Photo-induced halide redistribution in organic–inorganic perovskite films

Organic–inorganic perovskites such as CH3NH3PbI3 are promising materials for a variety of optoelectronic applications, with certified power conversion efficiencies in solar cells already exceeding 21%. Nevertheless, state-of-the-art films still contain performance-limiting non-radiative recombination sites and exhibit a range of complex dynamic phenomena under illumination that remain poorly understood. Here we use a unique combination of confocal photoluminescence (PL) microscopy and chemical imaging to correlate the local changes in photophysics with composition in CH3NH3PbI3 films under illumination. We demonstrate that the photo-induced ‘brightening' of the perovskite PL can be attributed to an order-of-magnitude reduction in trap state density. By imaging the same regions with time-of-flight secondary-ion-mass spectrometry, we correlate this photobrightening with a net migration of iodine. Our work provides visual evidence for photo-induced halide migration in triiodide perovskites and reveals the complex interplay between charge carrier populations, electronic traps and mobile halides that collectively impact optoelectronic performance.

processing techniques and measured in nitrogen, showing that the rises are generally observed in polycrystalline films, but not in single crystals. The films 4,5 and single crystals 6 were processed as described in the Supplementary Methods. The HPA/acetate 7 films were used throughout the remainder of the work. The polycrystalline films and single crystals were excited with a 532 nm CW laser with intensities of 160 mW cm -2 and 300 mW cm -2 , respectively.  wavelength of 532 nm, with an intensity of (a) ~600 mW cm -2 (~10 sun equivalent, total photon dose of ~0.7 kJ cm -2 ) and (b) ~6000 mW cm -2 (~100 sun equivalent, total photon dose of ~3 kJ cm -2 ). No significant changes were observed.

Supplementary Note 3. Model to Describe Recombination Kinetics
The model to describe the recombination kinetics in the presence of subgap states is derived in detail in our recent work 10  but excludes trapped electrons and corresponding photo-doped holes. We note that we include excitons here for completeness but due to the low exciton binding energy in these materials, their presence does not significantly affect the results.
In Supplementary Fig. 4

19
(3) The parameters are the rates of exciton formation, dissociation and decay, respectively. We can then analytically solve (1) with .
We set the following parameters to be the same as our previous work: Rpop = 2x10 -10 cm 3 s -1 , Rdep = 8x10 -12 cm 3 s -1 , and fix γ0 =1.6x10 7 s -1 . This leaves the only fitting parameter to be the trap density NT. We globally fit the data across three orders of magnitude in Supplementary Fig. 4

Supplementary Note 4. Relation Between Photoluminescence Intensity and Trap Density
To illustrate the relationship between average PL intensity and the trap density, we rewrite the equation (1) with the effective recombination rate containing the contributions from both the direct band-toband electron-hole recombination and the trap-assisted (Shockley-Reed-Hall) recombination according to where we used the notations 00 / eh x R R A   for direct band-to-band recombination (radiative and non-radiative), and The time-in integrated PL (over the pulse period of duration t0) calculated using Eq. (8) is T  e  T  T  e  T  rad e  rad  T  T  e  T  T  e  T   T   I  n n n n n t n n n t n n dt tt n n n t n n n t Substituting here Eq. (7) Finally, the second term in the denominator will dominate over the first because the trap densities NT are large, such that: The inverse relationship given by Eq. (12) is shown in Figure 1b of the main text.

Supplementary Note 5. Temperature-Dependence of Photoluminescence Rises
We show the time-resolved PL decays measured in vacuo at low temperature (190 K) in Supplementary Fig. 5a and at high temperature (340 K) in Supplementary Fig. 5b. We note here that the temperature range is chosen to avoid the phase transition in the perovskite at ~160K from a tetragonal to orthorhombic phase, and to avoid high temperatures (>350 K) where degradation will likely occur 12,13 . We fit the time-resolved PL decays measured over time under illumination for each temperature, and the extracted trap densities as a function of time under illumination are shown in Supplementary Fig. 5c. After reaching stabilized emission levels, the trap densities reach temperature-dependent values, with lower trap densities at low temperature (NT ~ 2 x 10 16 cm -3 at 190 K) compared to those at high temperature (NT ~ 2.5 x 10 16 cm -3 and 4.3 x 10 16 cm -3 at 295 K and 340 K, respectively). We note that this temperature-dependence of the trap density is consistent with the longer stabilized monomolecular lifetimes ( Supplementary Fig. 5d inset) and higher PL intensities ( Supplementary Fig. 5c inset) at lower temperature ( = 280 ns at 190 K) than at high temperature ( = 132 ns at 340 K). These observations are also consistent with our earlier report where we showed that the PLQE approaches 100% at 190 K 10 .

Supplementary Note 6. Arrhenius Fits
We fit single exponential functions to the rise in PL over time (Supplementary Fig. 6) to extract a time constant for each temperature. We then plot the rate constants k=1/ versus 1/T and fit the data to the Arrhenius relation k = Aexp(-Ea/RT) (where A is a prefactor, R is the ideal gas constant) to extract an estimate for the activation energy Ea, as shown in Figure 1c (main text). We note that we can also fit the PL rise curves using two exponential functions corresponding to a short and long time scale 14,15 , but Arrhenius fits for each component separately yield similar activation energies. It is likely that the curves follow much more complicated functions and will need a more detailed analysis, but we simply use exponential functions to give estimates for the time scales involved.

Supplementary Note 7. Excitation Intensity-Dependence of Photoluminescence Changes
We present the results for different excitation intensities in Supplementary Fig. 7, where we observe that the trap densities and the time taken to reach stabilized emission varies dramatically with intensity, but ultimately similar total photon dose to reach stabilization for each case (e.g. ~200-300 J cm -2 at room temperature). At low excitation fluences, corresponding to photo-excited densities of ~10 15 cm -3 per pulse (1 MHz repetition rate), stabilized emission is reached only at times >10,000 s (~3 hours). This compares to thousands of seconds (~10 minutes) at the intermediate fluences (~10 16 cm -3 per pulse, 1 MHz repetition rate) also shown in Figure 1a of the main text, and only ~20 seconds for the highest fluences (~10 17 cm -3 per pulse, 1 MHz repetition rate). Consistent with previous reports 10,16,17 , we see a transition at an excitation density of 25 ~10 17 cm -3 from the trap-limited monomolecular kinetics (as seen in Figure 1a) to bimoleculardominating kinetics, in which the traps are predominantly filled.
Since it is commonly accepted that one of the free carriers (likely the electron) is trapped within 100 ns -1 μs 10,18,21,22 , this long lifetime must be that of the residual (non-trapped) carriers and is likely an underestimate because not all carriers will be trapped (some recombine radiatively). By solving simplified rate equations for the two processes of trap filling (100-ns time scale with typical trap densities of ~10 16 cm -3 10 ) and trap depopulation (100 μs as a conservatively slow estimate 10 ), it will not take longer than a few 100s of microseconds for the system to reach equilibrium. This implies that the slow (minutes) transient phenomena observed in this work are not primarily attributed to simple trap filling effects.

Supplementary Note 9. Scanning Electron Microscopy (SEM) Grain Analysis
We show a grain size analysis via scanning electron microscopy (SEM) in Supplementary Fig. 8.

Supplementary Note 10. Local Photo-induced Cleaning
We show how the PL enhancement closely follows the laser profile in Supplementary Fig. 9.

Supplementary Note 11. Film Excitation and Relaxation in the Dark
Supplementary Fig. 11 shows the changes of the bulk PL intensity after varying lengths of time in the dark. Here, the films are illuminated (spot size waist w ~ 17 µm) until they reach a stabilized emission, then the laser is switched off for a fixed length of time and then switched back on, and the PL (at the same spot) continually monitored. Supplementary Fig. 11a shows the situation where the PL is reducing over time in the dark, where the inset shows the PL recovery value as a function of time left in the dark. This suggests that the photo-induced changes can be at least in part reversible and the films can eventually recover to lower stabilized emission levels over a time scale of hours, which is consistent with the microscale PL measurements (Supplementary Fig. 10). In contrast, Supplementary Fig. 11b shows a situation in an identical film measured under identical conditions, and in this case the PL continues increasing over time in the dark although eventually does seem to decrease over very long time scales. These results suggest that the changes that the illumination triggers can continue even while the film is kept in the dark. More generally, we see that the PL moves through long-term 'phases', i.e. over very long time scales (1000s of seconds), we see periods where the emission rises and other periods where it decreases even for the same film and spot. The rises or drops we see in the emission over time in the dark tend to follow the phase of the long term transient. This suggests that there are changes in the film that are instigated by illumination that are partly reversible but that will continue even without illumination. Similar long-term phenomena (long-term phases and associated rises or drops over time) were also reported in these perovskites by Gottesman et al. from photo-conductivity measurements 23 , and they are also reminiscent of the photocurrent behavior under different bias conditions in solar cells 24 . We emphasize that here we are studying neat films with no contacts or applied bias. We also note that we cannot exclude the possible contributions of atmospheric effects such as adsorbed oxygen or water species that remain in films even kept under vacuum for long periods of time [1][2][3] .
We note that the photon dose used for collecting a fluorescence image (7 J cm -2 ) is only ~2% of the photon dose the film was exposed to under simulated sunlight (360 J cm -2 ) suggesting that the changes observed in Figure 3 (main text) are induced by the long light soak and not by the laser excitation required to collect a fluorescence image. This idea is further supported by the negligible changes in PL observed for a control film that had not been exposed to simulated AM 1.5 sunlight ( Supplementary Fig. 12).

Supplementary Note 12. Time-of-flight Second Ion Mass Spectrometry (ToF-SIMS)
For the ToF-SIMS measurements, we first light soak the spot to be analyzed (indicated by the red circle in Supplementary Fig. 13a) and then immediately transfer the film in the dark to the ToF-SIMS instrument and put the sample under ultra-high vacuum (~ 40 minutes). Instrument calibration takes an additional 20 minutes, so in total the film is in the dark for ~ 60 minutes before being depth profiled. We believe the local changes in PL are still retained based on the results reported in Supplementary Figs. 11 and 12 (~9 hrs until stabilization) and therefore any changes in composition should also be retained. ToF-SIMS is a surface-sensitive technique with typical molecular ion escape depths of a few nanometers 25 . For depth profiling, the signal intensity is proportional to the composition at the top surface of the film after each sputter cycle. We identified and analyzed several different negative ion fragments and report the depth-summed ToF-SIMS images most representative of the data set ( Supplementary Fig. 13), including I -, I2 -, PbI2 -, PbI3 -, Pb2I5 -, Pb3I7 -, PO3 -, and C3HN2 -(a fragment of methylammonium). In Supplementary Fig. 13a, we   28 show the iodide counts summed through the film depth and also define regions of interest for the illuminated region (red circle), an adjacent region (green circle) and a background region far from illumination (blue circle). As ToF-SIMS is primarily a qualitative technique, obtaining absolute changes in iodine content is not possible without careful calibration, and is therefore beyond the scope of this work. In order to extract an approximate relative change in intensity of iodinecontaining fragments (R) between light-soaked, adjacent, and background regions, we used the following equation: where M i − (… ) denotes the intensity of the iodine-containing fragment in the region of interest. To a first-order approximation, we estimate R= -1.2% in the illuminated region and R= +1.4% in the adjacent region, indicating that iodine has been partially redistributed. These values are significant given we only observe ~0.5% variations in several background regions, though we emphasize again here that there is likely a large error without proper calibration.
We briefly consider whether illumination could induce local variations in sputtering rates and ion extraction leading to artifacts in the intensity maps. In this possible scenario, we would expect the adjacent region (red circle in Supplementary Fig. 13a) to have a similar depth profile and intensity as the background region (blue circle in Supplementary Fig. 13a) -both of which have not been illuminated. In contrast, we still observe distinct profiles and intensities in the adjacent and background regions (Supplementary Fig. 14). This strongly suggests that material is moving laterally outside the illumination region. In addition, there are no other high-yield iodinecontaining fragments with similar intensity maps as iodide. If illumination was changing the sputtering rate and probability of ion extraction, we would expect a systematic artifact in at least some of the other fragment intensity maps.

Supplementary Note 13. Energy-Dispersive X-Ray Spectroscopy (EDS)
We show a semitransparent scanning electron microscope (SEM) image of the film in Supplementary Figure 16a overlaid on a PL map, again highlighting bright and dark spots. We show energy dispersive X-Ray spectroscopy (EDS) measurements at a dark ( Supplementary Fig.   16b) and a bright ( Supplementary Fig. 16c) spot to monitor the weight fractions of lead and iodide.
Here, we use a high energy 10 keV electron beam to generate electron-hole pairs via inelastic scattering within the material; this same mechanism is exploited in order to detect electron beaminduced current (EBIC) measurements in photovoltaic devices 26,27 and cathodoluminescence measurements 28 . We estimate a peak electron-hole generation rate of 2x10 23 cm -3 s -1 (see below), hence we are probing the samples under carrier densities similar to the PL measurements and ~10 sun solar illumination conditions (~3x10 23 cm -3 s -1 ). We note that carriers generated from the electron beam are distributed further in the bulk compared to carriers generated by photoexcitation ( Supplementary Fig. 16d). We find that for the dark spot, which is associated with a rise in PL over time under illumination, the iodide content reduces on a time scale consistent with the rise times in PL and device open-circuit voltage at 1-sun equivalent optical irradiation (Figure 1b inset, main text) 10 . In contrast, the bright spot has very little change in iodide content over time under electron excitation. As a reference point, the lead content remains essentially unchanged in both cases. We note that the weight fraction values are dependent on the local interaction volume and include all detected elements including those in the substrate, and therefore the absolute weight fractions should not be compared. We also cannot rule out local differences in volatization of iodide-containing species under high-energy electron excitation. Nevertheless, these EDS results suggest that the dark regions with high trap densities correspond to regions with excess mobile iodide, in agreement with the ToF-SIMS measurements.
We estimate that the pulsed excitation for PL measurements (~1 MHz, 0.5 J cm -2 per pulse), gives a peak charge density at t=0 of ~10 16 cm -3 and an average density over the decay of ~10 15 cm -3 .
This is roughly equivalent to the charge density arising from 1-sun equivalent irradiation (upper bound charge density of ~10 15 cm -3 ) 10,12,17 . We approximate the electron-hole generation rate under optical excitation using the average power output of the laser and by using the absorption coefficient reported elsewhere 29 , and we plot the resulting profile in Supplementary Fig. 16d (black circles). We estimate the electron beam generation rate as a function of depth using established cathodoluminescence equations 30 taking into consideration the measured beam current (53.5 pA), perovskite density (4.286 g cm -3 ) 31 , and perovskite bandgap (1.55 eV) 32 ; we also plot this in Supplementary Fig. 16d (green squares).
The maximum generation rate is quite similar for both photon and electron-beam excitations. For the PL experiments, we obtain a maximum rate of ~3x10 23 cm -3 s -1 compared to ~2x10 23 cm -3 s -1 for the electron beam measurements. However, the spatial generation profiles are quite different. The optical excitation generation rate peaks at the surface and drops off exponentially through the film following the Beer Lambert Law, while the electron beam generates most of the carriers deeper in the sample as a result of inelastic scattering.

Supplementary Methods
Open-Circuit Voltage Rises. Solar cells were fabricated on FTO-coated glass (Pilkington, 7 Ω sq -1 ). Initially, FTO was removed from regions under the anode contact by etching the FTO with 2 M HCl and zinc powder. Substrates were then cleaned as for the microscope slides. A holeblocking layer of compact TiO2 was deposited by spin-coating a mildly acidic solution of titanium isopropoxide in ethanol, and annealed at 500°C for 30 min. The perovskite precursor solution was spin-coated and the substrates annealed as for the microscope slides. After cooling, the spiro-OMeTAD hole-transporting layer was then deposited from a 66-mM chlorobenzene solution containing additives of lithium bis(trifluoromethanesulfonyl)imide and 4-tert-butylpyridine.
Finally, 120-nm-gold electrodes were thermally evaporated under vacuum of ~10 -6 Torr, at a rate of ~0.1 nm s -1 , to complete the devices.
For the transient open-circuit voltage rises, the devices were illuminated with a continuous-wave laser source at a wavelength of 532 nm and an intensity of 60 mW cm -2 , giving an approximately equivalent photoexcitation density to the 100 mWcm -2 AM 1.5 spectrum. The data were acquired using a sourcemeter (Keithley 2400, USA) coupled to a customized Labview program.

Preparation of Single Crystal CH3NH3PbI3
Lead iodide (98%) and anhydrous γ-butyrolactone (> 99%) were purchased from Sigma Aldrich and used without further purification. Methylammonium iodide (MAI) was synthesized by reacting methylamine (33 wt% in EtOH, Sigma) with equimolar hydriodic acid (57 wt%, Sigma) in an ice bath. The reaction mixture was stirred for 60 minutes and the liquid was removed with a rotary evaporator. The crude MAI solid was redissolved with EtOH (> 99.5%, Sigma), then precipitated and washed with diethyl ether (> 99%, Sigma). The MAI was dried and transferred to a nitrogen atmosphere. Our procedure for single crystal growth was based off of the inverse temperature crystallization (ITC) method of Saidaminov et al 6 . A 1.2 M solution of both PbI2 and MAI (1:1 molar ratio) was prepared in anhydrous γ-butyrolactone. The solution, exposed to ambient conditions, was preheated to 60 °C. Approximately 2 mL of this solution was filtered using a 0.2 μm PTFE filter. The filtrate was added to a 4 mL vial which was sealed shut before submerging in an oil bath heated to 80 °C. Over the course of two days, the temperature of the oil bath was gradually increased to 110 °C. After the two days, a ~1 cm 3 CH3NH3PbI3 single crystal was removed from the growth solution and washed with two 2 mL aliquots of acetophenone. The crystal was then dried and transferred into a N2 filled glovebox. The crystal was cleaved in the glovebox using a razor blade and the PL was measured under nitrogen flow.

Preparation of CH3NH3PbI3 (PbCl2 precursor)
Thin films of CH3NH3PbI3-xClx (PbCl2 method) were formed by first preparing a 40% weight precursor solution consisting of 2.64 M MAI (Lumtec) and 0.88 M PbCl2 (Sigma-Aldrich 99.99% purity) dissolved in DMF 4,5 . The solutions were spin-coated onto oxygen-plasma-etched glass at 2500 rpm for 60 s in a nitrogen filled glovebox and the substrates subsequently dried at room temperature for 20 mins and then annealed at 90°C for 2 hours.

Preparation of CH3NH3PbI3 (Dripping Method)
Thin films of CH3NH3PbI3 were also prepared using the solvent engineering ('dripping') method described in detail elsewhere 33,34 . In brief, MAI was synthesized and purified as described above.