The nature of the methylamine–MAPbI3 complex: fundamentals of gas-induced perovskite liquefaction and crystallization

In this work we scrutinize the exact interaction mechanisms between methylamine and perovskite based on extensive experimental evidence.


Introduction
In the last decade, perovskite photovoltaics (PV) have been investigated by a tremendous number of scientists and researchers, reaching outstanding power conversion efficiencies (PCEs) which in the laboratory have already exceeded the record efficiencies of such mature technology as polycrystalline silicon, Cadmium Telluride (CdTe) and Copper Indium Gallium Selenide (CIGS). 1 Such rapid development was realized by continuous improvements of the molecular structure, lm morphology and physical properties of perovskite crystals. One successful method to achieve this goal was proposed in 2015 by Zhou et al., where a peculiar phenomenon was described in which methylammonium lead iodide (MAPbI 3 ) has be liqueed upon exposure to methylamine (CH 3 NH 2 /MA 0 ) gas. 2 This liquefaction process is reversible and upon removal of the MA 0 -gas, crystalline MAPbI 3 is formed again. The chemical reaction has been suggested in literature as following: CH 3 NH 3 PbI 3 (s) + xCH 3 NH 2 (g) $ CH 3 NH 3 PbI 3 $xCH 3 NH 2 (l) This process has oen been used as a post-treatment step to improve lm morphology and crystal quality. [3][4][5][6] Enhancements of photovoltaic properties of solar cells using methylamine gas were repeatedly reported due to an improved lm quality, charge carrier lifetime, perovskite lm homogeneity and more favorable crystallinity, 5,[7][8][9][10][11] bridging to an efficiency of above 20%. 12,13 On the technological side, Noel et al. have proposed to dissolve perovskite precursor in a mixed solvent of acetonitrile (ACN) and MA 0 to produce a solution which can be used for manufacturing high-efficient perovskite solar cells with PCE of 18%. 14 In this process the solvating capability of ACN was increased by methylamine bubbling, leading to formation of a clear perovskite solution. In 2018, we have developed a dropcasting method where a similar approach can be used for inltration of the porous scaffold of carbon-based printable, low-cost, hole conductor-free perovskite solar cells (C-PSCs). The champion device had a V oc of over 1 V and a stabilized efficiency of 12.6% certied by Fraunhofer ISE CalLab. 15 Consequentially, manufacturing the photo-absorber layer of a PV device using methylamine-perovskite interaction establishing itself as a promising, cost-effective approach, that can be implemented on a large scale module production. [16][17][18] The question what stands behind the observed mechanism of perovskite liquefaction was already raised in previous reports. Zhou et al. supposed that the lone-electron pair in the nitrogen (N) atom of methylamine gas interacts with PbI 6 octahedra, which leads to a liqueed state. 2 It has also been noted that the "molten state" can be achieved by exposing the precursor salt PbI 2 to methylamine or other amines which suggests that the reaction is driven by interaction between the alkylamines and Pb(II). 7,8,[19][20][21] The interaction between MAPbI 3 and MA 0 gas was also described as a moderate ligand coordination of the lead atoms changing the lead halide framework. 8 Simultaneously, Rand et al. have reported that the proton transfer from PbI 2 to alkylamines results in the formation of Pb-amide bonds, responsible for the perovskite liquefaction. 20 Also the formation of colloids has been attributed to the presence of a xMA 0 -MAPbI 3 complex which is strengthened by witnessing Tyndall effect in this complex. 8,22 However, despite the substantial research in the eld of MA 0 gas treatment of CH 3 NH 3 PbI 3 perovskites 2-6,8-10,13,14,21-23 the exact mechanism of the methylamine-perovskite complex formation (which will be referred to as MPC further in this work) and the chemical bonding in the liqueed state remain unclear due to lack of experimental evidence. In particular the following underlying questions of this method remain pending: (1) how does the stoichiometry of the MPC formation reaction affects the crystal parameters, (2) what dictates the crystallization dynamics during up-take and removal of the MA 0 -gas and (3) what happens to the MAPbI 3 lattice when it is exposed to MA 0 gas molecules. We accentuate that the answers to those questions in this work contain vital importance not only for the fundamental chemistry scientists but also for numerous researchers in the perovskite community, which aim to reach high device efficiencies.
In this work, we were able to isolate single crystals and to determine the crystal structures of intermediates from the reaction of MA 0 with MAPbI 3 . These empirical results clearly present how MA 0 can coordinate Pb 2+ and MA + (CH 3 NH 3 ) leading to the formation of crystalline materials, which are stable only in the presence of MA 0 -rich environment. We combine the knowledge of crystalline intermediates to elucidate how does methylamine interact with MAPbI 3 perovskite, what are the possible reaction pathways and in particular how the properties of MPC affect the opto-electronic properties of the crystallized perovskite.
We demonstrate by Raman spectroscopy that, upon introduction of methylamine, the lead-halide framework changes through partial replacement of I À with MA 0 during its coordination of Pb 2+ and MA + . We also show the bonding environment in the molten state and how it correlates with the optoelectronic properties of the recrystallized lms. Through realtime photoluminescence (PL) microscopic measurement during crystallization we reveal the underlying mechanisms happening inside a porous layer (ZrO 2 /graphite which are normally a part of C-PSCs) during the conversion of MPC to crystalline MAPbI 3 . In addition, we investigate the effect of methylamine/perovskite ratio in the solution on the crystallized absorber layer. Our extensive analysis sheds light onto the chemistry of the MPC, its crystallization kinetics and the mechanisms to control the fundamental properties of perovskite absorber, which dene solar cell efficiencies.

Precursor solution preparation
The PbI 2 /MAI (1 : 1)-DMF complex (purchased from TCI, 99.99% trace metals basis) was heated at 100 C under nitrogen atmosphere and thus a nearly pure MAPbI 3 perovskite powder was obtained. The liqueed perovskite using MA 0 gas ($99%, purchased from Linde) of certain amount has been further diluted using acetonitrile with a ratio of 1 mg MAPbI 3 to 1 mL of ACN (purchased from Carl Roth >99%).

Sample preparation
The glass plates with uorine-doped tin oxide on top were treated in an ultrasonic bath containing 3% deconex OP153 at 60 C followed by rinsing with DI water. An ultrasonic treatment at 60 C for 1 min in DI water was further operated to ensure removal of any detergent residual. Glass plates were then blown dry using nitrogen. ZrO 2 (Zr-nanoxide paste obtained from Solaronix SA) of 2 mm thickness was screen printed on top of glass and sintered at 500 C for 30 min, aer which the graphitebased counter electrode of 12 mm thickness was screen printed, followed by sintering at 400 C for 30 min. Samples lled with perovskite in ZrO 2 and graphite layers were kept under MA 0 environment for 30 minutes to facilitate a better pore impregnation, aer which the samples were annealed at 80 C for 30 minutes. Perovskite lms were spin-coated on glass substrates, pre-treated with plasma and annealed at 80 C for 30 minutes.
For the complete perovskite solar cell devices (shown in ESI †) with carbon-based electrode, a compact TiO 2 layer of 25-30 nm thickness was deposited by spray pyrolysis using a solution comprising 0.05 M of titanium diisopropoxy (bis)acetyl acetone (Sigma Aldrich) diluted with ethanol ($99.5% from Carl Roth). Mesoporous layers of TiO 2 (paste prepared via mixing Dyesol DSL-18NRT with terpineol in 1 : 0.75 weight ratio) and ZrO 2 of 500 nm and 1 mm thicknesses respectively were screen printed and sintered at 500 C. Aerwards the graphite-based layer (Elcocarb B/SP paste purchased from Solaronix SA) was screen printed and subsequently sintered as described above. The positive and negative electrodes were applied by ultrasonic soldering on the FTO. The applied silver paste on the electrodes completes the architecture of the cell. The device active area was 0.4 cm 2 .

Characterization
Raman spectra were measured on a Bruker Vertex 70 with a Bruker RAM II module Nd:YAG laser (l ¼ 1064 nm) in a range of 50-4000 cm À1 . X-ray diffraction (XRD) patterns of lms were collected with a Bruker D8 diffractometer (CuKa ¼ 0.1540598 nm) equipped with a Lynxeye detector. The PL images were obtained by OLYMPUS BX50 stereomicroscope and sCMOS camera ("Zyla 5.5 sCMOS" by Andor) with a long-pass lter while the partial illumination of the sample was provided by a 623 nm red light-emitting diode (Thorlabs, SOLIS-623C). Due to the presence of long-pass we can exclude that the recorded signal originates from radiative recombination of free charge carriers and not from the reectance of the excitation source. Time-resolved PL measurement were conducted using externally triggered CW laser (Coherent Orbis) of 660 nm wavelength with 100 mW power, with a fall and rise time (10% and 90%) below 2 ns. The laser was triggered to produce rectangularshaped signals at 7 Hz frequency. The current-density and voltage curves of the devices were measured with a source meter (Keithley 2400) at a scan rate of 50 mV s À1 using a solar simulator providing a 100 mW cm À2 illumination intensity (of AM 1.5G).

Result and discussion
Synthesis and characterization of intermediates MA 0 /MA + / PbI 2 Following the conditions in the course of the liquefaction transformation process, single crystals of several intermediates were isolated and characterized. Passing a stream of MA 0 / CH 3 NH 2 gas over powdered PbI 2 yielded a highly viscous liquid from which pale-yellow platelets were crystallized. The crystal structure determination (Table S1 †) revealed a neutral complex of Pb(II) with the composition (CH 3 NH 2 ) 4 PbI 2 . A simplied description classies (CH 3 NH 2 ) 4 PbI 2 as a neutral octahedral complex with two iodine anions in a cis-position (Fig. 1a, details Table S1 †). CH 3 NH 2 acts as a classical donor ligand to Pb 2+ .
Passing a strong stream of methylamine evaporated from the cold trap over MAPbI 3 was readily liqueed and colorless crystals were precipitated from the MAPbI 3 xMA 0 . They were stable at room temperature if the sample was kept in a closed vessel with pure methylamine as equilibrium gas phase. For the measurement the crystals were directly transferred from the mother liquid into the cold N 2 -stream of the sc-XRD (at 100 K). Further examination revealed a multiply twinned crystal where the idealized structure is based on a face-centered cubic unit cell. The renement (Table S1 †) yielded a structure model close to the well-known K 2 PtCl 6 type with octahedral [Pb(CH 3 NH 2 ) 6 ] 2+ complexes and isolated I À anions (Fig. 1b, details Table S1 †). This K 2 PtCl 6 -related structure is a very common motif for hexammine complexes of divalent transition metal halides [M(II)(NH 3 ) 6 ]X 2 (M ¼ V, Cr, Mn, Fe, Co, Ni; X ¼ Cl, Br, I). 24 In some cases, small colorless needles were formed as byproduct which can't be recovered as MAPbI 3 in the course of the re-crystallization process. The small needles were characterized by sc-XRD as Pb(OH)I. 25,26 Additionally, pale-yellow platelets were separated from the bulk material, which were characterized as (CH 3 NH 3 ) 5 (CH 3 NH 2 ) 2 Pb 2 I 9 . This compound contains [Pb 2 I 9 ] 5À units consisting of two face-sharing octahedra and MA 0 /MA + dimers [CH 3 NH 3 -NH 2 CH 3 ] + bridged by a hydrogen bond (Fig. S1, details Table S1 †). The formation is explained by moisture according the following reaction: Composition of MA 0 -MAPbI 3 complexes in solution of liqueed MAPbI 3 and evolution/precipitation of perovskite From the above-mentioned ndings, it is strongly suggested that the liquefaction of MAPbI 3 in the presence of MA 0 is driven by its coordination of Pb 2+ and MA + . To gain new insights into the interaction between MA 0 and CH 3 NH 3 PbI 3 , an MPC solution (diluted in ACN) with high excess of MA 0 was produced in a closed vial. The Raman spectrum was then measured as the vial was opened in ambient conditions at room temperature and MA 0 gradually le the liquid over a course of 125 minutes throughout which the pale-yellow liquid gradually turned to saturated clear yellow and, consequently with additional annealing, turned dark, resembling crystalline MAPbI 3 . In the liquied state (Fig. 2, purple) at high MA 0 pressure, the most intense peak is at 78 cm À1 , followed by a peak at 94 cm À1 . Upon opening of the vial, an intermediate state is introduced which has a dominant peak at 104 cm À1 and a smaller peak at 82 cm À1 . Finally, upon additional annealing the bands at 94 cm À1 and 82 cm À1 arise. Furthermore, the bands at 155 cm À1 and 290 cm À1 are only present in the MPC and both of them  disappear during crystallization, while the new bands at 120 cm À1 and 242 cm À1 occur.
In the context of MAPbI 3 Raman shi spectrum, the region between 70 and 120 cm À1 is usually attributed to Pb-I framework interactions, such as symmetric and asymmetric Pb-I bond stretching. [28][29][30][31] As soon as the vial with the MPC is opened, MA 0 molecules start to leave and both peaks shi, indicating a structural reorganization of the Pb-I framework (i.e. a change in Pb-I valence vibrations) during the transition between the liquid and solid phases. The Raman shis in wavenumber region of 150-3000 cm À1 is commonly attributed to numerous vibrational modes of organic cations such CH 3 bending, C-N stretching, N-H stretching and others 29,32 (complete Raman shi spectra are presented in Fig. S2 in the ESI †). Notably, we highlight the presence of a band at 290 cm À1 existing only in the liquid state under high pressure, which we affiliate with a rotational mode of MA + . In a recent report from Park et al. a similar band at 288 cm À1 was found to be responsible for a rotational mode of MA + cation, which is normally inactive, but can be seen here either due to the fact that in the MPC the local symmetry of PbI 3 framework is lower or because MA + form MAI-MA 0 x during the MPC formation as will be shown later. 30 As seen from Fig. 2, the band at 290 cm À1 gradually reduces during the crystallization, while the band at 120 cm À1 rises, which can be attributed to MA + libration mode. 29 In addition, the bands at 1030 cm À1 and 990 cm À1 which nearly ideally correspond to the CH 3 -NH 3 rocking and C-N stretching vibrations respectively for a free MA + in vacuum, vanish during crystallization. 32 Simultaneously, the bands assigned to the same vibrational modes in MAPbI 3 appear at 911 cm À1 and 961 cm À1 for CH 3 -NH 3 rocking and C-N stretching, respectively. Such band shis suggest the reduction of free MA + cations and the increase of the Pb-I framework symmetry due to insertion of MA + yields recrystallized MAPbI 3 . In the presence of humidity, H 2 O and CH 3 NH 2 are reacting in an acid base reaction to produce MA + cations and OH À anions. In this case degassing results in the formation of crystalline (CH 3 NH 3 ) 3 (-CH 3 NH 3 -NH 2 CH 3 ) 2 Pb 2 I 9 and colorless rod-shaped crystals of Pb(OH)I. (CH 3 NH 3 ) 3 (CH 3 NH 3 -NH 2 CH 3 ) 2 Pb 2 I 9 is build up by isolated [Pb 2 I 9 ] 5À dimers of face sharing PbI 6 -octahedra (Fig. 3). Charge balance is achieved by MA + as well as the dimeric (MA-MA 0 ) + cations. Further degassing yields also MAPbI 3 , but Pb(OH)I stays as a by-product, which emphasizes that humidity should be avoided during the recrystallization of MAPbI 3 with MA 0 . Thus, the MAPbI 3 liquefaction may be driven by two effects, MA 0 bonding to Pb 2+ atom via lone electron pair as well as by the formation of MA 0 -MA + dimers. Very recently (this manuscript had already been submitted at that time), another publication on the liquefaction of MAPbI 3 using methylamine gas was published. 27 In accordance with our own work, they also found the existence of methylammonium-methylamine dimers. The dimers can be regarded as large cations with low charge density, reducing the lattice energy of the corresponding salt, which yields a situation comparable to ionic liquids.

Real-time crystallization dynamics observed by photoluminescence
We investigated the crystallization dynamics using real-time photoluminescence microscopy. Such characterization tool allows us to link the emission of photons from radiative recombination with the surface chemistry of the crystallizing layer to understand the behavior of nucleation and absorber layer formation inside the mesoporous scaffold. 33 A layer of graphite (12 mm) on top of thick ZrO 2 layer (2 mm) on a glass substrate was lled with perovskite solution (MPC with ACN) and subsequently annealed at 70 C to crystallize the perovskite. Then, the substrate was placed in a transparent gas chamber (as depicted in Fig. S4 †), to which MA 0 gas is later supplied, inducing the perovskite liquefaction. Aer reaching the lique-ed state, a small outlet was opened in the box allowing MA 0 gas to leave and upon rise in temperature induced by a transparent hotplate, the methylamine molecules start to evaporate even more rapidly inducing a crystallization within the mesoporous layers. A more detailed description of the setup and the procedure is shown in Fig. S4 and S5. † We ascribe several processes to the different stages observed during the liquid-solid conversion of the perovskite crystals. The real-time PL measurement starts with solid perovskite without MA 0 (denoted as stage (i) in Fig. 4). Upon introduction of MA 0 to the system, in a rst instance the PL sharply rises, reaching a PL intensity 4 times higher than initially and subsequently drops to PL counts close to zero (stage (ii)) since the perovskite now is in a liquied form (stage (iii)). When the MA 0 is removed again, the nucleation takes place, and a rise in the PL signal (stage (iv)) is observed, followed by a slow decay (stage (v)) during which the annealing takes place (stage (vi)). Aer that in the nal stages ((vii) and (viii)) the PL signal slightly rises and stabilizes at a value close to the initial one. The real-time PL microscopy measurement is provided in detail in Video S1 in ESI. † We manifest that the MA 0 gas forms the MPC with MAPbI 3 via chemisorption of gas molecules to the perovskite crystal lattice by replacing I À atoms, thus coordinating Pb 2+ as well as MA + . Therefore, when MA 0 is introduced to a perovskite bulk, it diffuses through the grain boundaries and is adsorbed by the crystal surface resulting in the formation of separated nanocrystals. This phenomenon is accompanied by the sharp rise in PL (as seen in stage (ii) Fig. 4), which arises from the reduced number of available bulk recombination centers per perovskite crystal. In literature, high PL in nano-sized perovskite crystals and quantum dots has been observed too. [34][35][36] Then the PL signal drops to nearly zero and the absorber layer color becomes transparent-yellow, highlighting that the ABX 3 crystal structure is collapsed leading to the inability of the perovskite to have a periodic wave function. Upon the removal of MA 0 from the MPC during the crystallization process, the (PbI x MA 0 6Àx ) 2Àx octahedra rst start to form into clusters, as the number of free MA + decreases (seen from the decrease of Raman shi band at 290 cm À1 ), and the number of MA + embedded into the ABX 3 structure increases (seen from the increase of band at 120 cm À1 ). We note that (PbI x MA 0 6Àx ) 2Àx are charged species where the total charge of a cluster is determined by the number of residing I À atoms. These clusters act then as nuclei in the beginning of the crystallization process. The supposed process is sketched in Fig. 5.
Revealingly, the adsorption and desorption of MA 0 both tend to increase the PL signal, since the non-radiative recombination mechanisms in bulk are suppressed. Between stages (iv) and (v), when the PL signal reaches its maximum (t ¼ 1.55 min) the bulk recombination starts to dominate because of crystal growth whereas surface passivation by MA 0 starts to fade away, decreasing the overall PL. During the recrystallization for Fig. 4 Investigation of liquid-solid evolution of MPC in mesoporous layers. The photoluminescence is monitored in real-time on the glass/ ZrO 2 /graphite device. When the MA 0 is supplied through an inlet, the layer turns pale-yellow, causing a drop in the PL signal due to the collapse of the ABX 3 structure. Once, the inlet is opened, the MA 0 leaves the MPC, reforming perovskite and causing a rise in PL. When the sample is heated up to $70 C, it induces a more rapid leave of MA 0 and the subsequent crystallization. The full PL microscopy measurement can be found in Video S1 in ESI. † $20 min, stabilization is reached as is indicated by constant slightly higher PL. During the stabilization, improvement of perovskite quality through the change of grain structure and removal of residual MA 0 caused by annealing, has been reported before. 14 A similar overall trend, where the PL starts to rise, then drops and rises slightly again during annealing until stabilization has been observed also before for the 2-step solution processing, where PbI 2 instead of MA 0 passivates the grain boundaries 14,37 until its complete conversion to the CH 3 NH 3 -PbI 3 , aer which the PL decreases. 33,38 Inuence of MA 0 concentration in the MPC solution on the fully-crystallized photoabsorbers Accurate estimation of the reaction products during liquefaction is challenging, therefore we quantitatively compare the effect of various given amounts of MA 0 in the solution by reacting it with the known quantity of CH 3 NH 3 PbI 3 . We prepared three different solutions using x (ratio of moles of MA 0 per moles of perovskite) of 19.2, 14.4 and 11.5. The estimation of x is shown in the ESI. † The lowest value (x ¼ 11.5) already allows the complete melting of the perovskite, whilst x > 20 lead to formation of white precipitates in the solution.
Perovskite lms were deposited on glass substrates by spincoating the MPC (diluted with ACN) of different ratios x of MA 0 in it. The obtained XRD patterns of each sample are shown in Fig. 6d, where the (110) and (220) diffraction peaks increase with higher concentration of MA 0 in the complex, while most of other reections decrease. Such trend highlights how the concentration of methylamine can directly affect the preferential orientation of the crystal lattice. Looking at the most pronounced diffraction signal of (110) plane, an increase in full-width at half maximum (FWHM) from 0.095 over 0.119 to 0.165 can be observed as the amount of MA 0 in the MPC rises. The change in FWHM indicates that by varying the MA 0 ratio in the MPC, the perovskite grain size can be tuned (i.e. large grain sizes are grown using high amounts of MA 0 ), which can also clearly be seen in SEM top-view images in Fig. 6a-c and crystal area distributions (Fig. S7 †), obtained by analysing these images via ImageJ.
It has been numerously reported that the change of stoichiometry in the precursor perovskite solution leads to different crystallization dynamics, which can lead to variation in such fundamental material properties as crystal size, charge carrier transport and contact at the interface between perovskite and charge extracting layer. [39][40][41][42] A similar dependency on the solution composition is seen in the MPC, where, as a result of the MA 0 /perovskite ratio, crystals with different domain size have been grown which coincides with the observations by Chen et al. 3 The differences in FWHM in the XRD measurements correlate with the observations by SEM. We attribute such behavior to a stronger chemical coordination of Pb 2+ and MA + with the higher MA 0 concentration, which causes a more preferential orientation of the crystal lattice along hhk0i planes such as (110) and (220), 8,43 resulting in the growth of larger crystals. It is worth mentioning that the commonly observed peak at 2q ¼ 12.7 which is usually observed when PbI 2 is formed 28,37,44,45 was absent in all the samples. However, this residual PbI 2 could be screened by the formation PbI 2 -alkylamide impurities, resulting in the reduced carrier lifetime of the crystallized absorber. The peaks at 2q ¼ 9.84, 19.74 and 29.79 which are attributed to MAI reections 28,46 are also not present, concluding that upon crystallization the MPC forms a pure, stoichiometric MAPbI 3 photoabsorber from an additive-free "perovskite melt".
The nding that the morphology of the nal crystal layers changes as a function of the x (MA 0 /perovskite) ratio suggests that the crystallization kinetics could also be different within the mesoporous layers of the C-PSCs. While perovskite crystals using MPC can even reach sizes of >1 mm on planar substrate, 43 the crystal growth inside the porous layers of C-PSCs is clearly different, as the majority of pores have diameters of less than 20 nm. Therefore, crystallization is conned by the pore size, which means that the interface plays a big role in the nucleation process as it provides multiple nucleation sites. 47 To gain a better understanding of the effect of ratio x in the solution on the crystallized absorber layer in porous isolating ZrO 2 scaffold, we use steady-state and timeresolved photoluminescence (TRPL) intensity measurements. A series of perovskite-lled mesoporous layers with different x variations were characterized. As seen from Fig. S8, † the steady-state PL intensity image of layers crystallized from a low-concentrated solution show strong PL signal. With higher concentration of MA 0 the average number of PL counts decreases, while the overall homogeneity of the PL signal increases as a result of better inltration in the mesoporous layers. We quantify the homogeneity (h) of the PL image by comparing the histogram of the PL intensity shown in Fig. S9. † Due to a higher concentration of MA 0 in the solution, the liquied complex becomes less viscous enabling an easier inltration through the pores. Zhou et al. suggested that the excess of MA 0 in the MPC "trims" the clusters, leading to a decrease in cluster size and lower viscosity. 48 A TRPL was employed to derive the carrier lifetime (s). Generally, as it has been demonstrated in numerous works, the PL-decay is affiliated with the transient behavior of radiative  -c), where the crystal size increases with higher MA 0 concentration, which can also be seen from an decrease in FWHM of (110) and (220) diffraction peaks at highly concentrated (with MA 0 ) samples. The presented XRD patterns of three samples show how strongly MA 0 coordinates the recrystallization of MAPbI 3 along (110) and (220) planes (d). Scale bar -300 nm. and non-radiative recombination of charge carriers in a semiconductor. 9,14,[49][50][51] In the absence of a quencher (as in this case), the longer the PL-decay, the longer the excited diffusing carriers are present in the perovskite, and therefore the longer carrier lifetime and diffusion length, both of which are vital characteristics of any PV device.
We quantify the charge carrier lifetime by tting the PLdecay to a bi-exponential function, accounting for Schockley-Read-Hall recombination processes at surfaces and in the bulk. The TRPL results in Fig. S10 † show that the decay is much slower (longer carrier lifetime) in the photoabsorbers fabricated with MPC using low ratio of MA 0 /MAPbI 3 in it (273.65 ns and 140.67 ns for x ¼ 8.6 and 11.1, respectively). Contrarily, for the higher ratio of MA 0 the PL-decay is signicantly faster: 20.9 ns for x ¼ 14.8 and 22.56 ns for x ¼ 18.2. As seen from Table 1, large amounts of MA 0 (x > 12) yield an absorber with lower carrier lifetime having lower averaged PL, indicating that despite strong advantages from crystallographic perspective (as seen from XRD patterns) such ratios of MA 0 are not desired due to poor opto-electronic properties of the layers. Conings et al. have also found that the recrystallized layer aer the MA 0 treatment helps the inltration through the pores, but, on the other side, leading to an inferior electronic quality. 10 Since no quenching layers are involved, the observed reduction in carrier lifetime is likely related to a more efficient bulk/surface non-radiative recombination processes. This means that there are either more of faster recombination channels, most likely in the form of trapassisted recombination. Since only the stoichiometry of the MPC changes without introduction of any other additive, the trap capture cross-section (i.e. quenching efficiency of each trap) should remain unchained. We therefore suggest that at high x, high number of trap energy states inside the bandgap outweigh the advantages of smoother lm and better inltration and start to play major role in dening the electronic properties of the absorber. Such a dependency of s on x emphasizes the importance of a careful control of xMA 0 (in reaction (1)) which should be within a certain processing window when producing solutions for perovskite PV devices.

Conclusions
In this work, the nature of methylamine-MAPbI 3 complex formation and its re-crystallization during the degassing in perovskite solar cells is presented. We show through XRD single crystal measurements, that MA 0 simultaneously forms neutral complexes with both Pb 2+ and MAI leading to the collapse of the perovskite crystal structure. In the excess (>> 1 atm) of MA 0 this reaction induces the formation and precipitation of nonperovskite solid crystals of [Pb(CH 3 NH 2 ) 4 I]I and [Pb(CH 3 -NH 2 ) 6 ]I 2 . We highlight that the presence of H 2 O leads to the irreversible formation of Pb(OH)I, which act as impurities in the photoabsorbing layer. Therefore, the formation of pure methylamine-perovskite complex needs to be carried out under normal atmospheric pressure of the MA 0 , and in the absence of water.
During the degassing of MA 0 from the complex, the iodide bonds with Pb 2+ , forming PbI 6 octahedra, while MA + cations get embedded in the A-site of the original pseudo-cubic ABX 3 structure. Real-time photoluminescence (PL) microscopy as a measure of the crystal quality visualizes the interaction of MA 0 gas with perovskite crystals. We postulate that the liquefaction starts from diffusion of MA 0 gas molecules through grain boundaries, leading to a break-up of bulk crystals and therefore reducing bulk recombination. This process is then followed by a complete collapse of the perovskite structure, as the PL signal approaches zero. When the MA 0 gas is desorbed aerwards in inert atmosphere, in accordance with the PL recovery a reversible process takes place, where the isolated non-bulk crystals of MAPbI 3 start to re-emerge from a liqueed state till the crystal coalescence is complete.
Furthermore, we nd that the amount of MA 0 in the complex plays a vital role in the control of crystal size and trap-assisted charge carrier lifetime. In planar lms with high MA 0 concentration, the MA 0 -MAPbI 3 complex leads to growth of larger crystals, while the low concentration leads to smaller crystals. In perovskite-lled mesoporous layers time-resolved PL decay shows that high amount of MA 0 in the complex causes an increase in trap-assisted monomolecular charge carrier recombination suggesting the formation of numerous trap energy states.
With this work we show that the understanding of liquefaction processes and crystallization kinetics is paramount for the further development of perovskite photovoltaics.

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