Understanding the Degradation of Methylenediammonium and Its Role in Phase-Stabilizing Formamidinium Lead Triiodide

Formamidinium lead triiodide (FAPbI3) is the leading candidate for single-junction metal–halide perovskite photovoltaics, despite the metastability of this phase. To enhance its ambient-phase stability and produce world-record photovoltaic efficiencies, methylenediammonium dichloride (MDACl2) has been used as an additive in FAPbI3. MDA2+ has been reported as incorporated into the perovskite lattice alongside Cl–. However, the precise function and role of MDA2+ remain uncertain. Here, we grow FAPbI3 single crystals from a solution containing MDACl2 (FAPbI3-M). We demonstrate that FAPbI3-M crystals are stable against transformation to the photoinactive δ-phase for more than one year under ambient conditions. Critically, we reveal that MDA2+ is not the direct cause of the enhanced material stability. Instead, MDA2+ degrades rapidly to produce ammonium and methaniminium, which subsequently oligomerizes to yield hexamethylenetetramine (HMTA). FAPbI3 crystals grown from a solution containing HMTA (FAPbI3-H) replicate the enhanced α-phase stability of FAPbI3-M. However, we further determine that HMTA is unstable in the perovskite precursor solution, where reaction with FA+ is possible, leading instead to the formation of tetrahydrotriazinium (THTZ-H+). By a combination of liquid- and solid-state NMR techniques, we show that THTZ-H+ is selectively incorporated into the bulk of both FAPbI3-M and FAPbI3-H at ∼0.5 mol % and infer that this addition is responsible for the improved α-phase stability.


Synthesis of Single Crystals
All single crystals were fabricated in N2 from a seed crystal by following a previously published experimental protocol 1,2 . The control FAPbI3 single crystals are prepared by dissolving equimolar FAI and PbI2 in γ-butyrolactone (GBL). The solution is then stirred at 60 °C for four hours. Then the solution is filtered with a 25 mm diameter 0.45 µm GMF filter. 4 ml of the filtrate is then placed in a vial that contains a seed FAPbI3 crystal, and the vial is kept in an oil bath undisturbed at 95 °C for 12 hours. The crystals were then dried in a vacuum oven at 180 °C for 45 minutes. For the FAPbI3-M and FAPbI3-H single crystals, alongside equimolar FAI and PbI2 either 3.8 mol% MDACl2 or 0.63 mol% HMTA is added. Then the same procedure as for the reference FAPbI3 crystals was followed.
For SCLC measurements: To control the thickness, a small chamber was constructed using two thin glass plates of different thicknesses placed on the bottom of the glass vial with a gap in between and a cover glass on top. Free-standing, millimetre-sized crystals were removed from the vial once formed. A schematic of the single crystal growth and device fabrication is depicted in Supplementary Figure S1. Figure S1: Schematic illustration of the single crystal synthesis and device fabrication process: 1. Single crystal seeding; 2. Seed crystal in confined growth template; 3. Addition of precursor solution; 4. Crystal growth in oil bath at 95 °C; 5. Free-standing single crystal; 6. Gold evaporation and gold wire attachment for electrical characterization measurements. cubic unit cell, increasing to 87% and 96% by the addition of two more twin components. The main component was detwinned and used to solve the structure. In the case of FAPbI3-H, only 47% of diffraction spots could be fitted with a single cubic unit cell. This could increase to 57% and then 66% by adding two more twin components. Fitting a fourth twin component only indexed 2% more of the diffraction spots, and so we stopped at three twin components. The remaining diffraction spots are likely to be due to the presence of some randomly orientated grains. The main component attributed to 47% In general, many of the crystals grown and measured in this work appear to be single crystals. However, for metal halide perovskites there have been several reports showing that both thin film grains and "single" crystals can contain several distinct crystallographic domains [4][5][6] . It is unclear if there is a link between the density of twinning in our crystals and the enhanced α-phase stability we observe. It has been suggested that point-defects can be the origin of domain boundaries due to memory effects seen when heating and cooling domain boundaries through phase transitions 5,6 . Comparing the diffraction spots observed for each crystal shows that our more stable crystals (FAPbI3-M and FAPbI3-H) also show a significantly greater density of well-defined twins, rather than the randomly orientated crystallites seen in FAPbI3. It is possible that the inclusion mode of THTZ-H + is such to have caused this effect, perhaps by inducing point defects that in turn induce distinct twin domains that are less likely to lead to propagation of the photoinactive δ-phase. Equally, however, this may be a confusion of cause and effect; the observation of many more randomly ordered domains in the FAPbI3 crystals may instead be early evidence of the onset of phase degradation, especially as it is not straightforward to assess if the majority of these disordered domains correspond to crystallites in the αor δ-phase. We emphasise that all our SCXRD measurements were performed on crystals that were freshly grown (<1 day old, in the case of FAPbI3) and had been stored in a static N2 environment from time of fabrication to time of measurement. As above, the measurements themselves were performed under a flow of temperaturecontrolled N2. All these precautions were taken in order to minimise degradation of the as-fabricated crystals to the δ-phase prior to measurement, nonetheless we cannot eliminate this possibility.   Index ranges -9 ≤ h ≤ 9, -9 ≤ k ≤ 9, -9 ≤ l ≤ 9 -9 ≤ h ≤ 9, -9 ≤ k ≤ 9, -9 ≤ l ≤ 9 -9 ≤ h ≤ 9, -9 ≤ k ≤ 9, -9 ≤ l ≤ 9   is coupled into an optical parametric oscillator to obtain various wavelengths in a home-built setup. The laser is attenuated by an optical density filter to reach several fluences for each wavelength (typically between 1.5 and 0.01 μJ cm -2 ). The spot size of the pulsed light is 0.5 cm 2 , which is 10x larger than the single crystals used, ensuring that they are uniformly excited. A small DC bias (< 5 mV/μm) is applied across two in-plane (lateral) gold electrodes that have been evaporated directly onto the single crystal.

Reflections collected
We monitor the voltage drop across the variable series resistor through a parallel oscilloscope (1MΩ input impedance) to determine the potential dropped across the two in-plane Au electrodes on the sample. We then extract the photoconductivity directly from the data and estimate the sum-mobility (Σμ) taking recombination and the free carrier fraction into account. The estimation error is greatly influenced by the sample thickness as diffusion in the z-direction is not taken into account in our analysis, which explains the error bars seen in Figure S4. For more detail see our recent work 10,11 . FAPbI3 thin film is presented alongside the data to indicate, where the optical bandgap is located.

Section S4: Photodetectors
Single crystal planar photodetectors were fabricated by applying two Ag electrodes onto the single crystal, as shown in the schematic below. The Ag electrodes were patterned on top of each single crystal using Ag-ink. A needle of 100 µm diameter has been used to mask the crystal surface while applying Ag-ink to produce a channel of 100 µm. The electrical characteristics of the photodetectors were monitored by a probe station that is connected to an Agilent B1500A semiconductor parameter analyser and a continuous light source with wavelength of 780 nm.

Section S5: Raman spectroscopy
Raman spectra were acquired with a Jobin Yvon T64000 triple spectrometer and an Andor DU420A-OE CCD. The spectrometer was calibrated with the 520.7 cm -1 line of a Silicon wafer before each measurement. Samples were excited with a Ventus solo Nd:YAG laser (λ = 532 nm) and the incident laser irradiation of the samples was strictly kept at intensities below 5 W/cm 2 to prevent laser induced degradation. This has been a major source of confusion for FAPbI3 Raman mode assignment in the literature and we thus opted for longer exposure times at extremely low laser intensities.
Raman spectroscopy has been used extensively to distinguish the δ-and α-phases in FAPbI3, as the different symmetries of the two phases allow identification via their Raman signatures. Han et al. were the first to report Raman signatures for both phases, assigning peaks at 111 cm -1 and 135 cm -1 to the δ-and α-phase, respectively 12 . This assignment has been used in numerous studies [13][14][15] since, even though the cubic Pm-3m symmetry of the α-phase is predicted to be Raman inactive by group theory.
Two recent reports by Driscoll et al. 16 and Ibaceta-Jana et al. 17 dissent previous assignments and provide experimental evidence that the α-phase is Raman inactive. These reports match our observations.
Comment on selection of peak at 108 cm -1 : There are four modes of the inorganic framework in the δ-phase: 35, 55, 83 and 108 cm -1 . Raman modes become harder to detect the closer they appear to the laser line (0 cm -1 ) due to its linewidth.
This has two consequences: 1. Most Raman spectrometers can only detect the line at 108 cm -1 , as they use optical filters to cut out the laser line.
2. For modes closer to the laser line, there will be more of a background issue from the tail of the laser line. This could be subtracted, but for a quantitative analysis it's more straightforward to take a line slightly further away.
For compounds containing organic entities in an inorganic framework, like FAPbI3, we can separate the Raman modes of the framework from the modes of the organic part. Modes of FA are all substantially above 200 cm -1 ; modes of the haloplumbate framework in the range <200 cm -1 . By looking at the 108 cm -1 peak, we are probing the PbI bond. However, we prefer to think about this in terms of the symmetry of the PbI cage instead of individual bonds, as the symmetry defines whether something is Raman active or not.
We note that the single crystals analysed after aging for 1 year (FAPbI3 and FAPbI3-M) and 132 days (FAPbI3-H) and presented in Figure 1f originate from the same single crystal fabrication batches as the crystals used for the 33-day-experiment (the first 33 days). However, these crystals were stored in glass vials under ambient conditions, while the crystals in the 33-day-experiment were constantly directly exposed to air. The Raman spectra of freshly annealed (α-phase) and fully degraded (δ-phase) FAPbI3 single crystals are shown in Figure S9a. Both spectra were acquired with identical experimental settings, but the δ-phase was divided by 400. This shows that the Raman scattering cross section of αphase FAPbI3 is at least two orders of magnitude weaker than for the δ-phase 13 . The overall Raman signal of any given FAPbI3 sample will consequently be dominated by δ-phase peaks as soon as a small fraction of the probed layer is degraded.
This allows us to track the degraded fraction of the probed surface layer by monitoring the δ-phase peak intensities, even if the crystal is predominantly in the α-phase. We note that peak area is always proportional to the amount of δ-phase present in the crystal and, as is the case here, if peak width is constant then peak intensity correlates directly with peak area. The Raman spectra used for the single crystal stability data in main text Figure 1f are shown in Figure S9b-d.  The peak splitting in the broad delta phase peaks at 55 cm -1 and 83 cm -1 that occurs in some of the FAPbI3-M spectra in Figure S9c indicates a structural change within the δ-phase 12 . The peak splitting has been observed in all crystal types independent of additives and does not affect the peak at 108 cm -  For experiments represented in Figures 2a-c, S11, S14, S17, S18, S20-22: A two-channel Bruker Avance III HD Nanobay 400 MHz instrument running TOPSPIN 3 equipped with a 5 mm z-gradient broadband/fluorine observation probe is used. The signal from residual nondeuterated DMSO solvent is used for reference.

For experiments represented in Figures 3a, 3b:
Spectra were acquired with a Bruker AVIIIHD 600 equipped with a BBO prodigy probe. Directionless enhancement of polarisation transfer (DEPT) experiments were carried out as described by Doddrell, et al. 18 .  We note that the coupling observed as a result of 1 H interaction with the quadrupolar I = 1 14 N nuclei is typically only observed in the case of a near-zero electric field gradient at the quadrupolar nucleus, otherwise quadrupolar relaxation becomes too rapid to permit coupling 19 . This suggests a species with a high degree of molecular symmetry. Moreover, the substantial coupling constant observed precludes anything but a 1 J interaction, implicating direct 14 N-1 H bonding and leaving NH4 + as the only likely candidate species.
We note that the 1 J14N-1H coupling characteristic of ammonium is not initially observed in the spectrum of NH4Cl (pH unadjusted). This could be due to a breakdown in the tetrahedral symmetry of the cation caused by either the rapid exchange of acidic H + in solution, or the close association of the chloride ion with the cation. Addition of hydrochloric acid to the NH4Cl solution simulates the pH environment of

Comments on the assignment of signals to intermediates in MDACl2 degradation
From the concurrent evolution and disappearance of the signals at 7.87 and 4.35 ppm (7 and 6) in Supplementary Figure S14 we infer that these signals correspond to 1 (Supplementary Figure S11). These two signals are also shown to be spin-spin coupled by COSY (Supplementary Figure S13). Not only does the rapid proton exchange evident in coordinated amine-ammonium groups lead to observation of a single mixed signal, but it also effectively suppresses spin-spin coupling between 1 H involved in coordination and adjacent 1 H. If this were not the case, resolution of 1:2:1 triplet splitting of signals corresponding to 1 H neighbouring amine moieties could be expected, which we do not observe.
In light of the above, and based on our proposed mechanism, we suggest the most likely identity of the principal intermediate observed (signals 4 and 5) is NH4 + -coordinated MDA + . As set out in the main text, the singly protonated diaminomethane is expected to be stabilised with respect to either the diammonium or diamine equivalents. Moreover, spin-spin coupling between the ammonium and the with the resulting dication being less charge-dense than MDA 2+ , and thus likely more stable. B 2+ is mechanistically relatively stable to both backward reaction (via attack by ammonia, which is strongly basic and thus almost entirely protonated in solution or involved in coordination with ammonium groups, or via kinetically inhibited elimination) and forward reaction, at least while substantial quantities of MDA 2+ remain in solution as the latter is more charge-dense, and thus more electrophilic than B 2+ .
Coordination of the amine moiety of B 2+ by NH4 + is expected, as discussed above, and thus spin-spin coupling of the methylene 1 H nuclei (6) to an adjacent ammonium group (7) produces the expected quartet signal. The presence of quartet splitting in both signals 5 and 6 assists greatly in reducing the number of candidate species. However, given the nearly identical nature of the two sets of intermediate signals we have highlighted, it is not possible to conclusively distinguish which of these corresponds to MDA + -NH4 + and which to B 2+ -NH4 + . Our assignment is based on the observation that signals 4 and 5 disappear earlier in the degradation process suggesting the species giving rise to them occurs earlier in the degradation process.
We note that the apparent quartets observed for signals 5 and 6 might be 'AB quartets', i.e. heavily roofed doublet of doublets. These are typically associated with locked ring systems in which axial and equatorial 1 H nuclei are inequivalent, but spin-spin coupled with a large 2 J coupling constant. If this were the case then one of the two pairs of signals discussed could correspond to 1,3,5-trazinanium (C + ).
However, AB quartets typically show non-uniform spacing between their constituent peaks unless 2 J splitting and the chemical shift difference between the two isolated signals is perfectly matched. As we find that peak spacing is highly uniform in both signals (Figure 2c) and the exception described is highly unusual, we conclude this interpretation is very unlikely. Although the stepwise reduction in pH leads to a gradual downfield shift in signals corresponding to 1 H nuclei in water molecules, as expected, a single discrete downfield shift is observed in the signal corresponding methylene 1 H in HMTA due to the monoprotonation of HMTA to HMTA-H + . Further reduction in pH has negligible effect on this signal. We note that in the monoprotic form, the acidic H + need not be localized on a single amine functional group, but instead may be captured inside the cage via equivalent interactions with all amine groups.    For simplicity, the mechanism shown in the main text only produces a single methanimine (CH2=NH) from each HMTA consumed. This is sufficient to demonstrate mechanistically that THTZ-H + can form in-situ in solution containing multiple HMTA molecules and FA + . However, in a protic environment (such as that provided by mixture with excess FA + ), the remains of the HMTA cage show in Figure 3c can itself degrade further by additionand subsequent consumption of -HMTA acting as a nucleophile.
Overall, and disentangling the requirement for multiple HMTA molecules playing a role, each single HMTA molecule is expected to give rise to four methanimine molecules.
By contrast, each MDA 2+ cation added produces only a single methanimine. Thus, addition of 3.8 mol% MDACl2 produces a maximum 3.8 mol% mim + , while 0.63 mol % HMTA produces a maximum of 2.53 mol% mim + in solution.
We emphasise that, as with all mechanistic details presented in this work, the schemes set out should be interpreted as mechanistic justifications only, demonstrating thatby established chemical mechanismsit is possible for the observed species to have been produced from one another. We have not conducted extensive mechanistic studies, and have only identified a relatively small number of the species displayed in the mechanistic schemes.   Figure S20 were all made in the same crystal growth batch, but a different batch to those in S19, and their spectra were taken on a 300 MHz Bruker spectrometer. There is a marked difference in the appearance of signals corresponding to FA + in the spectra, both in whether two amidinium 1 H chemical environments are resolved, and in the degree of resolution of vicinal 1 H-1 H coupling between amidinium and methine 1 H nuclei. In Figure S20, the two amidinium environments (which differ in whether they are E or Z with respect to the methine 1 H environment, and are resolvable because of the reduced rate of rotation of the conjugated N-C-N system) are distinguishable in all samples. In Figure   S19, only FAPbI3-M shows the expected two singlets. It has previously been shown 22 that the linewidth of 1 H NMR spectra (of perovskite solutions) is sensitive to the acidity and/or halide content of the solution in which the perovskite materials are dissolved. In general, line-broadening can also occur due to unsuccessful or incomplete shimming of the sample before spectrum acquisition. Considering we have observed the THTZ-H + in the spectra of single crystals in which the amidinium environments were not resolved, and in those in which the amidinium environments are resolved, we do not believe this difference is related to the chemistry giving rise to enhanced α-phase stability. Equally, in Figure S20,

Comments on the appearance of low intensity signals in 1 H NMR spectra throughout
In main text Figures 2a-b and 3a, and in Supplementary Figures S16-17 show comparable phase stability to their peers where trace GBL is not detected.
Also apparent is a signal corresponding to H2O at 3.33 ppm. The presence of water in spectra corresponding to dissolved single crystals in particularly crucial. As the amount (indicated by integration of the signal) of water present seems to be variable, we do not attribute this only to H2O introduced with the d 6 -DMSO solvent, although this is common. Some water may come from incomplete drying of NMR tubes used to prepare samples. However, there is also likely a degree of moisture adsorption on the surface of the crystals that is highly dependent on factors such as surface to volume ratio, storage conditions and crystal age. Across our study, single crystals were typically stored in vials sealed under nitrogen. However, there is substantial variability in crystal size, and when multiple measurements are made on a crystal it may be exposed to ambient air for different lengths of time. The role that differing quantities of adsorped water may play in accelerating α-FAPbI3 phase transformation is unclear, but certainly worthy of further study. We note that the amount of water observed via 1 H NMR does not correlate with improved phase stability.

Comments on GBL washing experiments to investigate THTZ-H + surface layer hypothesis
There appears to be a smaller quantity of THTZ-H + in the spectrum of the GBL-washed FAPbI3-M crystal than that of the unwashed crystal. Although integration of such small signals is challenging, this appears to be born out by relative peak integrations of the FA + and THTZ-H + signals. However, the nature of this experiment (in which a crystal is irreversibly dissolved in DMSO-d 6 , and so the two spectra shown are from the dissolution of two different crystals) complicates the analysis possible. Given that surface area to volume ratio is essential in determining the expected THTZ-H + :FA + stoichiometry under the hypothesis that the former is only found on the crystal surface, crystal-to-crystal variation renders quantitative analysis of this nature impossible unless there is no THTZ-H + present whatsoever. As we do not observe this, we cannot conclusively say that THTZ-H + is only present on the surface of FAPbI3-M and FAPbI3-H crystals. .   Table S4 shows computed cation sizes for a range of candidate A-site cations reported for lead-halide perovskites, as well as mim + , MDA 2+ , THTZ-H + and HMTA-H + studied in this work. We highlight that a range of different methods for estimating organic cation sizes for application in halide perovskites have been reported [23][24][25] . In this work, we use only the approach reported in Filip et al. 23 throughout, for consistency, and we compare where possible with cation radii as defined by Kieslich et al. 24 . We estimate the size of the MDA 2+ , mim + , HMTA-H + , THTZ-H + and other cations as the radius of a sphere centred at the centroid of the cation charge density, which incorporates 95% of the total integrated charge.

Section S7. First Principles Calculations of Cation Steric Radii
We calculate the charge density within density functional theory (DFT) 26 as implemented in the Quantum Espresso code 27 , using the Perdew-Burke-Erzerhof parametrization of the generalized gradient approximation (DFT-PBE) 28 . We used the Optimized Norm Conserving Vanderbilt (ONCV) pseudopotentials 29 as found on the Pseudo Dojo repository 30 constructed for the PBE functional to describe the atoms. The structure of all isolated cations reported in Table S4 are optimised within DFT-PBE, by simulating vacuum using a large unit cell of 25 Å, compensating the cation of +1 with a background charge and using a plane wave cut-off of 75 Ry.
We note that, even our use of the stericrather than ionicradii fails to account for the substantial anisotropy of many of the organic cations discussed. The assumption that these organic cations may be considered spherical, and so assigned a radius of any kind, is in general increasingly invalid with increasing anisotropy/non-sphericity, and with the size of the organic cation. Moreover, such calculations fail to account for directed bonding interactions that exist between the A-site cation and components in the Pb-X inorganic scaffold, which may (depending on the geometry of the organic cation in question, and the distribution of its electron density) act to either stabilise or destabilise the 3D perovskite structure. All discussion presented here must be assessed with consideration of these limitations.
Given the size of THTZ-H+ comparative to DMA+, GUA+ and EA+, it is unlikely that THTZ-H+ can be completely incorporated into the A-site of the perovskite; however, we cannot rule out the possibility of partial incorporation in small concentrations. , where and are ionic radii of the B-and X-site ions, respectively, such that a 3D perovskite structure would for if 0.8 < ≤ 1.0, with = 1.0 being geometrically 'ideal'. Based on our calculated steric radii (Table S4) and Shannon radii of Pb 2+ (1.19 A) and I (2.20 A), we calculate the tolerance factor t for 3D ABX3 structures based on a selection of the cations listed in Table S4. These are marked by solid circles in Figure S22. We also compute values of t for these structures based on values for the ionic radii of Pb 2+ (1.42 A) and I - (2.49 A) estimated using the same protocol as discussed above. These are marked by open circles in Figure   S22.
We emphasise, however, both that the Goldschmidt tolerance factor is calculated on the basis of a single A-site cation and also takes no account of the globularity of organic cations 25 or any modes of bonding that may exist between the A-site cation and the haloplumbate structure in which it resides.
Although it is possible to linearly combine the A-site cation radii constituting A-site alloyed perovskites and calculate a tolerance factor for the 'average structure', such an approach is limited as it further fails to take account of the possibility offor exampleco-alignment or co-misalignment of adjacent nonspherical cations and/or regular octahedral tilting in the haloplumbate network, by which mechanisms an A-site alloyed perovskite structure might be stabilised. Further, as the exact mode of THTZ-H + incorporation remains unclear, it is important to note that the Goldschmidt tolerance factor offers no mechanism by which to account for ion substitutions besides direct (stoichiometric) replacement of the A-B-or X-site ions in the perovskite structure. This may not be sufficient to full account for THTZ-H + incorporation. Improvements on Goldschmidt's tolerance factor have been reported [40][41][42] , notably that of Bartel, et al., however most such efforts have focused on the limitations on ABX3 structural stability of adjacent halide separation and to our knowledge none have satisfactorily addressed the limit of nonspherical A-site cations or a provided a mechanism whereby non-stoichiometric substitutions can be accounted for.
Within these limitations, however, we note that even for a hypothetical "(THTZ-H)PbI3" perovskite the Goldschmidt tolerance factor calculated using our steric radii is less than the 'ideal' value of 1, but significantly larger than all reported APbI3 crystals with 100% A-site occupancy.  Table   S1 in the Supplementary Material of Filip et al. 23 ). The experimentally-reported phase stable region for 3D ABI3 perovskites (with only one A-site cation) is between Cs + (lower) and FA + (upper), as highlighted.
Both CsPbI3 and FAPbI3, have a metastable 3D cubic perovskite phase under ambient conditions.

Section S8: Solid-state Nuclear Magnetic Resonance Spectroscopy
The single crystal samples were packed into 1.  Integration of the selected regions indicates that the new species is present at ~0.5 mol% relative to FA in both materials, assuming that the number of protons in FA and in the new species is the same.  The formation of a solid solution is therefore the only mechanism that agrees with the experimental 1 H-1 H spin-diffusion and 127 I NQR data. On the other hand, the presence of GBL within all three types of crystals (reference FAPbI3, FAPbI3-H and FAPbI3-M) indicates that the solvent is kinetically trapped during crystallization as inclusions which exist independently of the THTZ -H + (note that those inclusions are present also in the reference FAPbI3 which has no THTZ -H + ).

Structural model of FAPbI3-H based on the 1 H-1 H spin-diffusion spectrum
For completeness, we also consider the hypothesis that THTZ-H + could template the growth of FAPbI3 such that the resulting perovskite lattice is more disordered relative to undoped FAPbI3 (octahedral tilting induced by surface templating, as reported by Doherty et al. 44 ). In this scenario, the dopant would only be present on the crystal surface but still lead to a qualitative change in the local iodide environments, as probed by 127 I NQR, throughout the crystal volume. While our results could be consistent with this hypothesis, we present a series of argumentswith supporting experimentationfrom which we eliminate this possibility.
First, we note the relatively large THTZ-H + content in the crystals (ca. 0.5 mol% relative to FA).
Considering the low surface area of the crystals (on the order of 0.001 m 2 /g), the amount of incorporated THTZ-H + is several orders of magnitude larger relative to what would be expected for a surface monolayer. For example, considering a cubic FAPbI3 crystal with an edge length of 1 mm and a density of 4.16 g cm -3 , a uniform layer of THTZ-H + (assuming density 1.5 g cm -3 and a FA:THTZ-H + molar ratio of 1:0.005) on the cube surface would have a thickness of about 100 μm. Such a thick surface layer of THTZ-H + (or THTZ iodoplumbate) would be expected to diffract X-rays, but we did not detect any such additional phases our XRD experiments. Further, it might be expected that such a capping layer could be removed by an appropriate solvent. We washed away the exterior layers of FAPbI3-M and FAPbI3-H crystals by repeated submersion in GBL (three washes) and conducted 1 H NMR solution NMR measurements on the resultant crystals. As shown in Supplementary Figure S21 the washing process did not remove signals corresponding to THTZ-H + . Quantitative analysis of a reduction in THTZ-H + content by this treatment is impossible due to crystal-to-crystal variation, as discussed above. It is possible that our washing process failed to remove all surface features or that THTZ-H + remaining in the washing solvent is retained on the surface during drying. However, we further note from our Raman spectroscopy that we do not observe that the use of either HMTA or MDACl2 additives has any effect on the position and width of the Raman modes of FAPbI3, as might be expected in the presence of substantial non-FA + organic components at the surface 45 (Supplementary Figure S10). Finally, we observe that crystals cleaved to reveal the bulk α-FAPbI3-M and α-FAPbI3-H material do not degrade at an enhanced rate (Figure 1e) as would be expected if these surfaces did not benefit from the hypothesised capping layer.

Section S9: Electron Probe Microanalysis (EPMA)
Electron probe micro-analyses were performed on a Cameca SX-5 FEG in the Department of Earth Sciences at the University of Oxford. Samples were mounted in epoxy resin, ground flat and polished under oil, using a combination of silicon carbide and diamond laps. Samples were coated with ~20 nm of carbon to minimise any charge build-up during the analysis. A 15 kV accelerating voltage was employed, with beam currents between 20 and 24 nA and a nominal spot size of 10 μm. Cl, I, and Pb were quantified using NaCl, Tl(Br,I), and galena (PbS) standards respectively. On-peak counting times for Cl (Kα) and Pb (Mα) were 60 s and 45 s for I (Lα), with half of on-peak counting times at each of the high and low background positions. Using these conditions the observed intensity of I X-rays is timedependent, decreasing somewhat throughout the analysis. Thus, we applied a time dependent intensity (TDI) correction (e.g., Nielsen et al. 46 ) to the I X-ray intensity whereby the on-peak intensity was sub-