Reduction of Hysteresis in Hybrid Perovskite Transistors by Solvent-Controlled Growth

The effect of crystallization process speed on the morphology of solution-processed methyl ammonium lead iodide (MAPbI3) thin films is investigated. Crystallization speed is controlled by varying the number of annealing steps, temperature, and resting time between steps. The resting period allows solvent-controlled growth (SCG) in which crystallization progresses slowly via an intermediate phase—during which solvents slowly evaporate away from the films. SCG results in fewer residues, fewer pinholes, and larger grain sizes. Consequently, thin-film transistors with SCG MAPbI3 exhibit smaller hysteresis in their current-voltage characteristics than those without, demonstrating the benefits of SCG toward hysteresis-free perovskite devices.

The morphology of the perovskite thin films plays a crucial role in the stability of the devices. In particular, the defect-mediated drift-diffusion of ions occurs via grain boundary defects [22,23]. As the perovskite thin films are formed by low temperature solution processes, the formation of polycrystalline films with grain boundaries is unavoidable [22,23]. Structural disorder at the grain boundaries results in defect states such as dangling bonds, elemental vacancies, voids, and pinholes. Besides hysteresis, these defects also have the potential to degrade carrier transport properties of the perovskite thin films, especially the carrier mobility. effectively performed oxidative treatments on the surface of the Al 2 O 3 . The O can react with O 2 to form ozone (O 3 ), which decomposes back to ·O and O 2 due to the presence of the 254 nm UV light [33][34][35]. These reactive ·O radicals are highly effective at removing organic residuals from metal oxide film surfaces, increasing the metal-oxygen bonds, and reducing oxygen-related defects. UV treatment thus results in a smooth surface morphology with less defects [33][34][35].
Materials 2021, 14,2573 3 of 13 nm, generated by a mercury vapor lamp). During treatment, the 185 nm UV light decomposed atmospheric oxygen molecules (O2) to oxygen-free radicals (·O), which effectively performed oxidative treatments on the surface of the Al2O3. The O can react with O2 to form ozone (O3), which decomposes back to ·O and O2 due to the presence of the 254 nm UV light [33][34][35]. These reactive ·O radicals are highly effective at removing organic residuals from metal oxide film surfaces, increasing the metal-oxygen bonds, and reducing oxygen-related defects. UV treatment thus results in a smooth surface morphology with less defects [33][34][35]. For film densification, we performed a final anneal for 2 h at 350 °C. We defined the source and drain electrodes by patterning a 60-nm-thick IZO layer deposited at 280 °C via sputtering. To avoid pattering, we spin-coated a positive photoresist (PR) (Sigma-Aldrich, Inc., Steinheim, Germany) layer on top of the source and drain electrodes and patterned it to form "banks" into which we spin-coated the MAPbI3. Several TFTs samples were fabricated using MAPbI3 films grown under different conditions.
We retrieved field-effect mobility (µFE) and subthreshold voltage, respectively, from the slope and intercept of the linear regression line of the |ID| 1/2 (VDS) plot in the saturation regime (VDS > VGS-VTH). VGS, VDS, and ID, denote the gate voltage, drain voltage, and drain current, respectively. We took the subthreshold swing (SS) as the minimum of (d log (ID)/d VGS) −1 and the hysteresis voltage (VHYS) as the difference between VGS values corresponding to ID equal to 1 nA in the forward and reverse sweeps.

Film Morphology
For better adhesion, we treated the surface of the Al2O3 with UV-ozone for 10 min before MAPbI3 deposition ( Figure 2). Surface properties of Al2O3 films with and without UV-ozone treatment were analyzed using atomic force microscopy (AFM) and contact angle (CA) measurements. Removal of organic solvent residues by the UV-ozone treatment decreases surface roughness as root mean square roughness (RRMS) decreased from 0.24 nm to 0.18 nm. This also improves the wettability of the Al2O3 film.
The contact angle of the Al2O3 G.I. decreased from 59° to 50° after UV-ozone treatment (AHTECH LTS., Gyeonggi-do, Korea), indicating higher surface energy ( Figure 2). We calculated the surface energy (γs) of the Al2O3 film before and after UV-ozone treatment using the equation γs = (γw/4) × (1 + cosθ) 2 where θ is the contact angle at equilibrium and γw is the water surface free energy and obtained the values 23.96 and 30.60 mJ/m 2 , respectively. Higher surface energy indicates better interfacial qualities between the MAPbI3 semiconductor and the Al2O3 G.I. It is important to note that without UV-ozone treatment the low wettability of the substrate results in fewer nucleation sites. Although fewer nucleation sites can lead to slightly larger grain sizes, the films without UV-ozone treatment cannot achieve functioning TFTs because they contain a large number of voids and pinholes, which deter lateral conduction in the thin films [36]. For film densification, we performed a final anneal for 2 h at 350 • C. We defined the source and drain electrodes by patterning a 60-nm-thick IZO layer deposited at 280 • C via sputtering. To avoid pattering, we spin-coated a positive photoresist (PR) (Sigma-Aldrich, Inc., Steinheim, Germany) layer on top of the source and drain electrodes and patterned it to form "banks" into which we spin-coated the MAPbI 3 . Several TFTs samples were fabricated using MAPbI 3 films grown under different conditions.
We retrieved field-effect mobility (µ FE ) and subthreshold voltage, respectively, from the slope and intercept of the linear regression line of the |I D | 1/2 (V DS ) plot in the saturation regime (V DS > V GS -V TH ). V GS , V DS , and I D , denote the gate voltage, drain voltage, and drain current, respectively. We took the subthreshold swing (SS) as the minimum of (d log (I D )/d V GS ) −1 and the hysteresis voltage (V HYS ) as the difference between V GS values corresponding to I D equal to 1 nA in the forward and reverse sweeps.

Film Morphology
For better adhesion, we treated the surface of the Al 2 O 3 with UV-ozone for 10 min before MAPbI 3 deposition ( Figure 2). Surface properties of Al 2 O 3 films with and without UV-ozone treatment were analyzed using atomic force microscopy (AFM) and contact angle (CA) measurements. Removal of organic solvent residues by the UV-ozone treatment decreases surface roughness as root mean square roughness (R RMS ) decreased from 0.24 nm to 0.18 nm. This also improves the wettability of the Al 2 O 3 film.  Figure 3 shows the XRD patterns of MAPbI3 films fabricated via one-step and multistep annealing processes. A "+" sign indicates a multistep process in the order of appearance. For instance, 70 + 120 °C stands for a two-step annealing process involving an initial annealing at 70 °C followed by second annealing at 120 °C. Each annealing step lasted for 10 min and the multistep annealing processes involved a 1-h resting time (at room temperature in a vacuum) between steps. culated the surface energy (γ s ) of the Al 2 O 3 film before and after UV-ozone treatment using the equation γ s = (γ w /4) × (1 + cosθ) 2 where θ is the contact angle at equilibrium and γ w is the water surface free energy and obtained the values 23.96 and 30.60 mJ/m 2 , respectively. Higher surface energy indicates better interfacial qualities between the MAPbI 3 semiconductor and the Al 2 O 3 G.I. It is important to note that without UV-ozone treatment the low wettability of the substrate results in fewer nucleation sites. Although fewer nucleation sites can lead to slightly larger grain sizes, the films without UV-ozone treatment cannot achieve functioning TFTs because they contain a large number of voids and pinholes, which deter lateral conduction in the thin films [36]. Figure 3 shows the XRD patterns of MAPbI 3 films fabricated via one-step and multistep annealing processes. A "+" sign indicates a multistep process in the order of appearance. For instance, 70 + 120 • C stands for a two-step annealing process involving an initial annealing at 70 • C followed by second annealing at 120 • C. Each annealing step lasted for 10 min and the multistep annealing processes involved a 1-h resting time (at room temperature in a vacuum) between steps.   Figure 3 shows the XRD patterns of MAPbI3 films fabricated via one-step and multistep annealing processes. A "+" sign indicates a multistep process in the order of appearance. For instance, 70 + 120 °C stands for a two-step annealing process involving an initial annealing at 70 °C followed by second annealing at 120 °C. Each annealing step lasted for 10 min and the multistep annealing processes involved a 1-h resting time (at room temperature in a vacuum) between steps. . X-ray diffraction (XRD) patterns of MAPbI3 films grown via one-step and multistep annealing processes at various temperatures. A "+" sign indicates a multistep process in the order of appearance. For instance, 70 + 120 °C stands for a two-step annealing process involving an initial annealing at 70 °C followed by second annealing at 120 °C. Each annealing step lasted for 10 min and the multistep annealing processes involved a 1-h resting time (at room temperature in a vacuum) between steps.
Compared to that of films annealed at 70, 100, or 120 °C, the ratio of the MAPbI3 peakto-PbI2 peak of the film successively annealed at 70 and 120 °C (70 + 120 °C) was the highest, indicating the superiority of multistep to one-step annealing processes. The MAPbI3 peak-to-PbI2 peak ratio dramatically decreased at temperatures >120 °C for both one-step and multistep annealing processes. It is important to note that multistep processes with a maximum temperature of ≤120 °C (e.g., 70 + 100 + 120 °C) yielded similar results to the 70 + 120 °C process. Additionally, resting time ≥ 1 h resulted in similar results (not shown Figure 3. X-ray diffraction (XRD) patterns of MAPbI 3 films grown via one-step and multistep annealing processes at various temperatures. A "+" sign indicates a multistep process in the order of appearance. For instance, 70 + 120 • C stands for a two-step annealing process involving an initial annealing at 70 • C followed by second annealing at 120 • C. Each annealing step lasted for 10 min and the multistep annealing processes involved a 1-h resting time (at room temperature in a vacuum) between steps.
Compared to that of films annealed at 70, 100, or 120 • C, the ratio of the MAPbI 3 peak-to-PbI 2 peak of the film successively annealed at 70 and 120 • C (70 + 120 • C) was the highest, indicating the superiority of multistep to one-step annealing processes. The MAPbI 3 peak-to-PbI 2 peak ratio dramatically decreased at temperatures >120 • C for both one-step and multistep annealing processes. It is important to note that multistep processes with a maximum temperature of ≤120 • C (e.g., 70 + 100 + 120 • C) yielded similar results to the 70 + 120 • C process. Additionally, resting time ≥ 1 h resulted in similar results (not shown here). As we deposited the MAPbI 3 in ambient air, oxygen in the air also strongly influenced the crystal quality. Previous studies have shown the presence of oxygen during annealing to reduce defects without affecting the surface morphology [37,38]. Moreover, it is suggested that oxygen reduces grain boundary defects, yielding better grain-to-grain connection, which is essential for good lateral conduction [24]. The presence of oxygen in the films ( Figure 4) was consistent with defect passivation during ambient air processing [21,24,25,38,39] and evaporation of MA via oxygen bonding at temperatures greater than 120 • C ( Figure 3). here). As we deposited the MAPbI3 in ambient air, oxygen in the air also strongly influenced the crystal quality. Previous studies have shown the presence of oxygen during annealing to reduce defects without affecting the surface morphology [37,38]. Moreover, it is suggested that oxygen reduces grain boundary defects, yielding better grain-to-grain connection, which is essential for good lateral conduction [24]. The presence of oxygen in the films (Figure 4) was consistent with defect passivation during ambient air processing [21,24,25,38,39] and evaporation of MA via oxygen bonding at temperatures greater than 120 °C (Figure 3). The 70 + 120 °C annealing process was superior due to the existence of solvent-controlled growth (SCG), which occurred during the resting period. For SCG to occur, crystal seeds are essential, and their formation requires thermal energy [40]. Experimentally, we found temperatures around 70 °C to be sufficient for the seed formation. Given the high boiling points of GBL and DMSO (204 and 189 °C, respectively) [24], high-temperature annealing is required after SCG to effectively complete crystallization. This temperature should be around 120 °C, according to Figure 3. As multistep processes with a maximum temperature of ≤120 °C yielded similar results, we chose the two-step annealing process (70 + 120 °C) for TFT fabrication to keep the process simple. Figure 5a,b depicts crystallization process mechanisms during one-step and two-step annealing processes, which respectively represent fast and slow crystallization. In the case of the one-step annealing at 120 °C the color of the thin films turned blackish immediately after placing them on the hotplate (see real image in Figure 5a). This indicates rapid growth of MAPbI3 poly crystals with a bandgap of ≅1.72 eV. As air bubbles can be easily trapped during processing in ambient air, rapid solvent evaporation may result in pinhole formation. Additionally, excessive thermal energy may result in MA evaporating with the solvents or via oxygen bonding (Figure 4), leading to void formation in the film ( Figure  5a). Extension of the annealing time or additional annealing steps at higher temperatures only results in additional evaporation of MA, leaving behind PbI2. MA is prone to evaporation due to its low boiling point and low sticking properties.
In the case of the two-step annealing process, crystallization progresses slowly via an intermediate state (Figure 5b). The XRD pattern of the 70 °C film shows that this intermediate phase was composed of both MAPbI3 (CH3NH3PbI3) and methyl ammine lead iodide (CH3)2NH2PbI3 crystals (Figure 3). During the intermediate phase, the film was very transparent, so much so that the cutting board under it was clearly visible (grid lines in real image), which further suggests the incomplete formation of MAPbI3. When the samples were left to rest for the duration of 1 h or longer after the 70 °C annealing, crystallization progressed slowly as solvents slowly evaporated from the film. Upon annealing at 120 °C after the resting period, (CH3)2NH2PbI3 disappeared and only the MAPbI3 peak appeared strong, indicating complete crystallization (Figure 3, 70 + 120 °C). The 70 + 120 • C annealing process was superior due to the existence of solventcontrolled growth (SCG), which occurred during the resting period. For SCG to occur, crystal seeds are essential, and their formation requires thermal energy [40]. Experimentally, we found temperatures around 70 • C to be sufficient for the seed formation. Given the high boiling points of GBL and DMSO (204 and 189 • C, respectively) [24], high-temperature annealing is required after SCG to effectively complete crystallization. This temperature should be around 120 • C, according to Figure 3. As multistep processes with a maximum temperature of ≤120 • C yielded similar results, we chose the two-step annealing process (70 + 120 • C) for TFT fabrication to keep the process simple. Figure 5a,b depicts crystallization process mechanisms during one-step and two-step annealing processes, which respectively represent fast and slow crystallization. In the case of the one-step annealing at 120 • C the color of the thin films turned blackish immediately after placing them on the hotplate (see real image in Figure 5a). This indicates rapid growth of MAPbI 3 poly crystals with a bandgap of ∼ =1.72 eV. As air bubbles can be easily trapped during processing in ambient air, rapid solvent evaporation may result in pinhole formation. Additionally, excessive thermal energy may result in MA evaporating with the solvents or via oxygen bonding (Figure 4), leading to void formation in the film (Figure 5a). Extension of the annealing time or additional annealing steps at higher temperatures only results in additional evaporation of MA, leaving behind PbI 2 . MA is prone to evaporation due to its low boiling point and low sticking properties.
In the case of the two-step annealing process, crystallization progresses slowly via an intermediate state (Figure 5b). The XRD pattern of the 70 • C film shows that this intermediate phase was composed of both MAPbI 3 (CH 3 NH 3 PbI 3 ) and methyl ammine lead iodide (CH 3 ) 2 NH 2 PbI 3 crystals (Figure 3). During the intermediate phase, the film was very transparent, so much so that the cutting board under it was clearly visible (grid lines in real image), which further suggests the incomplete formation of MAPbI 3 . When the samples were left to rest for the duration of 1 h or longer after the 70 • C annealing, crystallization progressed slowly as solvents slowly evaporated from the film. Upon annealing at 120 • C after the resting period, (CH 3 ) 2 NH 2 PbI 3 disappeared and only the MAPbI 3 peak appeared strong, indicating complete crystallization (Figure 3, 70 + 120 • C). SEM images of the one-step and two-step annealed films confirmed this mechanism. The film annealed at 120 °C contained many voids and/or pinholes as a consequence of the rapid growth (Figure 6a). Fewer voids existed in the 70 + 120 °C film (Figure 6b) as most of the solvents slowly evaporated during the resting period, thereby minimizing the rapid loss of precursors during the second 120 °C annealing step. Furthermore, rapid crystallization from numerous crystal seeds led to smaller grain size in the 120 °C film than 70 + 120 °C film. If the films are left to rest after spin-coating without providing sufficient heat energy through annealing at 70 °C, the intermediate phase will not occur (compare 120 °C and 27 + 120 °C films in Figure 3). Therefore, the first step of annealing at 70 °C is necessary to provide sufficient energy for the crystal seed formation.
Similar results were observed for the inorganic perovskite CsPbI3 where a δ-phase CsPbI3 was observed upon resting the precursor films in a nitrogen-filled glovebox for several tens of minutes [41]. The formation of this additional phase resulted in films becoming more continuous and having a reduced number of pinholes and larger gain size. Additionally, solar cells employing the CsPbI3 films obtained by SCG displayed higher efficiency and could tolerate above 500 h of continuous light soaking [41].
In addition to the crystallization speed, thermal stress also plays a role in the morphology of perovskite films. Perovskite films can accumulate residual stress during processing. Initially, the precursor solution moves freely without any constraint during the spin-coating process due to its low viscosity and lack of interaction with the substrate. SEM images of the one-step and two-step annealed films confirmed this mechanism. The film annealed at 120 • C contained many voids and/or pinholes as a consequence of the rapid growth (Figure 6a). Fewer voids existed in the 70 + 120 • C film (Figure 6b) as most of the solvents slowly evaporated during the resting period, thereby minimizing the rapid loss of precursors during the second 120 • C annealing step. Furthermore, rapid crystallization from numerous crystal seeds led to smaller grain size in the 120 • C film than 70 + 120 • C film. If the films are left to rest after spin-coating without providing sufficient heat energy through annealing at 70 • C, the intermediate phase will not occur (compare 120 • C and 27 + 120 • C films in Figure 3). Therefore, the first step of annealing at 70 • C is necessary to provide sufficient energy for the crystal seed formation.
Similar results were observed for the inorganic perovskite CsPbI 3 where a δ-phase CsPbI 3 was observed upon resting the precursor films in a nitrogen-filled glovebox for several tens of minutes [41]. The formation of this additional phase resulted in films becoming more continuous and having a reduced number of pinholes and larger gain size. Additionally, solar cells employing the CsPbI 3 films obtained by SCG displayed higher efficiency and could tolerate above 500 h of continuous light soaking [41].
that of the substrate [42]. This is the case for halide perovskites that have a CTE (approximately 50 × 10 −6 K −1 ) that is five times larger than that of glass (approximately 10 × 10 −6 K −1 ) [43,44]. When the samples were cooled to room temperature after annealing, the perovskite film was constrained from contraction by the underlying glass substrate, resulting in tensile stress building up in the perovskite film. Note that the layer immediately below the perovskite was a 40-nm-thick Al2O3 layer, which was much thinner than the glass (approximately 0.6 mm). The stress was thus determined largely by the glass. Rolston et al. found the residual stress to have a linear relationship with the temperature at which the film is annealed with higher annealing temperatures resulting in larger stress values [42]. They attributed the correlation between the residual stress and annealing temperature largely to the CTE mismatch between the film and substrate and calculated the predicted stress, σΔT, by where Ep is the perovskite modulus, νp is Poisson's ratio of the perovskite, αs and αp are the respective thermal expansions of the substrate and perovskite, and ΔT is the temperature gradient when cooling. Using an Ep of approximately 12.5 GPa and νp = 0.3 [42], σΔT is approximately 31 and 65 MPa, respectively, for annealing at 70 and 120 °C. Therefore, the one-step annealing process (120 °C) resulted in initial residual film stress that was much larger than that of the two-step annealing process (70 + 120 °C). Although the twostep annealed sample also underwent annealing at 120 °C, the stress was expected to be less due to the presence of the intermediate phase. Residual film stress can have a significant effect on the optoelectronic properties of the perovskite films and contribute to heat, light, and moisture-based chemical degradation as well as fractures in the device [45,46]. In addition to the crystallization speed, thermal stress also plays a role in the morphology of perovskite films. Perovskite films can accumulate residual stress during processing. Initially, the precursor solution moves freely without any constraint during the spin-coating process due to its low viscosity and lack of interaction with the substrate. However, when the solvent evaporates during the annealing process, conversion into the crystalline perovskite structure enhances chemical interactions with the substrate. The film is thus constrained from expanding or contracting freely, leading to film stress. This constraint is large when the thermal expansion coefficient (CTE) of the film is larger than that of the substrate [42]. This is the case for halide perovskites that have a CTE (approximately 50 × 10 −6 K −1 ) that is five times larger than that of glass (approximately 10 × 10 −6 K −1 ) [43,44]. When the samples were cooled to room temperature after annealing, the perovskite film was constrained from contraction by the underlying glass substrate, resulting in tensile stress building up in the perovskite film. Note that the layer immediately below the perovskite was a 40-nm-thick Al 2 O 3 layer, which was much thinner than the glass (approximately 0.6 mm). The stress was thus determined largely by the glass.
Rolston et al. found the residual stress to have a linear relationship with the temperature at which the film is annealed with higher annealing temperatures resulting in larger stress values [42]. They attributed the correlation between the residual stress and annealing temperature largely to the CTE mismatch between the film and substrate and calculated the predicted stress, σ ∆T , by where E p is the perovskite modulus, ν p is Poisson's ratio of the perovskite, α s and α p are the respective thermal expansions of the substrate and perovskite, and ∆T is the temperature gradient when cooling. Using an E p of approximately 12.5 GPa and ν p = 0.3 [42], σ ∆T is approximately 31 and 65 MPa, respectively, for annealing at 70 and 120 • C. Therefore, the one-step annealing process (120 • C) resulted in initial residual film stress that was much larger than that of the two-step annealing process (70 + 120 • C). Although the two-step annealed sample also underwent annealing at 120 • C, the stress was expected to be less due to the presence of the intermediate phase. Residual film stress can have a significant effect on the optoelectronic properties of the perovskite films and contribute to heat, light, and moisture-based chemical degradation as well as fractures in the device [45,46].

TFT Current-Voltage Characteristics
Transfer, diode (saturation), and output characteristics of the TFTs were measured as shown in Figure 7. It is interesting to see that hysteresis in the transfer characteristics was smaller in TFTs with SCG MAPbI 3 than those without (Figure 7a,b). Hysteresis is due to vacancy-mediated migration of ions in the MAPbI 3 [15][16][17]. Given that most of these vacancies occur in grain boundaries or around voids, reduction of hysteresis is consistent with the larger grain size and fewer voids in the SCG-based MAPbI 3 .

TFT Current-Voltage Characteristics
Transfer, diode (saturation), and output characteristics of the TFTs were measured as shown in Figure 7. It is interesting to see that hysteresis in the transfer characteristics was smaller in TFTs with SCG MAPbI3 than those without (Figure 7a,b). Hysteresis is due to vacancy-mediated migration of ions in the MAPbI3 [15][16][17]. Given that most of these vacancies occur in grain boundaries or around voids, reduction of hysteresis is consistent with the larger grain size and fewer voids in the SCG-based MAPbI3. Hysteresis can also originate from ferroelectric effects, trapping at the G.I./MAPbI3 interface, or ion migration inside the G.I. Given that the two sets of TFTs were fabricated with an identical G.I. process, the difference in their hysteresis characteristics is mainly due to changes in the MAPbI3 and not the aforementioned. I − , CH3NH3 + , and Pb2 + are the three ions in MAPbI3, and the activation energy for vacancy-mediated migration is the lowest for I − (0.1-0.6, 0.46-0.84, and >2 eV, respectively) [47][48][49]. I − was thus believed to dominate ion drift-diffusion in MAPbI3. Reduced hysteresis in SCG MAPbI3 TFTs was Hysteresis can also originate from ferroelectric effects, trapping at the G.I./MAPbI 3 interface, or ion migration inside the G.I. Given that the two sets of TFTs were fabricated with an identical G.I. process, the difference in their hysteresis characteristics is mainly due to changes in the MAPbI 3 and not the aforementioned. I − , CH 3 NH 3 + , and Pb 2 + are the three ions in MAPbI 3 , and the activation energy for vacancy-mediated migration is the lowest for I − (0.1-0.6, 0.46-0.84, and >2 eV, respectively) [47][48][49]. I − was thus believed to dominate ion drift-diffusion in MAPbI 3 . Reduced hysteresis in SCG MAPbI 3 TFTs was thus consistent with reduced drift-diffusion of I − due to the effective reduction in the total number of defects and grain boundaries. These results were consistent with a previous report on lead-free layered perovskite transistors where grain boundary passivation and grain size enlargement were found to reduce hysteresis and increase TFT reproducibility and reliability [50].
The diode characteristics measured by shorting the gate and drain electrodes showed a steeper slope for TFTs with SCG than those without (Figure 7c,d). The steeper slope is an indication of higher µ FE as shown in Figure 8. The output characteristics also showed about an order of magnitude higher current in TFTs with SCG than those without, consistent with better film morphology for TFTs with SCG (Figure 7e,f). The current crowding at low V DS values was also more pronounced in the TFTs without SCG than those with SCG, indicating higher resistance at source/drain contacts of the former-possibly due to the presence of many voids. thus consistent with reduced drift-diffusion of I − due to the effective reduction in the total number of defects and grain boundaries. These results were consistent with a previous report on lead-free layered perovskite transistors where grain boundary passivation and grain size enlargement were found to reduce hysteresis and increase TFT reproducibility and reliability [50]. The diode characteristics measured by shorting the gate and drain electrodes showed a steeper slope for TFTs with SCG than those without (Figure 7c,d). The steeper slope is an indication of higher µFE as shown in Figure 8. The output characteristics also showed about an order of magnitude higher current in TFTs with SCG than those without, consistent with better film morphology for TFTs with SCG (Figure 7e,f). The current crowding at low VDS values was also more pronounced in the TFTs without SCG than those with SCG, indicating higher resistance at source/drain contacts of the former-possibly due to the presence of many voids. Owing to the good film morphology achieved by SCG, the µFE improved from 0.02 ± 0.01 to 0.15 ± 0.08 cm 2 /V·s (Figure 8a,b). The VTH values were 6.1 ± 0.9 and 5.2 ± 1.1 V, respectively, for TFTs with and without SCG (Figure 8c,d). The slightly lower VTH in TFTs without SCG could be related to a slightly different surface potential due to defects or Owing to the good film morphology achieved by SCG, the µ FE improved from 0.02 ± 0.01 to 0.15 ± 0.08 cm 2 /V·s (Figure 8a,b). The V TH values were 6.1 ± 0.9 and 5.2 ± 1.1 V, respectively, for TFTs with and without SCG (Figure 8c,d). The slightly lower V TH in TFTs without SCG could be related to a slightly different surface potential due to defects or mobile ions. There were evidently more defect states in the TFTs without SCG than those with SCG, given that SS of the former (1.28 ± 0.19 V/dec) was approximately ten times larger than that of the latter (0.17 ± 0.03 V/dec) (Figure 8e,f). The total trap density (N T ) can be calculated from N T = {SS(log(e))/(k B T/q) − 1}(C ox /q) where k B , T, and q are respectively the elementary electric charge, Boltzmann's constant, and temperature in kelvins [51]. N T values of the TFTs with and without SCG were approximately 1.29 × 10 12 and 1.06 × 10 13 cm −2 ·V −1 , respectively [51,52]. The difference in N T was mainly due to the bulk states, given that the TFTs employ the same gate insulator.
Positive gate bias stress (PBS) stability of the MAPbI 3 TFTs was also investigated. The PBS was achieved by applying a constant V GS of 10 V to the TFTs while grounding the source and drain electrodes for 3600 s. The PBS was interrupted at different points to measure the transfer characteristics as shown in Figure 9. mobile ions. There were evidently more defect states in the TFTs without SCG than those with SCG, given that SS of the former (1.28 ± 0.19 V/dec) was approximately ten times larger than that of the latter (0.17 ± 0.03 V/dec) (Figure 8e,f). The total trap density (NT) can be calculated from NT = {SS(log(e))/(kBT/q) − 1}(Cox/q) where kB, T, and q are respectively the elementary electric charge, Boltzmann's constant, and temperature in kelvins [51]. NT values of the TFTs with and without SCG were approximately 1.29 × 10 12 and 1.06 × 10 13 cm −2 ·V −1 , respectively [51,52]. The difference in NT was mainly due to the bulk states, given that the TFTs employ the same gate insulator. Positive gate bias stress (PBS) stability of the MAPbI3 TFTs was also investigated. The PBS was achieved by applying a constant VGS of 10 V to the TFTs while grounding the source and drain electrodes for 3600 s. The PBS was interrupted at different points to measure the transfer characteristics as shown in Figure 9. Consistent with having fewer defects, the TFTs with SCG showed better PBS stability compared to those without. PBS results in trapping of carriers at the semiconductor/gateinsulator interface. These trapped carriers screened the gate field, resulting in the reduction of the effective applied VGS, which led to positive the turn-on voltage shift (ΔVON). However, the shifting of the VON of the MAPbI3 TFTs investigated herein was bidirectional, consistent with the contribution of other forms of hysteresis, particularly due to ion migration. The fluctuation of the VTH, especially at the early stages of the PBS, was much larger for the TFT without SCG-underscoring the importance of defect reduction by SCG. The similar ΔVON after 3600 s can be explained by the fact that the TFTs employed the same gate insulator.
The VHSY decreased from 1.46 ± 0.58 to 0.17 ± 0.03 V after SCG (Figure 8g,h). Overall, SCG also resulted in less variation in the extracted parameters, indicating better film uniformity. However, even after SCG, the hysteresis was still quite significant, especially in the off-state characteristics (Figure 7b). Additionally, there is still room to optimize the Consistent with having fewer defects, the TFTs with SCG showed better PBS stability compared to those without. PBS results in trapping of carriers at the semiconductor/gateinsulator interface. These trapped carriers screened the gate field, resulting in the reduction of the effective applied V GS , which led to positive the turn-on voltage shift (∆V ON ). However, the shifting of the V ON of the MAPbI 3 TFTs investigated herein was bidirectional, consistent with the contribution of other forms of hysteresis, particularly due to ion migration. The fluctuation of the V TH , especially at the early stages of the PBS, was much larger for the TFT without SCG-underscoring the importance of defect reduction by SCG. The similar ∆V ON after 3600 s can be explained by the fact that the TFTs employed the same gate insulator.
The V HSY decreased from 1.46 ± 0.58 to 0.17 ± 0.03 V after SCG (Figure 8g,h). Overall, SCG also resulted in less variation in the extracted parameters, indicating better film uniformity. However, even after SCG, the hysteresis was still quite significant, especially in the off-state characteristics (Figure 7b). Additionally, there is still room to optimize the ther-mal treatment employed herein and further improvement can be achieved by employing additional strategies to suppress the formation of vacancies and ion migration such as the incorporation of multi cation mixtures and Lewis base acid treatments [53][54][55]. Additionally, various surface treatments such as the insertion of SAMs [16] and the use of additives to passivate surface defects [24][25][26][27][28][29] can be applied to further mitigate the difference in the surface energy or roughness between the G.I. and the perovskite semiconductor to achieve smoother pinhole-free thin films.

Conclusions
The effect of crystallization process speed on the morphology of MAPbI 3 thin films processed from a solution in ambient air was investigated. The crystallization speed can be controlled by varying the number of annealing steps, temperature, and resting time between steps. The resting period allows solvent-controlled growth (SCG) in which crystallization progresses slowly via an intermediate phase during which solvents slowly evaporate away from the films. SCG decreases the total number of defects and grain boundaries in MAPbI 3 -thereby significantly reducing hysteresis in MAPbI 3 TFTs.
To achieve SCG, a two-step annealing process is required. It is recommended to do the first step at approximately 70 • C followed by resting for at least one hour before the second step at approximately 120 • C. Adding steps between the 70 and 120 • C annealing or increasing the resting time does not significantly yield better results. Annealing at temperatures larger than 120 • C degrades the films due to the evaporation of MAI whereas annealing at temperatures lower than 70 • C is not sufficient for crystallization. A one-step annealing at 120 • C results in rapid crystallization from numerous crystal seeds, which leads to smaller grain size and voids in the film. This work demonstrates a simple strategy to obtain hysteresis-free perovskite devices in ambient air and thus opens up the scope of using organic-inorganic hybrid perovskites for applications beyond optoelectronic devices.