Photodecomposition and thermal decomposition in methylammonium halide lead perovskites and inferred design principles to increase photovoltaic device stability

Hybrid lead halide perovskites have emerged as promising active materials for photovoltaic cells. Although superb efficiencies have been achieved, it is widely recognized that long-term stability is a key challenge intimately determining the future development of perovskite-based photovoltaic technology. Herein, we present reversible and irreversible photodecomposition reactions of methylammonium lead iodide (MAPbI3). Simulated sunlight irradiation and temperature (40–80 C) corresponding to solar cell working conditions lead to three degradation pathways: (1) CH3NH2 + HI (identified as the reversible path), (2) NH3 + CH3I (the irreversible or detrimental path), and (3) a reversible Pb(0) + I2(g) photodecomposition reaction. If only the reversible reactions (1) and (3) take place and reaction (2) can be avoided, encapsulated MAPbI3 can be regenerated during the off-illumination timeframe. Therefore, to further improve operational stability in hybrid perovskite solar cells, detailed understanding of how to mitigate photodegradation and thermal degradation processes is necessary. First, encapsulation of the device is necessary not only to avoid contact of the perovskite with ambient air, but also to prevent leakage of volatile products released from the perovskite. Second, careful selection of the organic cations in the compositional formula of the perovskite is necessary to avoid irreversible reactions. Third, selective contacts must be as chemically inert as possible toward the volatile released products. Finally, hybrid halide perovskite materials are speculated to undergo a dynamic formation and decomposition process; this can gradually decrease the crystalline grain size of the perovskite with time; therefore, efforts to deposit highly crystalline perovskites with large crystal sizes may fail to increase the long-term stability of photovoltaic devices.


Powdered Samples
Lead (II) iodide (PbI2, 99.9%) was purchased from Tokyo Chemical Industry Co., Ltd, lead (II) bromide (PbBr2, 99.999%) was purchased from Sigma-Aldrich, methylammonium iodide (MAI) and methylammonium bromide (MABr) were purchased from Dyesol Limited. All chemicals were used as received without any further purification. Hybrid perovskites in powdered polycrystalline material form were obtained mimicking the procedure to deposit perovskite thin films on substrates. Briefly, 1 mL of DMF (Wako Pure Chemical Industries) solution (~1 M) containing desired stoichiometric precursor quantities to synthesize MAPbBr3 or MAPbI3 was poured on a 10 cm in diameter mortar kept at 100 °C inside of fume hood. Precursor solution is slowly spread on the mortar surface helped by the pestle. DMF solvent evaporates in 1-2 minutes remaining a solid crystalline material on the mortar that is carefully collected. Perovskite phase purity proof, absence of PbX2 peaks and crystalline parameters are checked from powder XRD measurements, see Figure S1.

Thin-film samples for XRD and XPS degradation studies
FTO glasses (14 Ω/sq, Opvtech.) were cleaned by sequential sonication in deionized water, ethanol, acetone, and UV-Ozone treated before use. PbI2, PbBr2, MAPbI3 and MAPbBr3 precursor DMF solutions (~1 M) were spin-coated on the substrates and annealed at 100 °C for 5 min. XPS and XRD analysis of these thin-films is reported in Sections 8 and 9, respectively.

Powdered Samples
Typically, fresh samples of PbI2 (~180 mg) are loaded in the sample holder located inside the chamber, Figure S2a. Once high vacuum level is reached (<10 -6 Torr), the pressure gauge and MS spectrometer are switched on. The temperature of the sample under dark conditions and high vacuum conditions was observed to be slightly high (30-35 °C) due to unavoidable e-ionization (radiative heating) of MS. Light power pulses of white, red, and blue LEDs are programmed using an Autolab PGSTAT204 potentiostat including the LEDs and the LED driver box accessory (Metrohm AG). Simulated sun-light is obtained using a 150 W short-arc Xe lamp from Solar Simulator (PEC-L01, Peccell Technologies Inc.). Light pulses in the solar simulator were computer controlled remotely by a homemade program and actuator. Both light power delivered by solar simulator and LEDs were calibrated using a calibrated silicon photodiode accounting quartz window and distance between light source and sample holder (see section S5 for light power calibration details). Mass spectrometer traces were recorded using a quadrupole mass spectrometer (1-300 amu) equipped with an electron multiplier detector (RGA300, SRS Stanford Research Systems). Conventional Faraday cup detector in this mass spectrometer was not reliable to detect diiodine traces at nearly room temperature. MS raw signals were calibrated using sensitivity factors calculated following the procedure explained in section S6.

Thin-film Samples
Thin-film samples (see section S1 for sample preparation) were loaded in the same vacuum chamber described in Figure S2a where the MS experiments were conducted. A Portable Solar Simulator (PEC-L01) was mounted on top of the vacuum chamber allowing the light exposure on the samples through a quartz glass view port. The samples were exposed to solar simulator light (~0.55 Sun) at vacuum of ~10 -6 Torr during 4 h. Thin films were analyzed by XPS and XRD before and after degradation experiments, see sections S7-9.
S3. PbI2 degradation under illumination/dark conditions. Powdered samples and thin-films were prepared according to section S1. Degradation procedures follows section S2. Section S5 provides further details of light sources power calibration and emission wavelengths. Section S6 provides further details of relevant gas species sensitivity factors determinated using the electron photomultiplier detector in the MS equipment.

Control experiment of PbI2 decomposition as a function of LED power.
Programmed sequential increase in light power is applied to the sample with a total of 12 sequential pulses for each of white, red, and blue LED light source, Figure S2c-e. The relationship between maximum partial pressure measured vs. photon flux applied in each pulse is shown in Figure S2b for white and blue LEDs, respectively. It can be observed that diiodine MS signal exceeds the detection threshold during white LED illumination at Pulse no. 8 (~2.2 10 17 photon s -1 cm -2 ), but it exceeds the threshold during blue LED illumination at a lower photon flux at Pulse no. 5 with ~1.4 10 17 photon s -1 cm -2 blue photons. This different behavior can be explained based on the fact that white LED light illumination consists of a mixture of photons with two wavelengths (450nm and 550 nm) and 550 nm photons in white LED cannot produce I2. As photon flux on the sample is sequentially increased, the rate of diiodine release is observed to increase for blue and white LED illumination ( Figure S2b). Main difference between blue and white LEDs illumination pulses experiment is that the I2 release rate vs. applied photon trace is displaced to the right for white LED experiments. This is explained by the fact that white LED flux contains less quantity of high energy photons above the PbI2 band gap that are able to generate I2.
On the other hand, the monoiodine MS signal can be detected at lower photon flux than the diiodine MS signal. Monoiodine can be detected already at Pulse no. 1 and 4 for blue and white LED, respectively ( Figure S2b). The most interesting distinctive feature observed comparing monoiodine and diiodine signals is that the two monoiodine MS signals using blue and white LED converges asymptotically as a higher photon flux is applied ( Figure S2b). In Figure S2b, the expected quantity of monoiodine fragmentation inside MS during white LED experiment is indicated by a red line. However, monoiodine observed experimentally is higher than expected and it could be a signature of thermal evaporation of PbI2. Note that monoiodine (127 amu) detected by MS can be ascribed to the fragmentation pattern of I2 during the MS measurement. In fact, according to the NIST database, 1 the fragmentation pattern for I2 corresponds to a molecular peak of 254 amu with 100% intensity followed by a peak for the monoiodine fragment (127 amu) with a peak intensity of 52.5%. Figure S13a describes the fragmentation pattern of I2. Section S6 describes the diiodine/monoiodine calibration procedure for these MS measurements using this electron multiplier detector.

Control experiment of PbI2 decomposition in dark as a function of sample temperature
To decouple the effect of temperature during illumination tests, an experiment using artificial sun light pulses from a Xe lamp from a solar simulator and heating interval events in dark conditions is carried out. This experiment consists of two parts, 1) pulses of 3 min of duration using constant power Xe lamp illumination (55.2 mW cm -2 ) alternating with 1 min duration rest intervals in dark conditions and 2) long heating interval in dark conditions until the sample reaches high temperatures close to that obtained with light pulses, see Figure S2f-h.
As seen before during white and blue LED light illumination experiments ( Figure S2c-e), there is an appreciable level of I2 gas release only under Xe lamp illumination conditions. On the other hand, during the sample heating in dark conditions, no release of I2 gas is observed. However, the monoiodine trace can still be observed when the sample is heated above 70 °C. In view of these results, it is concluded that the release of I2 during controlled degradation of PbI2 only takes place during light irradiation using photons with energy greater than the PbI2 band gap (2.34 eV, 530 nm). 2 On the other hand, under dark and heating conditions, the most plausible origin of monoiodine excess detected would be resulted from thermally evaporated under high vacuum conditions PbI2 molecular entities or PbI2 clusters being monoiodine a fragment observed by MS. Under dark conditions and moderate temperatures (~70 o C) in high vacuum conditions (~10 -6 Torr), PbI2 evaporation takes place: This process described by Eq. 1 is followed by electron ionization inside MS inducing subsequent PbI2(g) molecules or clusters fragmentation: Therefore under dark and heating conditions, only monoiodine fragment is detected by the MS equipment.
[PbI] mass fragment can not be observed because it is out of the detection range of the MS equipment (1-300 amu). quadrupole mass spectrometer, top quartz window and Xe lamp or LED light sources with controlled on/off intervals. All components were housed inside a vacuum chamber with a base pressure of 5x10 -8 Torr. A pin hole (2 mm diameter) in an aluminum foil shield separated the sample holder and the ionizer filament in the MS avoiding direct illumination of the sample by the filament. Experiments were performed initially at ~35°C. Experimental setup scheme for calibration purposes differed slightly, see Figure S10. b) Experimental partial pressure peak maxima measured for di-and monoiodine MS traces vs. photon flux in each interval for white (open square white symbol) and blued LEDs (solid square blue symbol). Numbered labels correspond to the pulse numbers shown in panels (d) and (e). Extracted points correspond to only those where the partial pressures in panels (d) and (e) are above background. Continuous red line represents the hypothetically expected quantity of monodione observed by fragmentation inside the MS using white LED. c) Temperature recorded in the sample holder in the dark (grey area) and under illumination using three different LEDs (white, red, or blue) with increasing light power during 12 pulses of 60 sec of illumination followed by 60 sec of darkness. MS signals are calibrated with m/z values corresponding to partial pressure recorded simultaneously by MS for d) mono-and e) diiodine. Power delivered for each LED in each light pulse can be found in Section S5. f) Temperature recorded in the sample holder under dark (grey area) and illumination conditions (white area) using Xe lamp illumination pulses (3 min illumination and 1 min dark, 4 cycles) at constant light power of 55.2 mW cm -2 . After ~3000 s, heating in dark is carried out using an heating element mounted on the sample holder. g) mono-and h) diiodine calibrated m/z traces. i) Plane view (1, -2, 0) of a layered PbI2 supercell in P-3m1 trigonal space group. 3 Each (PbI2)n layer showed in the figure depicts a step during the release of I2 and generation of Pb (0) by the two exciton mechanism. 4 Step 1, left layer) Photons with higher energy than PbI2 band gap (2.34 eV, 530 nm) arrive to the PbI2 surface forming excitons. Two closely generated excitons can be separated if holes oxidize iodide anion and electrons reduce Pb 2+ center.
Step 2, middle layer) Generated iodine atoms are close enough to form I2.
Step 3, right layer) I2 molecule is released from PbI2 surface leaving two anion vacancies.

Determination of non-volatile products of photodecomposition on thin-films of PbI2 using XRD and XPS analysis.
See section S7. Determination of non-volatile products of photodecomposition on thin-films of PbI2, MAPbI3 and MAPbBr3 on FTO substrate samples using XRD and XPS analysis.
Recovery of PbI2 thin-film from degraded Pb 0 containing films using I2.
Degraded PbI2 thin-film containing Pb 0 can be recovered back after exposure to I2 gas in a closed glass bottle at room temperature and dark conditions. After few seconds it can be observed that the gray material observed in degraded film disappears forming again an orange-yellowish film on the top of the substrate. This I2 exposure is hold during 4 h to counteract the 4h duration photodecomposition procedure. XPS measurement are carried out to check the remaining Pb 0 content in the PbI2 sample, see Figure S3, where it can be observed that there is no Pb 0 traces remaining in the sample. The I/Pb ratio in the restored film is 1.62, which is close to that of pristine PbI2 thin-film (I/Pb ~ 1.8, section S7) considering experimental uncertainties. Similar output can be observed by XRD diffraction where it can be observed that the ~31° 2θ diffraction peak belonging to cubic Pb 0 phase have completely disappeared, Figure S4a. A comparison of the (001) diffraction peak in PbI2 phase in pristine PbI2 thin-film and recovered PbI2 thinfilm after I2 exposure is shown in Figure S4b. According to the Rietveld refinement, recovered PbI2 has larger crystalline domains (96 ± 6 nm) than pristine PbI2 film (71 ± 3 nm). On the other hand, microstrain is larger in recovered film (0.0046 ± 0.0003 %) than in pristine film (0.0027 ± 0.0003 %).
Recovery of MAPbI3 perovskite from degraded samples containing PbI2 by using CH3NH2 is not demonstrated experimentally here because there is enough literature supporting that MAPbI3 can be synthesized using PbI2 thin-films and CH3NH2/HI gases sequentially or simultaneously not needing explicit synthesis of the methylammonium iodide salt. 5

S4. Activation Energy calculation from MS traces
Activation energy for the reaction releasing diiodine is calculated using a modified constant initial rate method over all pulses generated during illumination runs. The reaction releasing diiodine from PbI2 is The dependence of the reaction rate on the reactant concentration is isolated considering a large excess of PbI2 surface ready to react at initial time so that their concentration remains essentially constant. Furthermore, only the raising temperature branch side of the pulse is considered for calculations. It simplifies the rate law because this constant and large excess concentration may be combined with the rate constant yielding a single effective rate constant as Here C is a lumped constant including pre-exponential kinetic term and constant initial concentration, Ea is activation energy and RT is the gas constant and temperature. Substituting (4) in eq. (3) and taking natural logarithm in both sides, a linearized form can be obtained for eq. 3, Equation 5 can be used to obtain graphically the slope Ea/R from all experimental pulses recorded using MS mass traces as partial pressures of I2, see Figure S5. Figure S5. Natural logarithm of the time derivative of the I2 partial pressure measured by MS during blue LED pulses vs. inverse of the temperature (black dots). All data points from pulses are drawn but only midearly points for each pulse obey the constant and large excess reagent concentration assumption. A straight and common negative slope can be easily visualized and the slope is obtained graphically by freehand using the frontier data. Later or sooner points do not obey the approximation and deviates from eq. 3 being them not considered for the slope. Note that this method is still valid even using non-calibrated MS data traces.

S5. Light power calibration
Light power delivered by solar simulator and LEDs were calibrated using a calibrated silicon photodiode (FDS100, Thorlabs, Inc) with spectral response from 350 nm to over 1100 nm. Calibration measurement accounts quartz viewport and distance from light source to sample holder. In principle the 2 mm thickness quart window shows > 90% of transmittance in the visible region, see Figure S6. Figure S6. Transmittance spectrum from the quartz window used in the experimental setup.
Light emission wavelengths for LEDs and solar simulator used in this work were measured by a Konica Minolta CS-2000 spectroradiometer. Increasing LED power light pulses were measured using the silicon photodiode ( Figure S7a) and photon flux ( Figure S7b) was calculated taking into consideration the before measured light emission wavelengths for LED ( Figure S8) and solar simulator ( Figure S9).

S6. Mass spectrometer detector calibration
Calibration of the quadrupole Mass Spectrometer using an electron multiplier detector (SRS Stanford Research Systems, RGA300) is carried out for partial pressure analysis of CH3I, NH3CH2 and I2 (diiodine) using direct comparison of the partial pressure analyzer output with a transfer standard pressure gauge. 6 Figure S2a shows the setup used for measurement and Figure S10 the modified setup for calibration purposes.

Iodomethane and methylamine sensitivity factors
Iodomethane (CH3I, reagent grade) was purchased from Wako Chemical Ltd., methylamine (NH2CH3) 40% water solution was purchased from Tokyo Chemical Industry Co., Ltd. Both reagents were used as received.
A few milliliters liquid sample of CH3I or CH3NH2 is loaded at atmospheric pressure inside the sample container which is connected to the leak valve. Vapor pressure at room temperature for pure CH3I liquid is obtained from NIST webpage and the vapor pressure of the NH2CH3 water solution from Romero et al work. 7 Because the small aperture delivered by a precision variable leak valve (MDC Vac. Prods., LLC.), the pressure and temperature in the glass vessel is considered constant through the experiment. Then, five different valve aperture positions are maintained during ~1 minute meanwhile partial pressure is recorded for CH3I (142 m/z peak) and CH3NH2 (31 m/z peak) using the electron multiplier detector and total pressure measured by the crystal/cold cathode pressure gauge (model CC-10, Tokyo Instruments, P symbol in Figure  S10). Pressure increments vs partial pressure increments from MS are plotted and linear fitting is carried out to obtain the calibration factor, see Figure S11. There is a linear relation between the partial pressure and the corresponding MS signals of the gases. Deviations from linearity in sensitivity factors (A) calculations are to be expected above 10 -5 Torr due to space charge effects in the ionizer and ion-neutral scattering interactions in the filter. MS sensitivity involves measuring the MS signals over several orders of magnitude of partial pressure to determine the range over which a linear relationship exists. The sensitivity factor for CH3I and CH3NH2 is calculated as the slope of the MS signal vs. partial pressure measure by external gauge over the linear range.

Monoiodine/ Diiodine sensitivity factors
Sensitivity factor calculation for the diiodine/monoiodine pair follows a procedure slightly different from above procedures for CH3I and CH3NH2 because of the high vapor pressure of solid iodine and fast recrystallization of iodine vapors in contact with cold surfaces. Such fast recrystallization does not permit to use the external glass container. High vapor pressure of solid iodine impedes direct sample holder loading and reaching simultaneously the working pressure for the MS equipment. Iodine pellets (I2, reagent grade) was purchased from Wako Chemical Ltd and used without further purification process. A few milligrams of iodine pellets were encapsulated in glass tube leaving a small orifice (0.1 mm or less) on the top of the glass capsule. The glass capsule is loaded in the sample holder equipped with heating element inside the vacuum chamber. Vacuum pump is started and the chamber reach an equilibrium pressure of ~ 1  10 -4 Torr at room T as measured from the external pressure gauge. In such a moment, the heating element is powered up accelerating the evaporation of solid I2. Pressure raises until reach a maximum followed by a pressure decay indicating the material is exhausted. MS equipment and detector are switched on at the moment that a safe pressure for the MS operation is reached. Decay in partial pressure signals for diiodine and monoiodine together with decay in total external pressure are measured and sensitivity factors are calculated from these pressure decays. An educated guess implies that the total measured pressure belongs to I2 gas and sensitivity factor is calculated taking the ratio of the time derivative of the MS signal and total pressures traces. Monoiodine sensitivity factor is obtained considering that diiodine fragmentation pattern from NIST database indicates a 52.5 % MS intensity signal for monoiodine compared to the 100% intensity for I2 molecular MS peak ( Figure S13a). Sensitivity factors calculated for di-, and monoiodine from pressure decays are shown in Figure S12.

Selected MS fragmentation pattern for molecules in this work
Fragmentation patterns m/z peaks retrieved from NIST MS library data 1 for compounds mentioned in the main manuscript are shown in Figure S13.  S7. Summary results on determination of non-volatile products of photodecomposition on thin-films of PbI2, MAPbI3 and MAPbBr3 on FTO substrate samples using XRD and XPS analysis.
Photodecomposition tests were carried out in ~200 mg of powdered samples to obtain a measurable quantity of volatile compounds to be detected by MS. Because photodecomposition is mainly produced at the sample surface, it is extremely difficult to check what non-volatile products from photodecomposition remains after degradation procedure on the top of such powdered samples. Therefore, thin-films of PbI2, MAPbI3 and MAPbBr3 on FTO substrates were prepared to check the remaining photodecomposition products after a 4 h under Xe lamp illumination test (0.55 Sun) inside the same vacuum chamber used for above MS experiments. Chemical compositions of such thin-films were analyzed before and after illumination test using X-ray photoelectron spectroscopy (XPS, Section S8) and X-ray diffraction (XRD, Section 9) measuresments, see Figure S15. Sections S8 and S9 provides further details on XPS and XRD measurements and analysis.
XPS and XRD measurements clearly confirm formation of metallic Pb 0 for all PbI2, MAPbI3 and MAPbBr3 on FTO substrate samples after photodecomposition, see Figure S15. It was observed that the freshly prepared PbI2 film showed a characteristic yellow color. However, it gradually turned to a dark grey color after light exposure. Concurrent to Pb 0 formation, XPS measurements showed a reduction of iodine content in the thin-film sample, which is consistent with the releases of I2 based on MS measurements. The atomic ratio of Pb 2+ : Pb 0 : I of the fresh PbI2 film was 1 : 0 : 1.8, which is in good agreement with expected nominal ratio (1 : 0 : 2); whereas the ratio for the degraded PbI2 film became 0.75 : 0.25 : 1.0. Such significant increase in Pb 0 and decrease in I content after light illumination indicate that PbI2 is decomposed to metallic-Pb accompanied by the release of I2 (see section S8 for experimental details and XPS results). MAPbI3 and MAPbBr3 samples also confirmed Pb 0 formation, but in these cases, it was also accompanied by PbI2 and PbBr2 formation how later observed by XRD, see Table S1 and Figure S15. XPS is a surface sensitive technique, and typically measures a few nanometers of the thin-film surface. On the other hand, XRD measurements carried out in the same degraded thin-films confirm the bulk formation of cubic Pb 0 for all samples and PbX2 phase for perovskite films (see section S9 for experimental details and XRD results). However, XRD measurements only take into consideration crystalline material formed. Accounting such differences between XPS and XRD techniques, the mismatch between the chemical compositions obtained can be understood. However, we note that both techniques corroborate the formation Pb 0 even the quantity trend observed for all samples.  Before light illumination, for the fresh PbI2 and MAPbX3 sample XPS spectrum of Pb 4f core electronic level showed the doublet peaks corresponding to spin-orbit splitting of Pb 4f7/2 and 4f5/2 at 138.3 eV and 143.1 eV, respectively. After light illumination, the degraded sample showed significantly lower intensity for the Pb 4f core levels, and new doublet peaks appeared at 137.0 eV and 141.9 eV corresponding to metallic-Pb. 8 XRD diffractograms for the corresponding thin-films are c), g) and k) before and d), h) and l) after 4 h light exposure. Crystal phases used for quantification were ICSD-68819 for PbI2, ICSD-96501 for the cubic F m -3 m Pb 0 phase, ICSD-238610 for tetragonal MAPbI3 and ICSD-252415 for cubic MAPbBr3.
S8. Further details on XPS measurements and analysis in thin-films.
The surface electronic properties of PbI2, MAPbI3 and MAPbBr3 were characterized by XPS (Kratos AXIS ULTRA HAS, monochromated Al-Ka = 1486.6 eV) in order to observe effect of the light exposure in vacuum on the surface chemical compositions. The binding energy (BE) was calibrated by measuring the Fermi edge (EF = 0 eV) and Au-4f7/2 (84.0 eV) on a clean Au surface. Freshly prepared samples were first analyzed by XPS. The BE scale of PbI2 spectra was calibrated using the adventitious carbon peak (C 1s) at ~285 eV as reference. 9,10 In our samples, residual amounts of adventitious carbon would be unavoidable due to air exposure prior to the XPS measurements. In addition, C 1s signal originating from residual solvents may be also expected. Great care was taken in order to minimize X-ray exposure time when acquiring PbI2, MAPbI3, and MAPbBr3 samples. X-ray induced sample damage was monitored by taking five consecutive scans and comparing these spectra. Acquisition time for each scan varied from 20 to 70 s depending on the core level regions. The five scans were averaged to a single spectrum if no significant changes were observed.
Peak fittings and standard deviation calculations were performed with CasaXPS 2.3.16. Shirley function was used to simulate the background signal due to inelastic scattering processes. 11 Raw XPS spectra of the Pb 4f, I 3d (for PbI2 and MAPbI3), Br 3d (for MAPbBr3), C 1s, and N 1s (for MAPbI3 and MAPbBr3) were fitted with Gaussian-Lorentzian (G-L) functions. Peak positions, full width at half maximum (FWHM), and the relative spectral areas were extracted from fitted curves. The intensity ratios between the 4f7/2 and 4f5/2 (Pb) and 3d5/2 and 3d3/2 (I and Br) doublets due to spin-orbit coupling were 1.33 and 1.50 (± 3% error), respectively. The concentration of the different elements (metallic-Pb, I, Br, C and N) relative to Pb 2+ was estimated from the fitted areas after normalization with the atomic sensitivity factors (ASF). [12][13][14] PbI2 thin-film XPS analysis XPS spectra revealed formation of metallic-Pb (Pb 0 ) in PbI2 film after the photodegradation under vacuum condition. Pristine PbI2 sample shows a typical yellow colored film (Eg ~2.3 eV), but it changed to a dark grey colored film after the light treatment. This color change under the light exposure is very likely due to the reduction of initial Pb 2+ to Pb 0 . 4 PbI2 XPS spectrum of Pb 4f core electronic level for fresh sample showed doublet spin-orbit coupling signals of Pb (4f7/2 and 4f5/2) at 138.3 eV and 143.1 eV ( Figure S15a). On the other hand, degraded PbI2 illustrated these peaks with less intensity, but also new doublet peaks at 137.0 eV and 141.9 eV ( Figure S15b), which were assigned to Pb 0 . Meanwhile, doublet signals of I (3d3/2 and 3d5/2) at 619.1 eV and 630.6 eV were observed in the fresh PbI2 sample ( Figure S16a). The peak heights, however, dramatically reduced after the photodegradation process. As release of I2 gas was monitored in the MS degradation experiment, the loss of I observed in the XPS spectrum would be also due to escaped I2 gas from the PbI2 film. While C 1s signal at BE of 295 eV did not significantly change its intensity ( Figure  S16b). This C is adventitious carbon. Atomic ratio of Pb 2+ : Pb 0 : I for fresh film was 1 : 0 : 1.8, which agrees quite well with expected nominal ratio of 1 : 0 : 2. Whereas the ratio in degraded PbI2 sample was 0.8 : 0.3 : 1 (Table S1 and Table S2). Significant increase in Pb 0 and decrease in I after the light treatment indicate that PbI2 is photodecomposed into Pb 0 and I2 gas.

MAPbI3 thin-film XPS analysis
Photodegradation of MAPbI3 perovskite implied reduction of Pb 2+ to Pb 0 as well as significant loss of I, C, and N.
Fresh MAPbI3 sample showed sharp and high intensity doublet signals corresponding to spin-orbit splitting of Pb (4f7/2 and 4f5/2) at 137.1 eV and 142.2 eV ( Figure S15e); whereas the degraded sample had less intense peaks at the BEs. Additionally, new doublet peaks at 137.0 eV and 143.9 eV ( Figure S15f) appeared in the degraded film. These findings suggest that Pb 2+ was reduced to metallic Pb 0 . In addition, peak intensities of I 3d3/2 and 3d5/2 doublet peaks at 619.5 eV and 630.1 eV, respectively, decreased after the light treatment on MAPbI3 film ( Figure S17a). The mechanisms of the loss of I from MAPbI3 is not as simple as the case of PbI2, but XPS spectra illustrated also dramatic perovskite C and N intensity peaks ( Figure S17b,c). Photodecomposition of MAPbI3 to HI, CH3NH2, CH3I, I2 and NH3 were recorded in our MS analysis ( Figure  2 shown in the main text). This indicates that it might have been released as I2, but also together with C as a form of CH3I. The peak position of adventitious carbon was shifted towards higher BE compared to that in PbI2 film. This is probably due to the overlap region with perovskite carbon peak and formation of unknown organic compound of polymeric nature in MAPbI3. 15 Atomic composition ratios of Pb 2+ : Pb 0 : I : CPVSK : N in fresh and degraded MAPbI3 were 1 : 0 : 3.1 : 1.3 : 0.9 and 0.8 : 0.2 : 1.6 : 0.3 : 0.1, respectively (Table S1 and S2). The fresh sample had an expected nominal ratio of 1 : 0 : 3 : 1 : 1, while the degraded sample possessed appearance of Pb 0 and significant decrease of I, CPVSK and N. Despite less amount of Pb 0 was observed in MAPbI3 compared to PbI2, the loss of I was even more significant in MAPbI3.

MAPbBr3 thin-film XPS analysis
Degradation of MAPbBr3 appeared to have the same trend as MAPbI3. Effects of the light exposure in vacuum on MAPbBr3 surface changes were monitored. XPS revealed the reduction of Pb 2+ to Pb 0 and decrease in Br, perovskite C and N signal intensities.
MAPbBr3 sample before the light treatment showed doublet spin-orbit coupling peaks of Pb 4f7/2 and 4f5/2 at 138.8 eV and 143.6 eV, respectively ( Figure S15i). These peak intensities decreased followed by formation of new doublets at 136.5 eV and 141.4 eV after the photodegradation treatment ( Figure S15j), which corresponded to the reduction of initial Pb 2+ to Pb 0 . Doublet spin-orbit splitting of Br 3d3/2 and 3d5/2 at ~68.5 eV and 69.5 eV, respectively, were recorded in MAPbBr3 film. Their peak intensities were apparently lower after the photodegradation ( Figure S18a). XPS signals corresponding to perovskite C ( Figure S18b) and N ( Figure S18c) core-levels also reduced after the photodegradation process.

S9. Further details on XRD measurements and analysis thin-films.
Powder X-ray diffractograms were recorded in GIXRD mode (detector scan, omega=0.5 o ) using a D8 Bruker Discover (Cu-Kα1 radiation) with 2θ degrees varying from 10 o to 55 o using 0.5 s of acquisition time for every 0.02 o 2θ intervals. Quantitative analysis for all powder samples were obtained fitting the entire XRD pattern with MAUD 2.71 software package. 16 Figure S19 shows XRD patterns of PbI2, MAPbI3, MAPbBr3 and PbBr2 before and after long term degradation tests (4 hours under 0.55 sun in vacuum conditions). Table S3 show percentage by weight (calculated molar percentage is shown in Table S1) obtained from refinements accounting the remaining corresponding lead halide derivatives and Pb 0 .