ottosson, M.; Hagfeldt, A.; Edvinsson, T. Inorg. Chem. 2015, 54, (46) Zhao, Y.; Zhu, K. J. Phys. Chem. Lett. 2014, 5, (47) Jeon, N. J.; Noh, J. H.; Kim, Y. C.; Yang, W. S.; Ryu, S.; Il Seol, S. Nat. Mater. 2014, 13, 897. DOI: /jacs.6b06320 J. Am. Chem. Soc. 2016, 138,

13 Journal of the American Chemical Society (48) Abate, A.; Leijtens, T.; Pathak, S.; Teuscher, J.; Avolio, R.; Errico, M. E.; Kirkpatrik, J.; Ball, J. M.; Docampo, P.; McPherson, I.; Snaith, H. J. Phys. Chem. Chem. Phys. 2013, 15, (49) Abate, A.; Staff, D. R.; Hollman, D. J.; Snaith, H. J.; Walker, A. B. Phys. Chem. Chem. Phys. 2014, 16, (50) Jacobsson, T. J.; Pazoki, M.; Hagfeldt, A.; Edvinsson, T. J. Phys. Chem. C 2015, 119, (51) Jacobsson, T. J.; Edvinsson, T. J. Phys. Chem. C 2012, 116, (52) Fondell, M.; Jacobsson, T. J.; Boman, M.; Edvinsson, T. J. Mater. Chem. A 2014, 2, (53) Jacobsson, T. J.; Edvinsson, T. J. Phys. Chem. C 2014, 118, (54) Tian, Y.; Peter, M.; Unger, E.; Abdellah, M.; Zheng, K.; Pullerits, T.; Yartsev, A.; Sundstrom, V.; Scheblykin, I. G. Phys. Chem. Chem. Phys. 2015, 17, (55) Dequilettes, D. W.; Zhang, W.; Burlakov, V. M.; Graham, D. J.; Leijtens, T.; Osherov, A.; Bulovic, V.; Snaith, H. J.; Ginger, D. S.; Stranks, S. D. Nat. Commun. 2016, 7, 1. (56) Li, G.; Ching, K. L.; Ho, J. Y. L.; Wong, M.; Kwok, H.-S. Adv. Energ. Mater. 2015, 5, (57) Philippe, B.; Park, B.-W.; Lindblad, R.; Oscarsson, J.; Ahmadi, S.; Johansson, E. M. J.; Rensmo, H. Chem. Mater. 2015, 27, (58) Tanuma, S.; Powell, C. J.; Penn, D. R. Surf. Interface Anal. 1994, 21, 165. (59) Kim, H.-S.; Lee, C.-R.; Im, J.-H.; Lee, K.-B.; Moehl, T.; Marchioro, A.; Moon, S.-J.; Humphry-Baker, R.; Yum, J.-H.; Moser, J. E.; Graetzel, M.; Park, N.-G. Sci. Rep. 2012, DOI: /srep (60) Xing, G.; Mathews, N.; Sun, S.; Lim, S. S.; Lam, Y. M.; Graetzel, M.; Mhaisalkar, S.; Sum, T. C. Science 2013, 342, 344. (61) Manser, J. S.; Reid, B.; Kamat, P. V. J. Phys. Chem. C 2015, 119, (62) Stamplecoskie, K. G.; Manser, J. S.; Kamat, P. V. Energy Environ. Sci. 2015, 8, 208. (63) Filip, M. R.; Eperon, G. E.; Snaith, H. J.; Giustino, F. Nat. Commun. 2014, 5, 1. (64) Geng, W.; Zhang, L.; Zhang, Y.-N.; Lau, W.-M.; Liu, L.-M. J. Phys. Chem. C 2014, 118, (65) Quarti, C.; Mosconi, E.; Ball, J. M.; D Innocenzo, V.; Tao, C.; Pathak, S.; Snaith, H. J.; Petrozza, A.; De Angelis, F. Energy Environ. Sci. 2016, 9, 155. (66) Miller, O. D.; Yablonovitch, E.; Kurtz, S. R. IEEE. J. Photovolt. 2012, 2, 303. (67) Stranks, S. D.; Eperon, G. E.; Grancini, G.; Menelaou, C.; Alcocer, M. J. P.; Leijtens, T.; Herz, L. M.; Petrozza, A.; Snaith, H. J. Science 2013, 342, 341. (68) Pazoki, M.; Jacobsson, T. J.; Hagfeldt, A.; Boschloo, G.; Edvinsson, T. Phys. Rev. B: Condens. Matter Mater. Phys. 2016, 93, (69) Tress, W.; Marinova, N.; Inganas, O.; Nazeeruddin, M. K.; Zakeeruddin, S. M.; Graetzel, M. Adv. Energ. Mater. 2015, 5, (70) Marinova, N.; Tress, W.; Humphry-Baker, R.; Dar, M. I.; Bojinov, V.; Zakeeruddin, S. M.; Nazeeruddin, M. K.; Gratzel, M. ACS Nano 2015, 9, (71) Sanchez, R. S.; Gonzalez-Pedro, V.; Lee, J.-W.; Park, N.-G.; Kang, Y. S.; Mora-Sero, I.; Bisquert, J. J. Phys. Chem. Lett. 2014, 5, (72) Kim, H.-S.; Park, N.-G. J. Phys. Chem. Lett. 2014, 5, (73) Snaith, H. J.; Abate, A.; Ball, J. M.; Eperon, G. E.; Leijtens, T.; Noel, N. K.; Stranks, S. D.; Wang, J. T.-W.; Wojciechowski, K.; Zhang, W. J. Phys. Chem. Lett. 2014, 5, (74) Unger, E. L.; Hoke, E. T.; Bailie, C. D.; Nguyen, W. H.; Bowring, A. R.; Heumueller, T.; Christoforo, M. G.; McGehee, M. D. Energy Environ. Sci. 2014, 7, (75) Azpiroz, J. M.; Mosconi, E.; Bisquert, J.; De Angelis, F. Energy Environ. Sci. 2015, 8, (76) Tress, W.; Marinova, N.; Moehl, T.; Zakeeruddin, S. M.; Nazeeruddin, M. K.; Gratzel, M. Energy Environ. Sci. 2015, 8, 995. (77) Jena, A. K.; Chen, H.-W.; Kogo, A.; Sanehira, Y.; Ikegami, M.; Miyasaka, T. ACS Appl. Mater. Interfaces 2015, 7, (78) Frost, J. M.; Butler, K. T.; Walsh, A. APL Mater. 2014, 2, Article (79) Meloni, S.; Moehl, T.; Tress, W.; Franckevicius, M.; Saliba, M.; Lee, Y. H.; Gao, P.; Nazeeruddin, M. K.; Zakeeruddin, S. M.; Rothlisberger, U.; Graetzel, M. Nat. Commun. 2016, 7, (80) Hoke, E. T.; Slotcavage, D. J.; Dohner, E. R.; Bowring, A. R.; Karunadasa, H. I.; McGehee, M. D. Chem. Sci. 2015, 6, 613. (81) Zhang, T.; Guo, N.; Li, G.; Qian, X.; Zhao, Y. Nano Energy 2016, 26, 50. (82) Son, D.-Y.; Lee, J.-W.; Choi, Y. J.; Jang, I.-H.; Lee, S.; Yoo, P. J.; Shin, H.; Ahn, N.; Choi, M.; Kim, D.; Park, N.-G. Nature Energy 2016, 1, DOI: /jacs.6b06320 J. Am. Chem. Soc. 2016, 138,

14

15 Supporting information Unreacted PbI2 as a Double-Edged Sword for Enhancing the Performance of Perovskite Solar Cells T. Jesper Jacobsson 1,2, Juan-Pablo Correa-Baena 2, Elham Halvani Anaraki 2,3, Bertrand Philippe 4, Samuel D. Stranks 5,6, Marine E. F. Bouduban 7, Wolfgang Tress 8, Kurt Schenk 9, Joël Teuscher 7, Jacques-E. Moser 7, Håkan Rensmo 4, Anders Hagfeldt 2,10 1) University of Cambridge, Department of Chemistry, Lensfield Road, Cambridge CB2 1EW, UK 2) Laboratory for Photomolecular Science, Institute of Chemical Sciences and Engineering, École Polytechnique Fédérale de Lausanne, CH-1015-Lausanne, Switzerland 3) Department of Materials Engineering, Isfahan university of Technology, Isfahan, , Iran 4) Department of Physics and Astronomy, Uppsala University, Box 516, 75120, Uppsala, Sweden 5) Research Laboratory of Electronics, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, Massachusetts 02139, United States 6) Cavendish Laboratory, JJ Thomson Avenue, Cambridge CB3 0HE, United Kingdom 7) Photochemical Dynamics Group, Institute of Chemical Sciences and Engineering, École Polytechnique Fédérale de Lausanne, CH-1015-Lausanne, Switzerland 8) Laboratory of Photonics and Interfaces, Institute of Chemical Sciences and Engineering, École Polytechnique Fédérale de Lausanne, CH-1015-Lausanne, Switzerland 9) École Polytechnique Fédérale de Lausanne, CH-1015-Lausanne, Switzerland 10) Department of Chemistry - Ångström Laboratory, Uppsala University, Box 538, Uppsala, Sweden Jacobsson.jesper.work@gmail.com, +46 (0) Anders.hagfeldt@epfl.ch, +41 (0) ~S1~

16 Experimental Methods Perovskite and device preparation As substrate for the devices, FTO NSG 10 was used. The substrates were cleaned in freshly prepared piranha solution composed of five parts concentrated H 2SO 4 and two parts 30 % H 2O 2. Proper protection should be used while handling this solution as it is highly aggressive. The substrates were soaked in the piranha solution for 10 minutes and then rinsed in water and then ethanol. They were thereafter treated in a UV-ozone cleaner for 10 minutes. A hole blocking layer of TiO 2 was deposited on the cleaned FTO substrates using spray pyrolysis. The spray solution was composed of ethanol, acetyl acetone, and titanium diisopropoxide (30% in isopropanol) in the proportions 90:4:6 by volume. Oxygen at a base pressure of 1 bar was used as a carrier gas. The glass substrates were heated to 450 C on a hotplate and kept at that temperature for 15 minutes prior to the spraying. After an additional 30 minutes at 450 C, the sprayed glass substrates were slowly cooled to room temperature. 10 ml of spray solution was used to cover 40 cm 2 of substrates. This procedure gives a compact layer of anatase with a thickness of around nm. On top of the compact TiO 2-layer deposited by spray pyrolysis, a mesoporous scaffold of TiO 2 nanoparticles was deposited by spin-coating. TiO 2 paste (30 NR-D) was bought from Dyesol and was dissolved in ethanol at a concentration of 150 mg/ml. On each substrate ( cm) 50 µl of the TiO 2 solution was applied and spin-coated at 4000 rpm, with an acceleration of 2000 rpm/s, for 10 s. A piece of scotch tape was used on one side of the substrates to prevent the mesoporous TiO 2 to form where the front contacts were to be deposit. The substrates with mesoporous TiO 2 were sintered at 450ºC in air on a hot plate for 30 minutes and then slowly cooled to ambient temperature. For XRD-measurements, and time dependent spectroscopy, the perovskites were deposited on soda lime glass (SLG) covered by 15 nm amorphous SnO 2. The SnO 2 film increases the wettability of the precursor solutions that give perovskite films of higher quality. The SnO 2 was deposited by atomic layer deposition, ALD, at 120 C using a Savanah ALD 100 from Cambridge Nanotech Inc. As a tin precursor, Tetrakis(dimethylamino)tin(IV), bought from Stem Chemicals Inc., was used. Ozone was used as the oxidising agent. The ozone was produced by an ozone generator fed with oxygen gas ( % pure, Carbagas) producing O 3 at a concentration of 13% in O 2. Nitrogen was used as a carrier gas with a flow rate of 5 sccm. The substrates with SnO 2 were directly prior to perovskite deposition treated in a UV-ozone cleaner for 10 minute. This process for deposition of an electron selective SnO 2 contact is described in more detail in a previous publication 1, and has been used as an electron selective contact in high efficiency planar devices 1. For PL-mapping measurements, the perovskites were deposited on thin microscope cover glass slides. For those measurements the perovskite was deposited on: glass, glass/tio 2, glass/al 2O 3, glass/tio 2/mesoporous TiO 2, glass/tio 2/mesoporous TiO 2/Al 2O 3. The Al 2O 3 was deposited by ALD and was a few nm thick. Prior to perovskite deposition, the substrates with mesoporous TiO 2 underwent a lithium treatment which has been found to be beneficial for the device performance 2. On the substrates, 100 µl of a 35mM Lithium bistrifluoromethanesulfonimidate (Li-TFSI) in acetonitrile was applied and spun at 3000 rpm for 10 s. The substrates were then thermally annealed in air at 450 C for 30 minutes and then slowly cooled to 150ºC where after they were brought directly into a glovebox for perovskites deposition. The best perovskite cells presented in the literature are based on mixed lead perovskites using a mixture of bromide and iodide on the halogen position and a mixture of methyl ammonium and formamidinium as organic ion. The best perovskites have a composition around MA 2/6FA 4/6Pb(Br 1/6I 5/6) 33. ~S2~

17 In previous work, we have observed that a slight molar excess of PbI 2 with respect to ([MA] + [FA]) in the precursor solutions from which the perovskite is deposited have been beneficial for device performance 4,5. In a previous work, the [Pb 2+ ] / ([MA] + [FA]) ratio for all precursor solutions was held at As the effect we intend to probe in these experiments is the importance of PbI 2, the FAI-concentration in the precursor solution was varied but [MABr] was held constant and equal to [PbBr 2]. This is a step away from the optimized protocol but makes the parameter window slightly cleaner for this particular investigation. The solutions were prepared in a glovebox with nitrogen atmosphere. Stock solutions of PbI 2 and PbBr 2 were prepared in advance whereas the final precursor solutions were prepared just before perovskite deposition. The solvent used for the perovskite solutions was a mixture of anhydrous dimethyl formamide, DMF, and anhydrous dimethyl sulfoxide, DMSO in the proportion 4:1 by volume. Three master solutions were prepared; PbI 2 and CH 3NH 3I in DMF/DMSO with a stoichiometric excess of PbI 2, PbI 2 and CH 3NH 3I in DMF/DMSO with a stoichiometric deficiency of PbI 2, and PbBr 2 and CH(NH 2) 2Br in DMF/DMSO. These three solutions were mixed in the right proportions to get the final precursor solutions from which the perovskites were deposited. All the final precursor solutions had the following concentrations: [Pb2+] = 1.25 M, [PbBr 2] = 0.22 M, [PbI 2] = 1.04 M, [MABr] = [PbBr 2] = 0.22 M. The concentration that was varied in the precursor solutions was [FAI]. When we henceforth state that a composition is +10 % with respect to PbI 2, we mean that [FAI] = [PbI 2] 0.9, and when we state a composition as -10 % we mean that [FAI] = [PbI 2] 1.1. Plus 10% thus mean a surplus of PbI 2 in the solution (or a deficiency of FA), and that should result in some excess PbI 2 in the films. There could in principle be PbBr 2 in the film as well, but as PbBr 2 more easily form the perovskite than PbI 2 that is unlikely 3. A secondary effect of this procedure is that the film thickness may vary a bit as a function of stoichiometry, but that should not be a mayor effect and can be compensated for. The MA and FA salts were bought from Dyesol and the lead salts were bought from TCI. All chemicals were used as received without further treatment. The perovskites were spin-coated in a glove box with a flowing nitrogen atmosphere with a fairly high flow in order to ventilate out solvent vapors. For each sample, 35 µl of the precursor solution was spread over the substrate, which thereafter was spin-coated using a two-step program. The first step was a spreading step using a rotation speed of 1000 with an acceleration of 200 rpm/s rpm for 10 s. That step was immediately, without pause, followed by the second step where the films were spun at 4000 rpm for 30 s using an acceleration of 2000 rmp/s. During the second step, when approximately 15 seconds of the program remains, 100 µl of anhydrous chlorobenzene was applied on the spinning film with a hand held automatic pipette. This last step, known as the antisolvent method, has a large impact on film morphology and result in significantly better device performance 6-8. It is, however, one of the steps that introduce a palpable degree of artisanship into the process. Directly after spin-coating, the films were placed on a hotplate at 100 C where they were annealed for min. At this temperature, the transformation into the perovskite was visually seen to occur within the time frame of a minute. After the heat treatment, the samples were cooled to ambient temperature where after the solid state hole-conductor was spin-coated on top of the films. A 70 mm solution of Spiro-MeOTAD (spiro) dissolved in chlorobenzene was used as a hole conductor. To improve the performance of the spiro, three different additives were added 9,10 : 4-tert-butylpyridine, 1.8 M Li-TFSI in acetonitrile, and 0.25 M Co[t-BuPyPz] 3[TFSI] 3, also known as FK209, in acetonitrile. The Spiro:FK209:Li-TFSI:TBP molar ratio was 1:0.05:0.5:3.3. The spiro solution was prepared the same day as the perovskite films were deposited. The spiro was deposited by spin-coating at 4000 rpm for 20 s. 50 µl of the solution was deposited on the spinning film, using a hand held automatic pipet, a few seconds into the ~S3~

18 spinning program. The samples were stored in a desiccator pumped under vacuum for a day and the back contact was then deposited. Before the back contact was deposited, the perovskite/spiro layer was removed from one end of the samples using a razorblade, acetonitrile, and a cotton bud in order to ensure contact between the FTO and the gold contact. The front and back contact were composed of an 80 nm thick gold film deposited by physical vapor deposition at a pressure of around Torr using an evaporator from Leica, EM MED020. Characterization UV-vis absorption measurements were performed on an Ocean Optics spectrophotometer HR c with deuterium and halogen lamps. In all measurements, a full spectrum from 190 to 1100 nm with 2048 evenly distributed points was sampled. 100 consecutive spectra were averaged in order to obtain good statistics. Steady state photoluminescence was measured with a Fluorolog, Horiba Jobon Yvon, FL A white tungsten lamp was used as luminous source, A monochromator was placed between the sample and the light source as well as between the sample and the detector. An excitation wavelength of 435 nm was used for all samples. The excitation spectrum was measured from 455 nm to 835 nm in steps of one nm. An integration time of 0.5 s was used for each wavelength. Measurements were performed both on perovskites deposited on substrates with mesoporous TiO 2 and on films deposited on substrates with 15 nm amorphous SnO 2. The excitation source and the detector were placed in 90º with respect to each other. The sample was oriented 60 with respect to the excitation source in order to decrease interference form reflected light. XRD measurements were measured using a Brucker diffractometer using a Bragg-Brentano geometry. Cu kα radiation, with a wavelength of 1.54 Å, form a cupper target was used as X-ray source. 2θ scans between 10 and 65º were collected using a step size of 0.008º. SEM imaging was carried out using a Zeiss Merlin scanning electron microscope. Photographs were taken using a Canon EOS 450 D with an EFS 60 macro lens. The IV-characteristics of the devices were measured using a home built system. To simulate solar light, an Oriel solar simulator with a xenon arc lamp, fed with 450 W input power, was used together with a Schott K113 Tempax filter (Praäzisions Glas & Optik GmbH). The light intensity was calibrated with a silicon photodiode equipped with an IR-cutoff filter (KG3, Schott). The IVcurves were measured with a digital source meter (Keithley 2400). No equilibration time or light soaking was applied before the potential scan. The starting point for the measurements was chosen as the voltage where the cell provided approximately 2 ma in forward bias. From that point, the potential was scanned to short circuit and back again using a scan speed of 20 mv/s. Thereafter, the dark current was sampled using the same scan speed. The cells were masked with a metal mask in order to limit the active cell area to 0.16 cm 2. The scan speed was slow enough to give efficiency data for the backwards scan that are in reasonable agreement with maximum power point tracking measurements. IV-curves measured on high efficiency devices with this setup have recently been confirmed to be in good agreement with data provided from independent certification agencies. The external photocurrent efficiency, EQE, was measured on a subset of the samples. This was done by a home built system composed of a 300 W xenon lamp, a gemini-180 double monochromator and a lock in amplifier. A white bias light of 50 W/m 2 was provided by a LED array. The EQE for each wavelength was extracted by measuring the difference in short circuit current between the white bias light, and the white bias light together with a superimposed monochromatic light, and scaling the signal with the intensity of the monochromatic light. The monochromatic light was chopped at 2 Hz and to get reasonable statistics, a fairly long integration time was used. The EQE was measured in steps of 10 nm from 340 to 850 nm. The EQE- system has had some performance problems. Absorption onsets and relative intensities are trustworthy but too much significance should not be read into absolute values. ~S4~

19 Transient absorbance spectra were recorded using femtosecond pump-probe spectroscopy. The pump beam (λ ex = 480 nm, 46 fs FWHM) was obtained by pumping a two-stage Noncollinear Optical Parametric Amplifier (NOPA) with the output of a Ti:Sapphire laser (CPA- 2001, Clark, 778 nm, 120 fs, 1KHz repetition rate). The pump fluence at the sample was 35 μjcm - 2 (100 nj, 605 μm in diameter). The probe pulse was generated by directing a part of the 778 nmoutput into a CAF 2 crystal, yielding a white light continuum (WLC, 400 nm-780 nm). The probe fluence at the sample was much lower than the one of the pump. Similarly, its diameter was smaller than the one of the pump to ensure homogeneity of the probed area. The dynamics of the photoinduced signals were obtained with a computer-controlled delay-line on the pump path. The probe beam was split before the sample into a beam going through the sample (signal beam) and a reference beam. Both signal and reference pulses were directed to a pair of 163 mm spectrographs (Andor Technology, SR163) and detected pulse-to-pulse with 512x58 pixels backthinned CCD detectors (Hamamatsu S ). The pump beam was chopped at half the laser frequency (500Hz). Satisfying signal-to-noise ratio was obtained by averaging 3000 spectra. TEM images were taken by a Talos FEI microscope. Samples were prepared with two different compositions (-10 % and + 20 %). Those samples were deposited directly on soda lime glass. The perovskite film was scraped of using a razor blade and deposited on the TEM grid by rubbing it in the perovskite scrapes. Photoelectron spectroscopy measurements were performed at two different synchrotron facilities having two different energy ranges. The depth sensitivity in the PES measurements depends on the inelastic mean free path (IMFP) of the photoelectrons, which is related to their kinetic energy. Therefore, changing the photon energy will modify the depth sensitivity 11. The depth sensitivity values reported in the present work were defined as three times the IMFP of the photoelectron, since 95% of the PES signal in a homogeneous material comes from a layer with this thickness. Hard X-ray PES (HAXPES) was carried out at BESSY II (Helmholtz Zentrum Berlin, Germany) at the KMC-1 beamline 12 using the HIKE end-station 12. The end-station is provided with a usable photon energy range from 2 kev to 12 kev, the photon energy is selected using a double-crystal monochromator (Oxford-Danfysik) and the photoelectron kinetic energies (KE) were measured using a Model R4000 analyzer (Scienta) optimized for high kinetic energies. In this work, photon energies of 2100 ev and 4000 ev were used by selecting the first-order light from a Si(111) and Si(311) crystals, respectively. The pressure in the analysis chamber was 10 8 mbar. Soft X-ray Photoelectron spectroscopy (SOXPES) measurements were carried out at the Beamline I- 411 at the Swedish National Synchrotron Facility MaxIV Laboratory in Lund, Sweden 13. The end-station is in this case provided with a usable photon energy range from 50 ev to 1500 ev. The photon energies were selected using a modified Zeiss SX-700 monochromator, and the photoelectron kinetic energies (K.E.) were measured using a Scienta R4000 WAL analyzer. The pressure in the analysis chamber was 10 8 mbar. Overview spectra were measured with a pass energy (Ep) of 500 ev while 200 ev was used for core peaks and valence band spectra. The spectra presented in this work were energy calibrated versus the Fermi level at zero binding energy, which was determined by measuring a gold plate in electric contact with the sample and setting the Au 4f 7/2 core level peak to 84.0 ev after curve fitting. The perovskite related spectra were intensity calibrated vs. the Pb4f 7/2 core level peak if not stated otherwise. The peak positions and areas were optimized by a weighted least-squares fitting method using CasaXPS software. Finally, the quantification tables and intensity ratios presented between different core levels were calculated from the experimental results after correcting the intensity by the photoionization cross section for each element at their specific photon energy, using database values 14,15. Charging was controlled by following peak positions and peak shape with variations in light intensity and time. No charging was observed in the measurements of the different perovskite ~S5~

20 materials. The same procedure was also used to check for X-ray-induced effects. No changes were observed in the spectra reported here. After preparation of the perovskite materials, each sample was stored in a sealed box together with common desiccants to avoid any moisture contamination. The box was only opened prior the PES analysis. Confocal photoluminescence (PL) maps were acquired using a custom-built time-correlated single photon counting (TCSPC) confocal microscope (Nikon Eclipse Ti-E) setup with a 100X oil objective (Nikon CFI PlanApo Lambda, 1.45 NA). Samples were photoexcited through the glass-side using a 405 nm laser head (LDH-P-C-405, PicoQuant GmbH) with pulse duration of <90 ps, fluence of ~2 μj/cm 2 /pulse, and a repetition rate of 0.5 MHz. The photoluminescence from the sample was collected by the same objective and the resulting collimated beam passes through a long-pass filter with a cut-off at 416 nm (Semrock Inc., BLP01-405R-25) to remove any residual scattered or reflected excitation light. A single photon detecting avalanche photodiode (APD) (MPD PDM Series 50 mm) is used for the detection, with the APD output connected to a timing module with a resolution of 4 ps (PicoQuant PicoHarp 300), which detects the arrival time of each photon for the TCSPC measurements. The sample was scanned using a piezoelectric scanning stage. The measurements were acquired using the commercial software SymphoTime 64 (PicoQuant GmbH). For the measurements when exciting and detecting through the top-side, a 40X objective (Nikon PlanApo, 0.95 NA) was used. Electroluminescence measurements were performed using a Biologic SP300 Potentiostat as voltage source and current meter. The emitted photon flux was detected by measuring the shortcircuit current of a Hamamatsu photodiode (1 cm 2 ), connected to a second channel of the potentiostat while sweeping the voltage applied to the solar cell. Additional XRD data As discussed in the main article, XRD data was sampled for films of seven different stoichiometries. The full set of diffractograms are given in figure S.1. Apart from the perovskite phase, the only crystalline phase observed is PbI 2, which, as expected, increases in amount for the more over-stoichiometric samples. PbI 2 is detected also in the under-stoichiometric samples but in a much smaller amount. Crystalline PbBr 2 is not detected in any sample, which is in line with previous measurements where PbBr 2 was shown to have a stronger preference for forming the perovskite than PbI 2 3 Depending on the composition, the perovskite crystal phase can be either tetragonal or cubic at room temperature. The crystalline phase here observed is cubic for all samples, which most easily is seen as the absence of a diffraction peak at 23.5º and that the (022) peak around 40.5º is a single peak rather than the double peak expected for the tetragonal phase. Based on previous measurements 3,16 as well as on theoretical considerations 17, that is the expected result. The width of the diffraction peak which is wider for the under-stoichiometric samples. That indicates a smaller crystallite size. No clear trend is, however, observed. The under-stoichiometric samples have one wider peak, whereas the rest of the samples have one narrower peak. An estimation based on Scherrer s equation gives a crystallite size of around 30 nm for the understoichiometric samples. For the over-stoichiometric samples, the same approach gives more than 100 nm but for crystallite this large it is hard to get a good estimate, and due to the instrument broadening this is bound to be an underestimation ~S6~

21 Figure S.1. XRD data for a number of different stoichiometry. Starting from the upper left corner and going from left to right the compositions are: -10 %, -5 %, 0 %, 5 %, 10 %, 15 %, 30% according the notation established in the experimental section. Data is normalised with respect to the (100) reflection around 14.2ºC. Additional absorption data Absorption data measured in transmission mode for seven different stoichiometries were given in the main article. These data are here reproduced in figure S.2.a. The overall absorption behaviour was rather unaffected by the stoichiometry, with the exception of the most under-stoichiometric sample (the one with the greatest deficiency of the PbI 2). That sample has a more plateau-like absorption in a region just above the band gap. The background signal due to scattering is low, and below 0.1 for all samples which is a sign of uniform films of high quality with respect to macroscopic inhomogeneity. The value of the background, hear simply defined as the minimum absorption value, was removed in the subsequent data treatment and are given in figure S.2.c. Figure S.2. (a) Absorption as a function of wavelength. Data are background corrected. (b) Square of the absorption against photon energy for extracting band gap energies. (c) The background signal due to scattering that is subtracted from the raw data to give the absorption curves in (a) The colour scheme is the same as in (a) and (b). ~S7~

22 The absorption data in figure S.2.a indicates that films with an excess of PbI 2 absorb stronger than the films with less PbI 2. Cross section SEM images presented in figure S.4 indicate that the films are of essentially the same thickness, and complementary measurements using an integrating sphere indicate that it may be differences in reflectivity that are behind the apparent absorption difference in figure S.2.a. That could potentially be an effect of difference in surface roughness observed in the SEM images in figure S.4 and S.5. Absorption measurements measured in reflection mode using an integrating sphere was measured for a few samples as illustrated in figure S.3. Those data indicate that the transmission measurements were overestimating the difference in absorption strength, and that the absorption coefficient not are strongly affected by the stoichiometry. A lower overall absorption, be it due to lower absorption itself or higher reflectivity, may correlate with a lower photocurrent for the under-stoichiometric samples as discussed in the main article. The rise in absorption for the -5% sample is less steep than for the other samples. That could possibly relate to the increased FWHM seen for the 001 diffraction peak (figure 2.b). The -10 % sample which has the same FWHM does, however, not show the same absorption behaviour. In figure S.2.b, the square of the absorption is plotted against photon energy. By doing that, a linear region is found for photon energies slightly above the band gap energy. If that linear region is extrapolated, the band gap is given as the intercept with the base line. The extracted band gap values are given in table S.1. They are centered around 1.64 ev. There is a tendency for the band gap to increase for the samples with more PbI 2 but the effect is small; eV and less than 0.01 ev if the sample with the smallest amount of PbI 2 is excluded. As the band gap is strongly dependent of the Br/I ratio this could indicate a slight compositional change. The effect is too small to be of real significance and both the absorption measurements and the XRD-data indicate that the crystalline perovskite phase is not strongly affected by the amount of PbI 2 in the system, even though the crystal grain size may differ. Figure S.3. Absorbance as a function of wavelength for samples of four different stoichiometries. The measurements were performed in reflection mode using an integrating sphere. Table S.1. Extracted band gap as a function of composition Sample Stock E g [ev] number 1-10% % % % % % 1.65 ~S8~

23 Additional SEM data Cross section SEM-images of complete cells are displayed in figure S.4, which show that the film thickness is rather unaffected by the stoichiometry. The over-stoichiometric samples may be slightly thinner, which would be expected as the amount of perovskite that could form from a given volume of stock solution is somewhat smaller for the over-stoichiometric samples given the synthesis protocol used. The difference is, however, small. Top view SEM images of a sample with -10% and + 20% stoichiometry was given in the main article. In figure S.5, the corresponding figures for a broader range of stoichiometries are given. Larger version, and images of different magnification are given in the end of the supporting information. Figure S.4. Cross section SEM images of cells with seven different stoichiometry. The most understoichiometric sample is the left one (a) and the most over-stoichiometric samples with most PbI 2 is to the right (g). Depending of the quality of the edge of the brake the individual grains are more or less clearly seen. The top most gold contact is only seen in (b), (c), (d) and (g), and in (a), (e) and (f) the best view was found slightly outside the gold contact region. The scale bare is 200 nm. Figure S.5. Top view SEM-images. From the upper left corner: -10%, -5%, 0, +5%, +10%, +15% and +20%. The scale bars are 200 nm. Additional photoluminescence data An interesting observation is that the PL-intensity is distinctly higher for the sample with a deficiency of PbI 2. The higher PL-intensity for the samples with less PbI 2 is a solid observation, but the magnitude of the difference is uncertain as the setup does not give absolute numbers which also have been observed to vary depending on sample and time. This effect was seen for most batches but not for every single one which is an indication that competing mechanisms are involved in determining the emissivity of the samples and that it to some extent is batch dependent. On the day of synthesis, the difference in intensity was approximately one order of magnitude, as seen in figure S.6.a. After a month in dry air, this difference remains, but it gets somewhat smaller as seen in figure S.6.b. In figure S.6.a and S.6.b there appears to be a difference in absolute intensity, which could be the case, but as the setup used in the measurements not give absolute numbers that is not a solid conclusion. The corresponding normalised data is given in figure S.6.c-d. The peak positions are given in table S.2. No systematic shift is observed in the peak positions, neither with respect to storage time or stoichiometry. This indicates that the ~S9~

24 perovskite composition is rather unaffected by the overall stoichiometry in the precursor solutions. Figure S.6. (a) PL-data measured the same day as the films were deposited. (b) PL-data measured after one moth of storage in dry air. (c) Normalised data measured on day 1. (d) Normalised data measured after one moth of storage in dry air. Figure S.7. PL data under 10 sun illumination, both directly while illuminated and after a few minutes of constant illumination. ~S10~

25 In figure S.7 PL data were sampled both directly wile illuminated and after a few minutes of constant illumination (10 sun). For the over-stoichiometric samples there appears to be an increase in the intensity, and it also appears that a phase separation is occurring. For the stoichiometric sample and for the under-stoichiometric sample the effect is less evident. Table S.2 Peak positions for the photoluminescence Sample Stock E g [ev] Peak position [nm] number Day 1 Day % % % % % % Additional photoelectron spectroscopy data The C1s spectra are given in figure S.8. The measurement at 4000 ev gives similar conclusion whereas the measurement at 758eV mainly detected the surface contamination species mentioned earlier. In figure S.9 the relative percentage of Pb, I and Br in the mixed perovskite with three different stoichiometry as function of the photon energy. The band gaps extracted earlier in this work were insensitive to the stoichiometry and the amount of excess PbI 2. This statement is confirmed by valence band spectra presented in figure 10. The spectra of the perovskite materials with composition -10%, 0 and +10% perfectly overlapped each other and a closer look at the valence band edge do not reveal any difference greater than the resolution of the technique which is ±0.05 ev. Figure S8: C1s core level peaks of the perovskite materials with three different stoiechiometries: -10%, 0, +10% in light grey, grey and dark grey respectively recorded at 4000 ev (left) and 758 ev (right). The C1s spectra of the SnO 2 substrate are shown as a reference to show the influence of the surface contamination on these spectra and how it mainly contributes to the signal when a low photon energy is used, i.e. a great surface sensitivity is probed. ~S11~

26 Figure S.9: Relative percentage of Pb, I and Br in the mixed perovskite with three different stoichiometry as function of the photon energy. +10% 0% -10% Valence band Binding Energy (ev) Figure S.10. Valence band spectra of the perovskite materials with three different stoichiometry: -10%, 0, +10% in light gray, gray and dark gray respectively recorded with a photon energy of 2100 ev. An expanded view of the valence band edge is presented on the right. Additional device data Making an efficient device is tricky, involves a fair bit of artisanship, and is affected by a number of environmental parameters that are challenging to control and at the moment not properly understood. This is the reason for the cell-to-cell and a batch-to-batch variation of device performance observed by many groups in the field, as well as in our data. The full set of JVcurves discussed in the article are given in figure S.11. The corresponding device data are given in table S.3 The absorption data in figure 2.c in the main article indicate that the under-stoichiometric samples have a lower absorbance which could explain a part of the drop in photocurrent. The magnitude of the current drop, together with the cross section SEM-images in figure S.4 and complementary absorption measurements using an integrating sphere, however, indicate that it cannot be ascribed to differences in film thickness and optical absorption. If obstruction of ion motion is the reason behind the decrease in the hysteresis for the understoichiometric samples, grain size could be a contributing factor. As discussed above, the understoichiometric samples have more polycrystalline grains, or grains with more stacking defects, which could contribute to decreasing the ion movement, and thus also the hysteresis. A relation between smaller grains and a lower hysteresis does, however, not translate into overstoichiometric compositions where the hysteresis has been observed to increase with decreased grain size 18. ~S12~

27 Figure S.11. The full set of JV-curves ~S13~

28 Table S.3. Device data extracted from the IV curves in figure S.1. An * marks data for the forward scan. The other data is for the reversed scan. Stock [%] Cell V oc [V] J sc [ma/cm²] FF η [%] V oc* [V] J sc* [ma/cm²] FF* η* [%] H -10 C C C C C C C C C C C C C C C C C C C C C C Electroluminescence Electroluminescence measurements were performed on the devices to verify whether the high open-circuit voltage correlates with high luminescence efficiencies as expected from fundamental theory. This was indeed the case. There was, however, no clear trend with respect to the PbI 2 excess. The samples showed strong variations from measurement to measurement, possibly due to a reversible instability related to hysteresis and introduced by degradation as the devices have been stored for some time before the measurements. The highest value of the external electroluminescence quantum efficiency (EQE EL) in this series was measured for a -5 % device and approached 0.8 % at a current of 20 ma/cm 2. This is in good agreement with a V oc of 1.2 V 5,19 and is as far as we know a record value. The EQE EL shown in figure 17.f increases approximately linearly with driving current as expected for a trap-recombination limited current 20. The hysteresis in the EQE EL indicates that defect formation as a function of applied voltage influences the radiative recombination yield. In this case, the yield is larger when scanning from high bias back to 0 V and then also closer to the linear relation, whereas it is lower but superlinear when scanning forwards from 0 V. This is due to enhanced defect recombination, where the effect of the defects (their number, their cross section, their occupation) is reduced in situ during the sweep, thus resulting in the superlinear relation. Figure S.12 EQE EL as a function of applied current for the -5% sample, scanned from 0 V to 1.5 V and back with a voltage sweep rate of 20 mv/s ~S14~

29 Additional TEM images Figure S.13. (a)-(c). TEM images of a PbI 2-deficient sample. (d)-(i). TEM images of a PbI 2-rich sample. (g) and (h) is an image of the same spot illustrating sample degradation under the electron beam. (i) The same part as in (h) together with a previously unexposed grain. The samples were prepared by scratching of a perovskite film deposited on SLG with a razor blade, in which the TEM-grid then was rubbed. ~S15~

30 Additional SEM images Top view SEM images taken at six different magnifications are given in figure S A larger version of the cross section images in the main article is given in figure S.20. Figure S.14. SEM images of samples with seven different stoichiometries. Starting from the upper left corner and going from left to right the compositions are: -10 %, -5 %, 0 %, 5 %, 10 %, 15 %, 20% according the notation established in the experimental section. The width of each panel is 1.7 µm. ~S16~

31 Figure S.15. The same samples as in figure S.7 but at a lower magnification. The width of each panel is 2.5 µm. Figure S.16. The same samples as in figure S.7 but at a lower magnification. The width of each panel is 3.3 µm. ~S17~

32 Figure S.17. The same samples as in figure S.7 but at a lower magnification. The width of each panel is 5.0 µm. Figure S.18. The same samples as in figure S.7 but at a lower magnification. The width of each panel is 10 µm. ~S18~

33 Figure S.19. The same samples as in figure S.7 but at a lower magnification. The width of each panel is 25 µm. ~S19~

34 Figure S.20. A larger version of the cross section images found in the main article. Starting from the upper left corner and going from left to right the compositions are: -10 %, -5 %, 0 %, 5 %, 10 %, 15 %, 20% ~S20~