The unusual interaction of DMF and DMSO solvent with perovskite first came to our attention when we observed the reversible transformation of MAPbI₃ films under DMF or DMSO vapor. Exposure to the vapor of either of these solvents causes the black perovskite films to become colorless and translucent. Similar exposure to other organic solvents had little or no effect. Annealing the colorless films causes the black perovskite phase to re-appear, but with a significantly altered film morphology (Figure S2). X-ray diffraction (XRD) measurements confirm that both the initial and final black films are MAPbI₃ (Figure S3). In solution phase, multiple transformations between intermediate solvent-intercalated complexes of MAPbI₃ in DMF and DMSO have been carefully elucidated by Nakamura et al.²⁰ In our work, we are concerned with elaborating the solvent-perovskite interplay in the solid state, and controlling the crystallinity and morphology of these transparent intermediate films to yield flat and dense perovskite layers.

First, in order to characterize the crystal structure and thermal characteristics of the transparent compounds, bulk samples of the solvent-intercalated materials were prepared under rapid timescales approximating the thin film growth conditions.

PbI₂ (461 mg, 1.0 mmol) and MAI (159 mg, 1.0 mmol) were dissolved in 1.0 mL of DMF, DMSO, or GBL. A few drops of each solution were placed at the center of a slide glass placed on a hot plate at 100, 120, and 120 °C for DMF, DMSO, and GBL solution, respectively. (Videos are available as part of the Supporting Information.). As shown in Figure 1 for the case of GBL, cubic, black perovskite solid crystallizes out of solution directly — the expected result. For DMSO and DMF, however, long, colorless, needle-like crystals formed initially. Only after additional heating did the crystals transform into black MAPbI₃. The intermediate compounds were isolated, and washed with acetonitrile to remove residual solvent prior to characterization.

Figure 1. Photographs of crystal growth process upon heating of several precursor solutions. MAI + PbI₂ (1:1) in GBL, DMF, and DMSO. The circular images are about 3 cm wide. Videos are available as part of the Supporting Information.

Single-crystal X-ray structure analysis revealed that precursor compounds dissolved in DMF and DMSO form MA₂Pb₃I₈·2DMF and MA₂Pb₃I₈·2DMSO intermediates, respectively (Figure 2). In both cases the solvent molecule is intercalated between a planar [Pb₃I₈]²⁻ layer and the MA⁺ cation. Note that neither complex maintains a perovskite stoichiometry: Both are deficient by one MAI formula unit for every three PbI₂ units, unlike the MAPbI₃·DMF complex we reported previously.^21,22 The different structure most likely arises from the faster rate of crystallization in the present work.

Figure 2. X-ray crystal structures and TGA curves of (top) MA₂Pb₃I₈·2DMF and (bottom) MA₂Pb₃I₈·2DMSO.

The loss of solvent molecules from the bulk intermediates was monitored by thermogravimetric analysis (TGA) as shown in Figure 2. Solvent loss begins at 77 °C and 73 °C, for MA₂Pb₃I₈·2DMF and MA₂Pb₃I₈·2DMSO, respectively. The total weight loss of 8.1 wt% and 8.5 wt% is in good agreement with the theoretical values for complete solvent extraction, 7.9 wt% and 8.4 wt%. We note in passing that MA₂Pb₃I₈·2DMF shows different behavior from MAPbI₃·DMF. In the latter case solvent loss began at a much lower temperature, around 58 °C.^21,22

Since solvent engineering protocols frequently make use of mixed DMF/DMSO solvents, it is important to determine the composition and structure of the solvent-intercalated intermediate complexes for different solvent ratios.^23–25 To examine this, precursor solutions of 1.0 M MAI and 1.0 M PbI₂ were made up with DMF:DMSO ratios from 1:0 to 0:1. The results, confirmed by X-ray structural analysis, are compiled in Table 1. MA₂Pb₃I₈·2DMSO formed in DMF-rich solvents up to 3:1 DMF:DMSO, demonstrating the clear preference for the intercalation of DMSO over DMF in the intermediate complexes. Only DMSO and DMF single crystal complexes were isolated, no mixed solvent complexes were observed.

Table 1. Intermediate Complexes Formed from MAI and PbI2 in Mixed Solvents

We now turn to examining the properties of the intermediate MA₂Pb₃I₈·2DMSO in the solid thin film state, including the influence of the film fabrication method, the morphology and orientation of the crystallites, and the details of the thermal conversion to MAPbI₃ perovskite. Attention is drawn to the difference in stoichiometry between MA₂Pb₃I₈·2DMSO and the annealed perovskite, MAPbI₃. This complex is deficient in one MAI formula per three PbI₂ units, so it is important to consider whether complete conversion of the film to MAPbI₃ is achieved, and what side products may be formed.

The thin films studied in the following sections were prepared from precursor solutions of 1.55 M MAI and PbI₂ (1:1) dissolved in 3:1 DMF:DMSO. The 3:1 DMF:DMSO solvent ratio was found to give an optimal balance between good solubility of the precursor materials and long drying times of the intermediate films.²¹ The precursor solution was spin-coated on a mesoporous TiO₂ substrate, and toluene antisolvent was dripped over the spinning substrate for 1 s starting precisely 8 s after the 5000 rpm spin program started, and just before the spinning stopped. Full details of the fabrication process are given in the experimental section and illustrated schematically in Figure S1.

By inducing rapid precipitation of the intermediate crystallites the toluene antisolvent plays a critical role in determining the flatness of the intermediate film, and therefore the morphology of the final perovskite layer as well. Scanning electron microscope (SEM) images of the films before annealing reveal that toluene antisolvent addition facilitates the formation of needle-shaped microcrystals approximately 50 nm wide and 500 nm long (Figure 3a). When the toluene dripping was omitted and the films were allowed to dry naturally, the films were opaque or translucent (Figure 3b), rather than transparent. The intermediate crystals were found to be much larger, over 500 nm in width and to over a micrometer in length. With antisolvent, a black perovskite film with a flat and uniform surface was observed after annealing. Without antisolvent, the color of the film changed to dark-brown while retaining the rough morphology of the intermediate film. Such a rough and granular film is not suitable for efficient solar cells. Thus, antisolvent is necessary to control the crystal size of the intermediate complex, MA₂Pb₃I₈·2DMSO, and ensure full conversion of the intermediate film to perovskite. In controlling the crystal size, and, as we will demonstrate later, the orientation of the needles with respect to the substrate, the antisolvent addition is directly responsible for forming dense and flat perovskite layers.

Figure 3. Top-view SEM images and XRD patterns of the films fabricated by (a) with or (b) without toluene dripping before and after annealing. Inserted photographs show a top view of each film. Asterisks indicate peaks of FTO substrate. Black arrows in (b) imply unconverted MA₂Pb₃I₈·2DMSO.

The XRD measurements shown in Figure 3 confirm that the MA₂Pb₃I₈·2DMSO intermediate film precipitated with the aid of toluene antisolvent is completely converted to perovskite on annealing. While the desirable result, it is surprising, considering that as a result of the 2:3:8 stoichiometry of MA:Pb:I, there is insufficient MAI present within the intermediate crystals for full conversion. Given that our precursor solution contained MAI and PbI₂ in exactly 1:1 ratio, it seems reasonable to speculate that residual, amorphous MAI exists in the dried film and is taken up into the perovskite during thermal conversion. However, the possibility of balancing quantities of amorphous PbI₂ cannot be ruled out as an alternate explanation. Regardless of which scenario is correct, the key point is that, for the intermediate film prepared without antisolvent, the MA₂Pb₃I₈·2DMSO diffraction peaks remain after annealing, indicating that the films have only partially converted to perovskite. This is reasonable considering that in the bulk experiment described earlier, only the surface of the intermediate complex transformed to perovskite after annealing. Antisolvent addition therefore ensures full conversion of the intermediate material to perovskite, again through control of the crystallite size.

The perovskite film morphology is also likely influenced by the orientation of the intermediate crystallites. The difference in the relative peak intensities between the simulation and the observed XRD pattern of MA₂Pb₃I₈·2DMSO intermediate indicates the presence of a preferred crystal orientation in the as-spun film. Single-crystal X-ray structural analysis confirmed that the long axis of the needle-shaped MA₂Pb₃I₈·2DMSO crystal corresponds to the [2 –1 1] direction, which coincides closely with the direction of the Pb-I network, as shown in Figure 4a. Cross-sectional (Figure 4b) SEM images of the as-spun precursor film show that the individual needle-shaped crystals tend to stack with their long axes nearly parallel to the substrate. The effect of the preferred orientation was evaluated by Rietveld refinement of the XRD pattern using March-Dollase function²⁶ in the RIETAN-FP software.²⁷ The pattern could be reproduced by setting the preferred-orientation vector to the [2 –1 1] direction, and the March distribution parameter r converged to 5.52(7). Figure 4c shows the dependence of the relative peak intensities on the value r. Generally, r = 1 represents the absence of any preferred-orientation effect, and r > 1 (<1) represents the preferred orientation vector being parallel (perpendicular) with respect to the substrate. Thus, it can be concluded that the MA₂Pb₃I₈·2DMSO crystal has a preferred orientation with its long-axis direction (Pb–I network direction) in the substrate surface plane. The crystal growth of the perovskite most likely preferentially proceeds following the same horizontal directions, filling voids in the film rather than creating roughness-inducing stacks or whiskers perpendicular to the substrate. Thus, the horizontal orientation of the intermediate crystallites may be an important additional factor to ensure the growth of a smooth perovskite layer. (Figure 4d)

Figure 4. (a) Photograph of the mounted MA₂Pb₃I₈·2DMSO crystal showing the orientation of the crystal planes. (b) Cross-sectional SEM image of the as-spun film with toluene dripping. (c) Rietveld refinement of the intermediate phase. XRD patterns of as-spun film (blue line), several calculated XRD profiles (black line), and their difference (green line). The large r value indicates that the crystals align parallel to the substrate. (d) Illustration of the formation of aligned MA₂Pb₃I₈·2DMSO crystallites on the substrate during spin coating.

Examining the TGA results of the bulk intermediate crystals as shown in Figure 2, it might be expected that solvent loss might begin around 70 °C. Instead, for the intermediate film samples prepared with toluene antisolvent dripping, significant color change was observed after heating up to 55 °C. In the film, both solvent loss and the transformation to perovskite crystal structure occurs below that of the bulk solid.

The various stages of film growth were explored by heating the intermediate film on a hot plate in a nitrogen-filled glove box and stopping the annealing at different points on a temperature ramp to a maximum of 100 °C. Figure 5 shows top-view SEM images at each of the holding temperatures. Prior to annealing, the intermediate layer is made up of MA₂Pb₃I₈·2DMSO microcrystals about 50 nm wide (Figure 5a). During the first heating stage at 40 °C, the films began to change from colorless and transparent to brown, indicating the onset of conversion to perovskite (Figure 5b). The conversion onset is significantly lower than for the bulk MA₂Pb₃I₈·2DMSO crystals (see Figure 2). The difference in temperature is almost certainly due to the much smaller crystal sizes present in the intermediate film. We found that 55 °C is the critical temperature where the intermediate complex fully changes to the deep black typical of perovskite (Figure 5c). Despite the apparent complete conversion, we also found the overall crystal size increased and the number of grain boundaries decreased during additional “ripening” at around 75 °C (Figure 5d). Finally, a finishing step at 100 °C creates large grains which extend across the full height of the film (Figure 5f). The SEM results clearly show that multiple processes occur when the film is annealed in a stepwise manner over a range of temperature from 55 °C–100 °C.

Figure 5. Top view SEM images of (a) the as-spun perovskite layer with toluene antisolvent dripping, and of similar films after annealing at (b) 40 °C, (c) 55 °C, and (d) 75 °C. For 100 °C, the final annealing temperature, (e) top view and (f) cross-sectional SEM images are given. Photographs are inset in each top view image to show the corresponding change in color/transparency.

When the as-spun film was heated immediately to 100 °C for 30 min, the resulting perovskite layer had poor uniformity and contained many voids (Figure S4). This generally resulted in moderate device performance with lower average V_OC and FF (Table S1). As seen in Figure 5, the stepwise-annealed films showed significantly higher quality. Based on the above results, a stepwise temperature ramp with fixed holding times: 40 °C for 5 min, 55 °C for 5–15 min, 75 °C for 5 min, and 100 °C for 30 min, was adopted as our standard, optimized annealing process for converting the MA₂Pb₃I₈·2DMSO intermediate to perovskite layers.

To show the advantage of our method of fabricating dense and flat perovskite layers, glass-FTO/compact-TiO₂ (20 nm)/mesoporous-TiO₂ (∼100 nm)/MAPbI₃ (480 nm)/2,2′,7,7′-tetrakis(N,N-di-p-methoxyphenylamine)-9,9′-spirobifluorene (spiro-OMeTAD) (200 nm)/Au (80 nm) solar cell devices were fabricated (Figure 6a) and characterized. The cross-sectional SEM image is shown in Figure 6b. Large perovskite grains can be readily observed which span the distance between charge transporting layers. These grains are not single crystals, however, showing instead multiple crystal domains according to high resolution transmission electron microscopy (HRTEM) (Figure S5) as noted by Kim et al,²⁸ with the measurement area constrained to 0.1 cm² using a shadow mask. The forward scan J-V curve of the best performing device is shown in Figure 6c, with J_SC = 23.4 mA cm⁻², V_OC = 1.16 V, and FF = 0.75. The PCE is 20.3%. For a MAPbI₃ perovskite device, with a typical structure using commonly available charge transport layers, this recorded open circuit voltage and efficiency are exceptionally good results. In addition, our fabrication method yielded highly reproducible devices (average PCE = 19.1 ± 0.6%). The best and average photovoltaic parameters are summarized in Table 2, and the incident photon to current conversion efficiency (IPCE) is shown in Figure 6d. These devices were confirmed to have similar stability to those reported in our related paper.²²

Figure 6. (a) Device structure and (b) cross-sectional SEM image of the perovskite solar cells. (c) Best performing J–V curves (forward scan: red, reverse scan: blue) and (d) IPCE spectrum (black line) together with integrated J_SC (red line). The integrated J_SC from the IPCE measurement was 23.2 mA cm⁻².

Table 2. Photovoltaic Parameters of the Perovskite Solar Cells