Reversible Decomposition of Single-Crystal Methylammonium Lead Iodide Perovskite Nanorods.

Perovskite solar cells offer remarkable performance, but further advances will require deeper understanding and control of the materials and processing. Here, we fabricate the first single crystal nanorods of intermediate phase (MAI-PbI2-DMSO), allowing us to directly observe the phase evolution while annealing in situ in a high-vacuum transmission electron microscope, which lets up separate thermal effects from other environmental conditions such as oxygen and moisture. We attain the first full determination of the crystal structures and orientations of the intermediate phase, evolving perovskite, precipitating PbI2, and e-beam induced PbI2 during phase conversion and decomposition. Surprisingly, the perovskite decomposition to PbI2 is reversible upon cooling, critical for long-term device endurance due to the formation of MAI-rich MAPbI3 and PbI2 upon heating. Quantitative measurements with a thermodynamic model suggest the decomposition is entropically driven. The single crystal MAPbI3 nanorods obtained via thermal cycling exhibit excellent mobility and trap density, with full reversibility up to 100 °C (above the maximum temperature for solar cell operation) under high vacuum, offering unique potential for high-performance flexible solar cells.

Introduction

Organo-lead trihalide hybrid perovskites (MAPbX3, where MA is CH3NH3 and X is a halogen atom such as I, Br, or Cl) have enabled remarkable progress[1] in the field of solar cells, including power conversion efficiency (PCE) of over 20%, due to their high optical absorption coefficients,[2] long charge-carrier diffusion lengths,[3] high defect tolerance,[4] and tunable band-gaps[5] with low-cost production processes. The key to high efficiency is the quality of crystallization and morphology,[6−13] which is achieved using antisolvent engineering[10,14−16] to precipitate an intermediate phase. This process requires thermal annealing to convert the intermediate phase to the crystalline perovskite,[10,14,17] but heat is also one of the main drivers (together with moisture,[18,19] oxygen,[19,20] light,[21] etc.) for perovskite decomposition. Annealing beyond the minimum temperature needed for perovskite formation leads to decomposition with formation of PbI2, degrading the performance.[22]

MAPbI3 bulk single crystals were first studied in 2015.[24−26] Subsequently, perovskite bulk single crystals with different compositions were synthesized and the properties of these materials were also characterized in detail.[27−29] All of the reported perovskite bulk single crystals exhibited superior properties, including red-shifted absorption, favorable charge transport properties, and good stability,[27−32] leading to great success in the application of these materials in X-ray detectors, γ-ray detectors, and visible light photodetectors.[33−37] Early perovskite single crystals had millimeter-scale thicknesses, far greater than the carrier diffusion length, which was problematic for photovoltaics. Therefore, methods were developed to grow micrometer-thick perovksite single crystal thin films (SCTFs), such as cavitation-triggered asymmetrical crystallization,[38] vapor phase epitaxial growth,[39,40] a space-confined method,[41] surface-tension assisted growth,[42] and a top-down method.[43] Still, several challenges remain to grow high-quality and large area SCTFs, to form good interface contacts between the SCTFs and substrates, and to preserve their properties over a range of operating conditions. These factors ultimately control the efficiency of solar cells.

Hence, to further improve the efficiency and stability of perovskite solar cells requires a fundamental understanding of the various phases and their transition, in particular perovskite decomposition with PbI2 precipitation. In situ transmission electron microscopy (TEM) has already enabled direct observation of perovskite degradation at high resolution,[23] and here we extend this powerful technique by growing single crystalline intermediate-phase nanorods and observing their transformations in situ during thermal processing in a high-vacuum TEM, which effectively guards the nanorods against oxygen and moisture.

It has been generally believed that decomposition of MAPbI3 occurs by irreversible loss of MAI via evaporation or reaction with oxygen or moisture.[22,44,45] Here we find a very different and remarkable behavior. When MAPbI3 is protected from extrinsic reactants, perovskite decomposition to PbI2 still occurs even at temperatures too low for MAI evaporation. Remarkably, this decomposition is reversible upon cooling. The reversibility appears critical for the long-term device endurance at the elevated temperatures that occur during normal solar cell operation. Our quantitative measurements let us develop a model explaining the reversible decomposition as entropically driven, with the excess MAI accommodated in stacked perovskite sheets.[46] Thus, the decomposition is not a cumulative destructive process but rather a fundamental part of the equilibrium phase diagram.

Annealing the intermediate-phase nanorods at 100 °C transforms them to single-crystal perovskite with excellent electrical properties; the hole mobility is nearly the same as that of bulk single crystals,[24] and the trap density is much lower than epitaxial films.[47] The electrical properties of these single crystal perovskite nanorods are temperature-dependent, reflecting the decomposition of the perovskite: with increasing temperature, the electrical properties degrade as the amount of PbI2 precipitation from the perovskite increases. Most importantly, the properties are fully reversible under thermal cycling up to 100 °C, which exceeds the maximum temperature for solar cell operation. Using electron diffraction and high-resolution, in situ TEM during thermal cycling, we obtain a full determination of the sequence of transformations and the crystal structures of the resulting phases. These include the original intermediate phase, the evolving perovskite, and PbI2 formed by decomposition. By varying the e-beam dose, we clearly distinguish between the actual phase evolution and e-beam effects.

Results and Discussion

To determine the initial structure of the nanorod, we carry out TEM measurements. Figure a shows a typical TEM bright field (BF) image of a broad view of the nanorods obtained at room temperature (RT). The nanorods are formed on a SiO2 membrane grid by spin-coating the precursor solution to which controlled amounts of γ-butyroactone (GBL) and dimethyl sulfoxide (DMSO) solvent are added (see the Methods section for more details.) As expected, the nanorod is comprised of an intermediate phase,[10] as identified by X-ray diffraction (XRD) and selected area electron diffraction (SAED). Figure S1 in the Supporting Information compares our XRD spectrum with that of the intermediate phase from ref (10). Figure b shows the SAED pattern, indicating that the nanorod is single crystal with the orthorhombic structure (space group Cmc21, a = 0.4621 nm, b = 2.7484 nm, c = 2.6923 nm, α = β = γ = 90°).[29] The diffraction spots are elongated in the direction normal to the nanorod axis, which we interpret as reflecting the presence of polysynthetic twins with a distribution of lamella widths (Figure c). We note that the structure is stable under e-beam exposure at a dose rate of 28 × 102 e–/nm2 s up to about 10 min at room temperature during TEM measurements.

Figure 1

Images of intermediate phase nanorods characterized by TEM. (a) BF image of a large view of the nanorods on a SiO2 membrane TEM grid. The scale bar is 1 μm. (b) BF image of a nanorod marked in a square in part a with the corresponding SAED pattern. The scale bar is 100 nm. (c) HRTEM image of the white square region in part b with the corresponding FFT. The lattices with red and white indicate the main and twin crystal structures, respectively. The width of twin varies. The scale bar is 2 nm. The lattice spacings of (15̅0) and (18̅0) are marked as 0.365 ± 0.005 and 0.337 ± 0.005 nm, respectively. (d) Schematic of indexing the diffraction patterns of intermediate (purple), PbI2 particle (orange), and PbI2 matrix (sky-blue). (e) Schematic of indexing the diffraction patterns of intermediate (purple), PbI2 particle (orange), cubic perovskite (blue), and tetragonal perovskite (violet).

To directly monitor the conversion of the intermediate phase into perovskite, we used in situ TEM during the annealing cycle for the sample in Figure . It is well-known that heating transforms the intermediate phase into perovskite,[10] and the perovskite decomposes to PbI2 under e-beam exposure. Here we deliberately used an e-beam with a dose rate of 28 × 102 e–/nm2 s which is sufficient to instantly evaporate MAI from a perovskite, without damaging the intermediate phase. This leaves a PbI2 crystal (called PbI2 matrix) with crystal orientation inherited from the perovskite,[49] as shown below.

Figure a–d shows a series of BF images of representative nanorods extracted from a video where the sample was heated in stages to 80, 100, 130, and 155 °C, respectively. At 80 °C, we see a nanorod plus a few tiny PbI2 particles (see arrows in the BF image of Figure a), and we determine the detailed crystal structure using SAED (Figure a inset). The nanorods still consist primarily of an intermediate phase oriented to [001]. In addition they include some PbI2 crystal matrix (space group, R3̅m:H, a = b = 0.4557 nm, c = 2.0937 nm, α = β = 90°, and γ = 120°) oriented to [4̅401], indicating that it was previously a perovskite with a cubic structure aligned in [001] prior to beam-induced decomposition.[49] The PbI2 particles (space group, P3̅ml, a = b = 0.4557 nm, c = 0.6979 nm, α = β = 90°, and γ = 120°) are oriented to [0001], indicating that they precipitated from the perovskite prior to e-beam decomposition. The alignment of the three crystals are indexed in Figure d and illustrated in Figure S2. Specifically, [150] and [15̅0] of the intermediate phase are tilted clockwise and counterclockwise ∼6.6° about [0001] of PbI2 particles with respect to [101̅0] and [011̅0] of PbI2 particles, respectively, and both planes of the intermediate phase have shorter lattice spacings (0.365 nm) than those of PbI2 particles (0.394 nm). Moreover, (08̅0) of intermediate phase and (1̅100) of PbI2 particles are parallel, but the former has a shorter lattice spacing (0.337 nm) than the latter (0.394 nm).

Figure 2

In situ TEM observation during annealing cycle. (a–h) Sequence of BF images acquired from a video obtained during heating and cooling cycles for perovskite conversion, decomposition, and recovery. The images (a–d and e–h) were obtained at 80, 100, 130, 155, 130, 100, 80 °C, and room temperature, respectively. The dark particles with strong contrast are PbI2. The scale bar is 50 nm. (i–k) BF images of the nanorods taken at room temperature after annealing cycles, with maximum annealing temperatures of 80, 100, and 130 °C during the respective cycles. The scale bar is 50 nm. The white arrows indicate PbI2 (R3̅m:H) matrix, and orange and purple arrows indicate PbI2 (P3̅m:I) particles and intermediate phase, respectively. (l) An indexed SAED pattern obtained at 100 °C with a weak e-beam without evaporating MAI from perovskite. (m–o) HRTEM images of the PbI2 (R3̅m:H) matrix from parts d and h and of a PbI2 (P3̅m:I) particle from part d. The scale bar is 2 nm.

The index in Figure d reveals that [011̅4] and [101̅4] of the PbI2 matrix are tilted clockwise and counterclockwise ∼15.7° about [0001] of PbI2 particles with respect to [101̅0] and [011̅0] of PbI2 particles, respectively, and both planes of the PbI2 matrix have much shorter lattice spacings (0.317 nm) than those of PbI2 particles (0.394 nm). Additionally, (112̅0) of the PbI2 matrix are parallel to (1̅100) of PbI2 particles, but the former has a much shorter lattice spacing (0.223 nm) than the latter (0.394 nm). Furthermore, all of (200) of the intermediate, (112̅0) of the PbI2 particle, and (11̅08) of the PbI2 matrix are parallel and have identical lattice spacings (0.226 nm). As mentioned above, we note that thermally and electron beam induced PbI2 structures are different. Because a cubic perovskite aligned in [001] is nearly identical to PbI2 matrix aligned in [4̅401] except for {100} and {110} which are only involved in the perovskite, the relationship of structural orientation among intermediate, perovskite, and PbI2 particle can be inferred from Figure d as exhibited in Figure e. It is noteworthy that the spot elongation present before annealing disappears, presumably due to the removal of a considerable amount of DMSO.

Complete transformation to single crystal perovskite was achieved at 100 °C, reflected by the PbI2 matrix crystal oriented in [4̅401] and no intermediate phase left in the nanorod, with a small amount of PbI2 particles precipitated as exhibited in Figure b. To confirm that the PbI2 matrix is converted from perovskite and their diffraction patterns are correspondent except for {100} and {110} in perovskite, SAED was carried out using a weak e-beam (dose rate of 1 × 102 e–/nm2 s) for less than a minute, which avoids beam-induced decomposition but is insufficient for BF imaging to quantify the precipitated PbI2 particles.[49] The SAED pattern acquired before taking the BF image at a higher dose rate in Figure b is shown in Figure l, confirming that the nanorod is composed of a perovskite aligned in [001] and PbI2 particles aligned in [0001], in the crystal orientation described in Figure e. With increasing temperature, the sizes of PbI2 particles become larger while maintaining their crystal orientation with the perovskite (Figure c,d), and this decomposition by heating leads to increasing roughness of the underlying nanorod. The crystallinity of the PbI2 matrix and particles are confirmed by using high-resolution transmission electron microscopy (HRTEM) images as shown in Figure m–o, respectively.

Real-time observation also allows us to quantify the growth kinetics of the PbI2 particle at different temperatures. To avoid beam-induced effects, we frequently moved the sample to observe fresh nanorods at nonirradiated sites at each time and temperature. We measured the volume fractions of the PbI2 particles as time progressed, Figure a. At each temperature, the PbI2 particles appear stable with time. Evidently, when temperature is increased, they grow too quickly for us to observe, reaching a new stable size within a few seconds.[50] The temperature dependence of the stable volume fractions of the PbI2 during annealing is plotted in Figure b and is found to be consistent with Arrhenius behavior (inset of Figure b). Details of measurement method and the particle size distribution are given in the Supporting Information and Figures S4 and S5. To ensure against e-beam effects on PbI2 precipitation, we also monitored one nanorod for an extended time. No observable change in the sizes and number of the PbI2 particles was identified at 155 °C for ∼5 min (Figure S3).

Figure 3

Quantification of the PbI2 precipitated from perovskite decomposition. (a) Plots of volume fraction of PbI2 particles vs time t during the thermal cycle. Approximating the particle image by an ellipse, the radius r is calculated as the geometric mean of the semimajor and semiminor axes; that is, the radius of a circle with an equivalent area. (b) Temperature dependence of PbI2 volume fraction relative to total volume for heating (red) and cooling (blue) cycles. The gray circles were obtained from 10 additional experiments with other nanorods. The inset is an Arrhenius plot of the data of part b, fitted with an activation energy 0.6 eV. The details are shown in Figure S5d–h. (c) PbI2-MAI phase diagram showing the boundary between single-phase and two-phase regions; the brown line was calculated from fitting the model to the data in part b.

Following the sequence discussed above, we cooled the sample through the same sequence of temperatures (Figure a) and examined its morphology and phase content at each temperature. The BF images of typical nanorods in Figure e–h were acquired at 130, 100, 80, and 25 °C, respectively. Surprisingly, the size and density of the PbI2 particles decrease with decreasing temperature and at 25 °C only a single-crystal PbI2 matrix remains (inset of Figure h,n). This implies that the nanorods had converted back to tetragonal perovskite thermally before being decomposed by the e-beam. To confirm this, as before, we performed SAED of a nanorod at each temperature using a weak beam (i.e., dose rate of 1 × 102 e–/nm2 s)[49] to avoid beam damage (Figure S6). The SAED patterns taken at 25 °C of the nanorods that experienced thermal cycles up to 100, 130, and 155 °C are shown in Figure S6d,h,l, respectively. All three exhibit the diffraction pattern of single crystal perovskite, confirming that the decomposition is completely reversible during thermal cycling. To confirm the reversibility of PbI2, we repeated the annealing cycle 10 times at various temperatures (gray data in Figure b). The data for all heating and cooling cycles collapse onto a single curve of PbI2 fraction vs temperature. We note that the in situ results shown in Figures and 3 are consistent with the nanorods in the whole area of the sample.

The thermal decomposition of MAPbI3 is generally attributed to irreversible processes–evaporation of MAI or reaction with oxygen or moisture.[22,44,51] In remarkable contrast, we find that when protected from environmental reactants, the decomposition is completely reversible up to 165 °C. Our thermogravimetric analysis (TGA) data obtained at a vacuum similar to the TEM show that the evaporation of MAI is negligible up to 180 °C (see the Methods section for more details and Figure S7). Moreover, the compositions of nanorods were measured using the line-scan mode in nanoprobe energy dispersive X-ray spectroscopy (EDX) during different thermal cycles (see the Methods section for more details and Figure S8). We find that the I/Pb ratio in the perovskite region of the nanorods is initially 3 (as for stochiometric MAPbI3) and increases to significantly more than 3 with increasing temperature. We confirmed the perovskite character of these regions using HRTEM after the EDX measurements. It is known that MAPbI3 can accommodate a large range of excess MAI via intercalation, and as the amount of excess MAI increases, the structure may transform progressively into stacked perovskite sheets where the perovskite still persists as a 3D crystal.[46] We therefore suggest that MAPbI3 decomposes into MAI-enriched MAPbI3 and PbI2. Such lossless decomposition requires protecting the system from moisture and oxygen, which we achieve by maintaining high vacuum in our TEM (1 × 10–7 Torr). To confirm this, the same annealing experiments were performed in an in situ gas holder (2 × 10–2 Torr) and an in situ XRD (3 × 10–2 Torr). In these cases, the PbI2 fraction increased more rapidly with temperature than in the high-vacuum TEM, and the PbI2 fraction continued to increase even during the cooling cycle (Figures S9 and S10), consistent with an irreversible reaction.

It is common to have reversible decomposition upon cooling, as occurs in spinodal decomposition,[53] where a material transforms from a single phase at high T to two-phase coexistence at low T. Here we have the opposite, reversible decomposition upon heating. Yet we believe the underlying cause is the same, transformation between a higher-entropy state at higher T and a lower-enthalpy state at lower T. In our case, we propose that decomposition actually raises the entropy. The initial single phase, MAPbI3, is an ordered compound with low entropy. Decomposition converts this into two phases, one or both of which are off-stochiometry and thus less ordered (higher entropy) than the original compound.

To explain how heating could drive reversible decomposition, we consider a highly simplified model, which contains only the minimal elements required to illustrate the effect. The elementary building blocks in this model are the molecules MAI and PbI2, with stochiometric MAPbI3 being an ordered compound consisting of two molecules per formula unit and with the possibility to insert excess MAI molecules into the MAPbI3. (We approximate the PbI2 as pure, for simplicity.) Precipitation of PbI2 leaves excess MAI in the MAPbI3, increasing the enthalpy but also the entropy. From these assumptions, the equilibrium fraction f of PbI2 vs temperature isto the lowest order in f, where α is the number of possible sites for insertion of MAI per formula unit of MAPbI3, ν is the number of internal configurations (e.g., rotations) for each site, H is the enthalpy change per MAPbI3 unit for decomposition, and k is the Boltzmann constant. Equation gives an excellent description of the data in Figure b with H = 0.6 eV and αν = 3 × 105, and it allows us to calculate the phase boundary of the MAPbI3 as shown in Figure c, consistent with the values measured by EDX (Figure S8). We emphasize that the actual perovskite/intercalation system is far more complex than this, so the fitted parameters H and αν may not directly correspond to physical properties of this system, but the model clearly illustrates how the observed phenomena can arise thermodynamically, because decomposition into nonstochiometric material increases the entropy.

For photovoltaic applications, the electrical properties are critical. To characterize the temperature dependence of hole transport in the nanorod, we fabricated two-terminal single-nanorod devices. Pt electrodes were deposited at both ends using FIB. An SEM image of the device is shown in the inset of Figure a (see the schematic illustration of the device in Figure S14a); the channel length is 1.9 μm and the rod’s width is 248 nm. We confirmed that the intermediate phase of the nanorod remained intact after FIB, using SAED measurements. The device dark current was measured in real time while annealing the device in vacuum (same pressure as in the TEM) to estimate the trap density and carrier mobility using the space-charge-limited current (SCLC) model[24,47,54] (Figure S11). As seen in Figure a–d (the thermal cycle to 155 °C), the J–V traces of the nanorod at varied temperatures feature a general trend. There is a linear (ohmic) J–V relation at low bias. With increasing bias, there is a remarkable increase of the current injection, which we attribute to filling of traps. At still higher bias, we observe the child regime, with a quadratic dependence of the current density on the applied voltage. The trap density (nt) of the nanorod was calculated by trap-filled limit voltage (VTFL), and the carrier mobility (μ) was evaluated in the child regime using the Mott–Gurney law; see the Supporting Information for relevant equations.

Figure 4

Characterization of dielectric and charge transport properties of individual nanorod devices. (a–d) J–V plots of the single nanorod device during the annealing cycle to 100 °C; RT before annealing, 100 °C, 80 °C, and RT after the thermal cycle, respectively (see Figure S11 for other cycles). The device is shown in the SEM image in the inset of part a. The scale bar is 2 μm. (e) Temperature dependence of dielectric constant at 1 MHz of a thin film of the material identical to the nanorod during annealing cycle to 155 °C. (f,g) Temperature dependence of hole mobility and charge trap density of the nanorod during the annealing cycle. Four regimes are specified as RT to 80 °C (regime I), 80 to 100 °C (regime II), 100 to 155 °C (regime III), and 155 °C to RT (regime IV). The dotted arrows indicate the cooling cycles after heating to 100 °C (green), 130 °C (orange), and 155 °C (blue). Note that the purple data are the repeated heating cycle to 100 °C after complete conversion to perovskite via thermal cycle to 100 °C.

To calculate carrier mobility and trap density in the nanorod, we first determined the temperature dependent dielectric constants (ε) by measuring capacitances of the same material in the thin film form during the annealing cycle[55] (see the Supporting Information for details and Figure S12). The variation of film properties with temperature is nearly identical to those of nanorods, implying that these behaviors are general and not limited to a particular morphology (see the in situ TEM BF and the relevant plots in Figure S13). The ε vs T plot (Figure e) exhibits nearly ideal reversibility, presumably reflecting the reversibility of decomposition and the almost identical ε of perovskite and intermediate phases. Dielectric constants measured at different temperature cycles show similar reversibility (Figure S12e–h). Based on the rule of mixtures, the ε of perovskite and PbI2 with their volume fractions should determine the ε of the total system,[57] allowing us to estimate[58] temperature dependence of ε of the MAI-rich single crystal perovskite (see the Supporting Information for details and Figure S12c), which is infeasible to obtain at such a wide range of temperatures due to perovskite decomposition.

Temperature dependence of the carrier mobility and trap density are plotted in Figure f,g. We first address annealing to 100 °C, since this is the minimum temperature for complete perovskite conversion and is above the maximum temperature for typical solar cell operation. During heating, the mobility slowly decreases at first (regime I), followed by a large decrease by a factor of 20, down to 5 × 10–2 cm2 V–1 s–1 at 100 °C (regime II). During cooling, the mobility fully recovers (regime IV, green data). The trap density initially decreases with heating, slowly until 80 °C (regime I) and then more rapidly, dropping by 33% from 80 to 100 °C, to 6.74 × 1014 cm–3 (regime II). During cooling the trap density decreases again by 32% (regime IV, green data). We repeated the cycle to 100 °C and confirmed that after the initial conversion from the intermediate phase to perovskite, both properties are fully reversible (see the purple data in Figure f,g). This suggests that the traps and mobility reduction reflect the excess MAI and/or PbI2 precipitates during reversible decomposition.

It is important to note that annealing at 100 °C produces single crystal perovskite nanorods with mobility (1.41 cm2 V–1 s–1) nearly equal to single crystal perovskites,[24] and trap density (4.58 × 1014 cm–3) an order of magnitude smaller than epitaxial perovskite thin films.[47] While there is already significant decomposition (∼3%) and an order of magnitude decrease in mobility at 100 °C, the excellent electrical properties are preserved after thermal cycling, reflecting the reversible decomposition. Solar cells can get quite hot during operation, but still their maximum temperature is typically well below 100 °C.[23,51] This suggests that in a device with an appropriate protective coating,[59−64] thermal decomposition still occurs unavoidably, but it is not a cumulative path to degradation and failure. Rather, the decomposition appears fully reversible, so it does not limit the operational lifetime.

Next, we measured the transport properties on different samples during the thermal cycle to higher temperatures, 130 and 155 °C. As the maximum temperature increases from 100 °C (green data) to 155 °C (blue data), mobility and trap density degrade over the entire cooling cycle with the additional regime III (Figure f,g). There is still a partial recovery (a factor of ∼40 improvement) upon cooling, again reflecting the reversible decomposition (Figure S11a–h). We repeated the annealing-cycle measurements at these three maximum temperatures with five devices for each, providing consistent trends (the deviation of the data is ∼12%). These trends directly correlate with the degree of roughness visible in Figure h,j,k. Other properties (fPbI which reflects the amount of excess MAI in perovskite and ε) are reversible over this range of temperature. We therefore believe that the degradation results from the roughening of the nanorod occurring during the process of reversible decomposition of perovskite at higher temperatures.

It seems surprising that the mobility gets worse even while the trap density is improving during conversion from intermediate phase to perovskite (∼100 °C). It is reasonable that initially mobility decreases due to formation of interphase boundaries between the intermediate phase and perovskite, followed by a faster decrease caused by scattering by the roughness, excess MAI in the perovskite, and perhaps other defects introduced during decomposition. We also suggest that the decrease in trap density reflects the elimination of planar defects as illustrated above (Figure i), which are aligned along the current path. Thus, this trend does not significantly contribute to the mobility.

Conclusion

We believe that the results shown here are of fundamental importance for understanding and optimizing perovskite processing for applications including photovoltaics. Our results provide new insight into the conversion from an intermediate phase and the decomposition induced by subsequent heating as well as the effect on dielectric and charge transport properties. All of these are essential in optimizing the power conversion efficiency and stability of photovoltaics. Synthesizing single crystal nanorods in an intermediate state is shown to be an ideal approach to quantify the intrinsic aspects of the aforementioned phenomena in an in situ TEM whose high-vacuum state lets up separate thermal effects from other environmental effects. The fabrication of single crystal perovskite nanorods by spin-coating and subsequent annealing is simple and cost-effective, compared to other methods such as vapor phase deposition,[65−67] and it may enable flexible devices based on single crystal perovskite nanorods.[68,69]

Methods

Material Preparation

CH3NH3I (MAI) was synthesized by reacting 27.86 mL of CH3NH2 (40% in methanol, Sigma-Aldrich) and 30 mL of HI (57 wt % in water, Sigma-Aldrich) in a 250 mL round-bottom flask at 0 °C for 4 h with stirring. The white-colored MAI powder was obtained using a rotary evaporator and redissolved in 80 mL of ethanol, recrystallized from 300 mL of ethyl acetate, and dried at 60 °C in a vacuum oven for 24 h. The prepared MAI and PbI2 (Sigma-Aldrich) were mixed in a 1:1 molar ratio, and then 0.04 M precursor solution dissolved in a mixture of γ-butyrolactone (GBL) and dimethyl sulfoxide (DMSO) (7:3 v/v) was stirred vigorously at 60 °C for 12 h. The concentration range of the precursor in solution to form nanorods is relatively narrow (0.03–0.07 M); at lower concentrations, the rods are too thin and becomes dissociated, while at higher concentrations, the density of rods is so high that they become entangled and form films. The mixed precursor solution (10 μL) was spin-coated onto the SiO2 membrane grid (3 mm size, TEM windows) by 3,000 rpm for 30 s with toluene (30 μL) drop-casting. The toluene was dripped onto the mesh grid at about 27 s after the beginning of spin-coating. During spin-coating, the mesh grid was fixed and supported by acid-resistant friction tape on the center of the supporting substrate (about 1.5 × 1.5 cm2 slide glass, Marienfeld Superior from Germany) and subsequently UV/ozone-treated for 10 min to control the surface energy of the SiO2 membrane. We note that the duration of UV/ozone-treatment and concentration of solution are critical for the formation of nanorods. We note that using the membrane grids made of other materials such as carbon and Si3N4 can be also used under slightly different conditions to fabricate nanorods, and these membranes also work properly for the annealing cycle within the range of temperatures selected in our experiments. We also note that no unexpected or unusually high safety hazards were encountered during the entire sample preparation processes.

Characterization and Device Fabrication

In Situ TEM Measurement

Investigation of phase transformation of the nanorod from the intermediate phase to the perovskite phase with the evolution of PbI2 nanoparticles during annealing were carried out using an in situ heating holder (652 double tilt, Gatan) within a high-resolution transmission electron microscope (TEM, Tecnai G2 F30 S-Twin operating at 300 keV, FEI). We note that the typical thickness of the nanorod ranges 10–17 nm, at which thickness we can sufficiently identify the contrasts of the PbI2 particles. In situ TEM interrogation of the stability of the perovskite in the nanorod at a high pressure (2 × 10–2 Torr) was carried out using an in situ TEM gas holder (1300 series, Hummingbird Scientific) with a modified gas flow system. We coated the precursor solution as mentioned above to form intermediate phase nanorods, followed by annealing cycle to quantify the PbI2 evolution, change in morphology, and etc. Real time elemental analysis was performed in STEM and EDX modes using another in situ heating holder (Lightening D7, Dens solutions) which allows us to identify the elements and their compositions of intermediate phase, perovskite phase, and perovskite phase overlapped with PbI2 particles during annealing.

In Situ XRD Measurement

An in situ X-ray diffraction (XRD) measurement of the phase transition was carried out using the synchrotron XRD located at a 5D beamline at Pohang Light Source in South Korea. The X-ray energy was kept at 10 keV (Cu, Kα, λ = 1.5406 Å) by using a double-bounce Si (111) monochromator. We engaged heat to examine the PbI2 precipitation from MAPbI3 in a vacuum of 3 × 10–2 Torr.

AFM measurement for the analysis of PbI2 volume fraction. Atomic force microscopy (AFM) was performed using a Bruker multimode 8 in “tapping” mode to quantify the geometry of the nanorod. 2D and 3D mapping, and line scans were acquired for the rods. The scan rate is 20 Hz and 256 × 256 pixel density using a cantilever with a spring constant of 40 N/m. For the analysis of PbI2 volume fraction, we determined the shape of the nanorod using AFM and assumed the PbI2 particle was spherical.

Thermogravimetric Analysis

Thermogravimetric (TGA) analysis was carried out to measure weight loss of MAI during the annealing process at 9 × 10–7 Torr, a similar pressure to the vacuum in TEM. The ramping rate was 10 °C/min. The TGA model we used is TGA-HP50 manufactured by TA Instruments.

In Situ Analysis of Dielectric and Charge Transport Properties

To investigate the dielectric and charge transport properties of thin films and individual nanorods, respectively, the thin film and nanorod with intermediate phase were spin-coated onto the SiO2/Si substrate using the aforementioned methods (see the first paragraph in the Methods), and then two-terminal devices were fabricated as follows. The Pt electrodes were deposited by FIB for the thin film and nanorod (see the real devices and their schematic illustrations in Figure S14): for the thin film, the channel length is 200 nm, film thickness is 50 nm, and the film width is 200 μm, while for the nanorod, the typical channel length is 2 μm and the rod’s width is 250 nm. The dielectric and charge transport properties of the thin film and nanorod devices were measured in a vacuum sealed chamber whose base pressure is 2 × 10–7 Torr, similar to the vacuum level in TEM, using a semiconductor characterization system (Keithley 4200-SCS).