Gas-solid reaction based over one-micrometer thick stable perovskite films for efficient solar cells and modules

Besides high efficiency, the stability and reproducibility of perovskite solar cells (PSCs) are also key for their commercialization. Herein, we report a simple perovskite formation method to fabricate perovskite films with thickness over 1 μm in ambient condition on the basis of the fast gas−solid reaction of chlorine-incorporated hydrogen lead triiodide and methylamine gas. The resultant thick and smooth chlorine-incorporated perovskite films exhibit full coverage, improved crystallinity, low surface roughness and low thickness variation. The resultant PSCs achieve an average power conversion efficiency of 19.1 ± 0.4% with good reproducibility. Meanwhile, this method enables an active area efficiency of 15.3% for 5 cm × 5 cm solar modules. The un-encapsulated PSCs exhibit an excellent T80 lifetime exceeding 1600 h under continuous operation conditions in dry nitrogen environment.

O rganic−inorganic lead halide perovskite solar cells (PSCs) have drawn a great deal of attention in the photovoltaic research community due to their high efficiency and simple manufacturing process [1][2][3][4][5] . The power conversion efficiency (PCE) of PSCs has reached over 22% 6 . Similar to other kinds of thin film photovoltaics, film qualities of perovskite layers can heavily influence device performance of PSCs. Thus, a delicate control of film quality of perovskite layers is key to achieving both superior performance and high reproducibility. Many solution processable perovskite formation methods have been developed to fabricate high-quality perovskite films, such as onestep method 2,7-9 and two-step method 10 . These methods usually involve a transformation of 2D lead halide inorganic framework to the 3D perovskite structure. This structure transformation is often accompanied with the formation of morphological defects and crystallographic structure dislocation within the grain and at grain boundaries, where defects are present as trap states causing serious charge recombination and limiting the charge carrier diffusion length (less than 1 μm) 11,12 . To ensure efficient charge collection in the presence of defects, perovskite films are usually made sufficiently thin. So far, high efficiency PSCs mainly adopt perovskite films with thickness ranging from 400 to 800 nm (Supplementary Table 1). The low thickness fluctuation tolerance and morphological defects of thin perovskite films decrease reproducibility of the fabrication process. Moreover, thin films are prone to thickness variations and pinhole defects, which increase significantly with the area.
It is obvious that thicker perovskite films are likely to reduce the risk of forming voids and pinholes due to the larger thickness fluctuation tolerance. A high-quality thick perovskite film is desirable to achieve a high device yield and reproducibility in making large-area solar modules. Besides, thicker perovskite films can not only improve the light harvesting 13,14 , but also broaden light response region by utilizing the below-band absorption 15 . As such, it is advantageous to construct a high-quality thick perovskite film with enough carrier diffusion length in PSCs to endorse both higher efficiency and manufacturing viability. However, thick perovskite films over 1 μm have been found to be generally less efficient than thin-films devices due to the poor mobility, limited carrier diffusion length and are seldom explored (Supplementary Table 1). Moreover, conventional perovskite formation methods are usually not suitable for fabricating largearea high-quality thick perovskite films. Thus, improving the quality of thick perovskite films is critical to balance the thickness and efficiency. Also, the development of new perovskite formation techniques is desirable to construct high-quality and thick perovskite films.
Recently, a perovskite formation process based on methylamine (CH 3 NH 2 ) gas−solid reaction has been developed to prepare highquality CH 3 NH 3 PbX 3 (X = I, Br, Cl), mixed halides CH 3 NH 3 +based perovskite and Cesium/CH 3 NH 3 + mixed cations perovskite films [16][17][18] . This gas−solid reaction-based method introduces a liquid intermediate, which leads to induced morphology reconstruction, defects healing 19,20 . Meanwhile, CH 3 NH 2 gas treatment can help reduce the defect density in perovskite films 19,20 and also elimination of I 2 13 . Besides, CH 3 NH 2 -induced δ-NH 2 CH=NH 2 PbI 3 in (CH 3 NH 3 ) x (NH 2 CH=NH 2 ) 1−x PbI 3 is also reported to passivate trap states to improve device performance 21 . Moreover, excess solvent is absent during this perovskite formation process, which avoids the possible detrimental effect-related residue solvent. Therefore, CH 3 NH 2 gas-based processes are promising for depositing high-quality perovskite films across large area for PSC device fabrication 22,23 .
In terms of the synthetic techniques developed for perovskite, using chloride containing raw materials to prepare perovskite is a commonly used method 3 . Although the chlorine content in resultant perovskite films is usually much lower than the nominal ratio of chlorine added in the precursor solution and in some cases the chorine content in fully annealed perovskite films is below detection limit, both the optoelectronic properties and morphology can be tuned by appropriate chlorine incorporation 24,25 . The resultant carrier diffusion length improvement 11,12 , defects density reduction 26 , optimization of film growth [27][28][29] , and charge tranport 30,31 make this method effective in obtaining high-quality perovskite films. In particular, CH 3 NH 3 PbI 3−x Cl x (i.e., CH 3 NH 3 PbI 3 incorporated with a small amount of Cl) has been reported to exhibit a longer carrier diffusion length over 1 μm compared with pure CH 3 NH 3 PbI 3 , and as a result device performance improved dramatically 11,12 .
These earlier works have inspired us to combine the CH 3 NH 2 gas-based perovskite formation method and partial substitution of iodine ions by chorine ions to fabricate high quality over 1-μmthick perovskite films with a sufficient charge diffusion length, which is not only beneficial for better light absorbing but also desirable for fabricating large-area solar modules with high yield and reproducibility via low-cost printing techniques 13,32 .
Here we report a simple perovskite formation method to fabricate over 1-μm-thick perovskite films based on the fast gas −solid reaction of chlorine-incorporated hydrogen lead triiodide (HPbI 3 (Cl)) and CH 3 NH 2 gas. The comprehensive characterization results reveal that with the synergistic effect of CH 3 NH 2 gas and partial substitution of iodine ions by chorine ions, the resultant CH 3 NH 3 PbI 3 (Cl) (MAPbI 3 (Cl)) films with a thickness over 1 μm and low thickness variation exhibit excellent film quality. The resultant PSCs gave an average PCE of 19.1% and low PCE standard deviation (±0.4%), which indicates the excellent reproducibility of this method. Meanwhile, this method enables an active area PCE of 15.3% for 5 cm × 5 cm solar modules. Besides, the un-encapsulated PSCs exhibit an excellent T 80 lifetime exceeding 1600 h under continuous operation conditions in dry N 2 environment. Our film stability study also offers the indepth understanding for the underlying mechanisms responsible for device stability improvement.

Results
Perovskite formation. Conventional perovskite formation methods are usually not suitable for fabricating high-quality thick perovskite thin films ( Supplementary Fig. 1). Here, we developed a fast gas−solid reaction of HPbI 3 (Cl) and CH 3 NH 2 gas to prepare high-quality thick perovskite films. In this method, hydrogen lead triiodide (HPbI 3 ) was used as the starting material aiming at taking advantage of its fully coordinated [PbI 3 ] − structure to reduce iodide-vacancy and improve film quality ( Table 2 and Supplementary Note 1). As a result, perovskite films with thickness about 1.1 μm are used as our benchmark for further studies. Partial substitution of iodine ions by chorine ions step depicted in Fig. 1a was achieved by reacting CH 3 NH 3 Cl (MACl) with HPbI 3 to form HPbI 3 (Cl). Considering that MACl exists in the solid state at room temperature but sublimes at elevated temperatures (higher than 100°C), the introduction of MACl can not only achieve a small amount of substitution of iodine ions by chorine ions, but also tune the morphology of HPbI 3 (Cl) films 27,35 . Top set in Fig. 1b shows the top-view scanning electron microscope (SEM) images of HPbI 3 (Cl) films prepared using HPbI 3 /MACl precursor solution with different MACl contents. It is found that upon increasing the MACl content, the size of voids between islands-like crystals in the HPbI 3 (Cl) films gradually decreases, and the crystals gradually turn from the hexagonal shape to the square shape and then to crystals with round edges ( Supplementary Fig. 7). The enhanced coverage of HPbI 3 (Cl) films is expected to be beneficial for the subsequent CH 3 NH 2 gasbased perovskite formation process.
To further investigate the incorporation of chlorine into the HPbI 3 , we used X-ray diffraction (XRD) spectroscopy to characterize the obtained HPbI 3 (Cl) films. It is reported that the HPbI 3 films show a hexagonal structure 20,34,36 . Upon increasing MACl content from 0 to 0.20, the low-angle peak around 11.7°shifts to a higher two theta value and the peak intensity increases significantly ( Supplementary Fig. 8). This result suggests that the substitution of Iion with smaller Clion can shrink the crystal lattice, and HPbI 3 (Cl) still preserves the pseudo-hexagonal phase in the case of MACl content lower than 0.4. When the MACl content is increased to 0.40, the XRD peak intensity of the HPbI 3 (Cl) film at around 11.7°reduced substantially. This observation suggests that when the MACl content is too high, the crystal structure of HPbI 3 (Cl) cannot maintain the previous structure, which is also supported by the observed shape difference of crystal islands by SEM (Supplementary Fig. 7). This is because the large difference between the ionic radii of Iand Clmakes the crystal structure unstable when the content of chlorine is too high. Such a phenomenon has also been observed in MAPbI 3−x Cl x . When a small amount of Cl was incorporated, MAPbI 3−x Cl x can still maintain the pseudo-cubic phase 11,12 . However, a continuous solid phase of MAPbI 3−x Cl x theoretically cannot form at a high chlorine content due to the large size mismatch between chloride and iodide 30 . Besides, a slight splitting of the XRD peaks around 11.7°and the formation of MAPbI 3 are observed for high MACl content of 0.70 and 1.00 cases. The former observation may be due to the incomplete removal of MACl or phase separation of HPbI 3 (Cl), and the latter observation suggests that the reaction of MACl with HPbI 3 can form MAPbI 3 . The nonvisible perovskite peaks in HPbI 3 (Cl) films at low MACl content from 0.05 to 0.40 may be due to the low amount of MAPbI 3 seeds and detection limit of XRD. Here, we can deduce that the possible reaction for the formation of HPbI 3 (Cl) at low MACl content up to 0.40 is as follows: where the chlorine-incorporated HPbI 3 , i.e., H 1−x Pb 1−x I 3−4x Cl x is termed as HPbI 3 (Cl) hereafter; for MACl, x + y represents the initial amount of MACl in the HPbI 3 /MACl precusor solution, x represents the amount of MACl used for chlorine incorporation, y represents the amount of MACl that sublimes during thermal annealing.
The as-prepared HPbI 3 (Cl) films were then exposed to CH 3 NH 2 gas atmosphere to form perovskite via the reaction, CH 3 NH 2 + H + → CH 3 NH 3 + 20 . As shown in Fig. 1b bottom set, the apparent grain size of the MAPbI 3 (Cl) films gradually increases upon the increase of MACl content. However, too much MACl is found to cause significant deterioration of the film morphology, especially for the MACl content of 1.00 case ( Supplementary Fig. 9). The structural properties of MAPbI 3 (Cl) films were further examined by XRD (Fig. 2a, b and Supplementary Fig. 10). All the samples show a preferred orientation along the (110) crystallographic plane, and the corresponding peak intensity becomes stronger with consecutive substitution of iodine ions by chorine ions. This preferable (110) crystal orientation is induced by the CH 3 NH 2 -based perovskite formation process 19,37 . The gradually increased peak intensity is ascribed to the effect of chlorine on perovskite film growth 27,35 . Furthermore, it was found that the low-angle XRD peak around 14.1°shifts to a higher two theta value and the peak intensity also increases upon increasing the MACl content. The same as the aforementioned discussion about the HPbI 3 (Cl), substitution of I - by smaller-size Clcan also lead to the shrinkage of the crystal lattice of perovskite, which is consistent with the cell parameters variation ( Supplementary Fig. 11). A slight splitting of the XRD peaks is also observed in the case of the MACl content of 1.00, in which the large size mismatch between chloride and iodide precludes the formation of a continuous solid phase of MAPbI 3−x Cl x , and results in phase segregation and large cell parameter variation ( Fig. 2b and Supplementary Fig. 11). The observed additional peak around 28.2°is likely associated with the hydrate phase of perovskite 38 . Although MACl is introduced into precursor with nominal ratios of MACl/HPbI 3 from 0.1 to 1, the remaining contents of chlorine in HPbI 3 (Cl) and MAPbI 3 (Cl) can be significantly lower than the nominal content of chlorine in precursor solution. This is due to the fact that MACl tends to sublime at elevated temperatures. Furthermore, chlorine tends to form volatile chlorine-containing species, which also tends to gradually diffuse and sublimate from the film, especially in the presence of water 25,27 . Thus, after thermal annealing in open air (relative humidity of about 40−60%) the remaining chlorine content in the resultant MAPbI 3 (Cl) films is much lower than the chlorine content in the precursor solution. The low content of chlorine in the MAPbI 3 (Cl) films is proven by the tiny change in absorption edge (Fig. 2c) 39 . It should be noted that the fluctuation of absorption for the case of MACl content of 0.70 and 1.00 may be due to the observed slight phase segregation and large roughness (Supplementary Fig. 12) 25,27,35,39 . Similar to chorine inclusion, phase segregation is also observed in the bromine inclusion case with high content, i.e., (MAPb(I x Br 1−x ) 3 ) (0.1 < x < 0.8) 40 , in which perovskite undergo phase separation into iodide-rich and bromide-rich regions under light illumination 41 . In the case of bromine inclusion with low content, the existence of bromine is found to assist grain growth, improve stability, and device performance [40][41][42] .
To further confirm the existence of chorine in MAPbI 3 (Cl) films, we employed secondary-ion mass spectrometry (SIMS) to characterize the depth profile of chlorine in the perovskite film with MACl content of 0.1 (which gives the best device performance as discussed in the device performance section). As shown in Fig. 2d, the Cl depth profile exhibits a continuous and gradual increase, and then shows a significant spatial overlap on the Ti profile. This observation is consistent with other reports revealing that the distribution of chlorine within MAPbI 3 (Cl) is often inhomogeneous 25,43-45 . One possible cause responsible for the inhomogeneous distribution of chlorine is due to the tendency of chlorine to form volatile chlorine-containing species gradually diffusing and sublimating from the top surface of the film, especially in ambient air, resulting in a gradient increase of the Cl content from the film surface to bottom 25 . Although the XRD results of the MAPbI 3 (Cl) films do not show obvious peaks of PbCl 2 or MAPbCl 3 , it is worth noting that in addition to incorporation into perovskite, the chlorine may form some chloride compounds that are nondetectable by XRD due to the detection limit or the amorphous nature of these compounds. Here, we can complete the second step of MAPbI 3 (Cl) formation reaction depicted in Fig. 1a as follows: where the chlorine-incorporated perovskite, i.e., MAPbI 3−x Cl x is termed as MAPbI 3 (Cl) hereafter.
Device performance. To study the device performance, we implemented our smooth 1.1-μm-thick MAPbI 3 (Cl) films in the   Fig. 3a, a best PCE of 20.0% was achieved. The integrated J SC from external quantum efficiency (EQE) is in excellent agreement with current density-voltage (J−V) measurements, with a discrepancy below 2% (Fig. 3b). The representative J-V curves under different scan directions indicate the negligible hysteresis effect ( Supplementary Fig. 14). In addition, a stabilized PCE approaching 19% for device (with efficiency of 19.2% from J−V measurements) is also achieved by holding the voltage at the maximum power point for 500 s (Fig. 3c), which further confirms the negligible hysteresis effect. When compared with reports listed in Supplementary Table 1, the MAPbI 3 (Cl)based devices in this work can also deliver a comparable high efficiency with perovskite film thickness over 1.1 μm as shown in Fig. 3d. This suggests that at least the 1.1-μm-thick MAPbI 3 (Cl) layer deposited here does not show a detrimental effect in carrier transport. To further inspect the devices, we performed the crosssectional-view SEM (Fig. 4a). It is found that the grains extend across the entire absorber layer minimizing grain boundaries and providing facile pathways for efficient charge transport. All these results prove that we obtained high-quality thick perovskite films, and high efficiency PSCs can be fabricated based on these films. Reproducibility is crucial for mass production of low-cost optoelectronic device applications. When compared with PSCs based on perovskite films fabricated using the antisolvent method, our 1.1-μm-thick MAPbI 3 (Cl) film-based devices show significantly improved reproducibility represented by a much smaller PCE standard deviation (decreasing from ±1.2% to ±0.4%, Supplementary Fig. 15 and Supplementary Table 4). The high reproducibility of device performance is attributed to the high reproducibility of 1.1-μm-thick MAPbI 3 (Cl) films. To confirm this point, we measured the thickness and roughness of 16 samples of the 1.1-μm-thick MAPbI 3 (Cl) films (Fig. 4b, c and Supplementary Figs. 16, 17). The average thickness and roughness together with their standard deviations are determined to be 1.13 ± 0.03 μm and 5.4 ± 0.4 nm, respectively. In addition, the device performance of 1.1-μm-thick MAPbI 3 (Cl) devices is higher than that of the devices based on 450-nm-thick perovskite film deposited by the antisolvent method. The higher efficiency for the former case is mainly due to the improved J SC and FF, which is also supported by the EQE results in Supplementary  Fig. 15. This improvement is not only due to the thicker film but also its improved film quality. Moreover, a slightly wider spectral response is observed for the 1.1-μm-thick MAPbI 3 (Cl) film-based devices. These results suggest that high-quality, thick perovskite films can benefit the enhancement of light harvesting capability, especially at the near-infrared region, towards better device performance.
Development of scalable fabrication processes is key to further industrialization of the PSC technology. We demonstrate that our new perovskite formation method is scalable and can be used to fabricate large-area solar modules. Figure 4d shows the optical photograph of 1.1-μm-thick MAPbI 3 (Cl) film on 5 cm × 5 cm substrates with the mirror-like smooth surface. It is also found that the film is much more uniform over the entire 5 cm × 5 cm substrate than that of antisolvent case (Supplementary Figs. 18, . We then fabricated six-cell modules (active area = 12.0 cm 2 , geometric fill factor = 48%) as shown in Fig. 4e. A best active area PCE of 15.3% under reverse scan was achieved with a J SC of 3.66 mA cm −2 , V OC of 6.65 V, and FF of 0.630 (Fig. 4f), which is among the top PCE values obtained for perovskite solar modules with an active area greater than 10 cm 2 35,46-56 . Furthermore, our new method shows good reproducibility in the large-area device fabrication with an average module PCE and standard deviation of 13.6 ± 0.8% (Supplementary Table 5), which is much better than that of the antisolvent case (8.6 ± 1.6%, Supplementary  Fig. 20 and Supplementary Table 6). These results further confirm the excellent reproducibility of our 1.1-μm-thick MAPbI 3 (Cl) films on large scale, which is a key advantage for manufacturing in the realistic industrial large-scale setting. Note that small J-V hysteresis was observed for solar modules ( Supplementary  Fig. 21).
Charge carrier transport behavior. As shown in Fig. 3d, we obtained high efficiencies with 1.1-μm-thick perovskite films. The promising device performance inspired us to conduct systematical investigations of charge transport properties of our PSC devices to find out what factors lead to the high device performance. We first performed conductive atomic force microscopy (c-AFM) measurements on the 1.1-μm-thick MAPbI 3 (Cl) film and perovskite film prepared by the antisolvent method (denoted as MAPbI 3 (AS)). It was found that the former shows lower roughness (Supplementary Note 3) and rounded large grains with almost four times larger grain size, and its corresponding fewer grain boundaries are expected to reduce the defects density ( Supplementary Fig. 22). Moreover, the 1.1-μm-thick MAP-bI 3 (Cl) film also shows nearly an order of magnitude higher lightinduced current (Fig. 5a), which indicates its good electrical conductivity. This observation can be ascribed to the improved crystallinity of the MAPbI 3 (Cl) films with preferential (110) orientation (Fig. 5b), as a result of synergistic effect of CH 3 NH 2 gas 19,57 and partial substitution of I − by Cl −11,12 .
In addition to the improvement in terms of the film morphology and conductivity, the defect nature in perovskite films may also play a role in influencing charge carrier transport properties 58 . To investigate radiative and nonradiative charge carrier recombination channels within perovskite films, we carried out time-resolved photoluminescence (TRPL) measurements on glass/perovskite/polymethyl methacrylate (PMMA) samples ( Supplementary Fig. 23). Figure 5c shows the TRPL decay curves obtained from both the glass side and PMMA side for the 1.1-μm-thick perovskite films with (MAPbI 3 (Cl)) and without (denoted as MAPbI 3 ) chlorine incorporation. Considering the limited penetration depth (tens of nanometers) of the light source, the PL emission signal only contains the charge carrier information on the surface layer that is facing the light source 59 . The time constants (τ) are calculated to be 54 ns for the MAPbI 3 film no matter whether the laser light incidents from the PMMA side or the glass side, which indicates that perovskite on the bottom and top of film has nearly the same charge carrier properties. In the MAPbI 3 (Cl) case, the PL lifetime obtained from the glass side is longer than that obtained from the PMMA side. This observation suggests that perovskite on the bottom has better charge carrier properties than the perovskite on the top in MAPbI 3 (Cl) film. This observation may be correlated with the inhomogeneous distribution of chlorine within MAPbI 3 (Cl) film, where chlorine within a certain range of concentrations can benefit charge carrier properties. Besides, the MAPbI 3 (Cl) film shows longer PL lifetimes than that of the MAPbI 3 film. This improvement can be ascribed to the better crystallinity and a lower defect density of the MAPbI 3 (Cl) film 60 .
To quantitatively evaluate the defect density, we fabricated sandwich devices by inserting the perovskite films between FTO and gold, and characterized the evolution of the space-chargelimited current (SCLC) for different biases, as shown in Fig. 5d−i. In general, the presence of mobile species in perovskite can lead to complication in interpreting the SCLC measurement results, and often precludes a robust quantitative analysis based on such measurements. On the other hand, we observed low hysteresis on the basis of J-V measurements, which indicates a negligible effect of mobile species. Thus, in this case we can safely deduce defects density (N t ) within perovskite films based on the SCLC measurements according to the equation 13,61 : where ε and ε 0 are the dielectric constants of perovskite and the vacuum permittivity, respectively, V TFL is trap-filled limit voltage , L is the thickness of the perovskite films, and e is the elementary charge. We estimated the defect density to be 2.13×10 15 and 2.51×10 15 cm −3 for the 1.1-μm-thick MAPbI 3 (Cl) film and MAPbI 3 film, respectively, prepared by the HPbI 3 (Cl)/CH 3 NH 2 method (Table 1). These defect density values are significantly lower than antisolvent case (1.48×10 17 cm −3 ). The similar magnitude in terms of defects densities derived from the SCLC measurements in contrast with the clear difference in terms of lifetime derived from PL measurements is possibly due to the fact that the former measurement gives the defect information in the entire films, and the latter measurements only provides the defect information for perovskite on the top or bottom of the film.
When the bias voltage further increases in the SCLC curves, the current shows a linear relationship with the square of the voltage, where we can deduce the mobility (μ) from the Mott −Gurney law 61 : The mobility are estimated to be 1.12 and 0.83 cm 2 V −1 s −1 for the 1.1-μm-thick MAPbI 3 (Cl) film and MAPbI 3 film, respectively. The much higher carrier conductivity is consistent with the higher device performance of the MAPbI 3 (Cl) case, which again verifies the positive effect of small content chlorine incorporation on the optoelectronic properties of perovskite films. In addition, these mobility values around 1 cm 2 V −1 s −1 are two orders of magnitude higher than that in antisolvent case (0.014 cm 2 V −1 s −1 ), which further confirms the excellent charge carrier transport properties of CH 3 NH 2 gas-based perovskite films. The improved carrier properties of the 1.1-μm-thick MAPbI 3 (Cl) films when implemented into PSCs is also verified by impedance spectroscopy characterization of devices under light illumination ( Table 1 Supplementary Fig. 24 and Supplementary Note 4).
Device stability and perovskite film stability. Operational stability is another key aspect of PSCs. We studied the stability of the small size devices under the operation conditions (Fig. 6a). It was found that the 1.1-μm-thick MAPbI 3 (Cl)-based devices showed much better stability than the MAPbI 3 (AS) devices. Especially, the MAPbI 3 (Cl) devices maintained over 90% of their initial performance for 800 h under continuous light illumination with   maximum power point tracking (MPPT). While the PCEs of the MAPbI 3 (AS) devices decreased to 90% of their initial performance after approximately 300 h. An initial increase of PCEs for both samples is observed in the first 20 h of the stability test. A similar phenomenon was also observed previously 55,62 , and was ascribed to the effects such as elevated temperature (approximately 46°C), light or field induced ion movement, light-induced traps formation, interfacial charge accumulation 55 , or spiro-OMeTAD conductivity variation 63 . In addition, both devices showed a fast, exponential decay region after reaching the highest performance points, which is followed by a slower linear decay 64 . By fitting the linear region, the MAPbI 3 (Cl) devices exhibit a linear slope of −0.0014% h −1 , which is 30% slower than the antisolvent case with a linear slope of −0.0020% h −1 . Using these slope values, it is expected to take 1660 and 870 h of operation to reach the T 80 points for MAPbI 3 (Cl) devices and the MAPbI 3 (AS) devices, respectively. The nearly doubled T 80 parameters for the MAPbI 3 (Cl) devices suggests the excellent long-term stability of the cells based on our Cl-incorporated thick films. Note that the stability measurements in this study were conducted under operation conditions, i.e., continuous illumination, MPPT, dry N 2 gas environment simulating the case of encapsulation, and therefore the results obtained here can be used as a realistic indicator of the device operation lifetime of our PSCs. It should be noted that all of the components in PSCs play an important role in device stability. Considering that MAPbI 3 (Cl) devices and MAPbI 3 (AS) devices are assembled using the same device configuration, the impact of charge selective layers and electrode on device stability can be excluded out. Hence, the significant stability difference between MAPbI 3 (Cl) devices and MAPbI 3 (AS) devices is most likely caused by the quality of the perovskite films such as crystallinity, grain size and/or defects, etc. 13,22,34 .
To reveal the underlying mechanism responsible for the improved stability, we prepared four kinds of films, i.e., film A: 1.1 μm MAPbI 3 (Cl) film prepared by the HPbI 3 (Cl)/CH 3 NH 2 method; film B: 1.1 μm MAPbI 3 film by the HPbI 3 /CH 3 NH 2 method; film C: 0.5 μm MAPbI 3 (Cl) films prepared by the HPbI 3 (Cl)/CH 3 NH 2 method; film D: 0.45 μm MAPbI 3 (AS) films prepared by the antisolvent method. We tested the stability of these films under light illumination, a high relative humidity of about 100% and gentle thermal environment of 50°C (denoted as LHT) as shown in Supplementary Fig. 25. It is found that that the stability of these four films is in the order film A > film B > film C > film D (Supplementary Figs. 25 to 28). The improved stability of film A is firstly attributed to its improved grain crystallinity and larger grain size, which can be clearly seen by comparing the XRD peak intensity and full width half maximum of fresh film A with fresh film D (Supplementary Fig. 28b, c and Supplementary Note 5). This is because the ions in highly crystalline perovskite films have higher chemical binding energies to prevent themselves from migrating, which benefits light and thermal stability 65 . This is confirmed by SEM results shown in Supplementary Fig. 29. We found that the decomposition of perovskite mainly started from grain boundaries for all films after 1 h of LTH testing, where H 2 O and O 2 can easily react with perovskite films due to the facts that (1) chemical binding at grain boundaries with low crystallinity is much weaker than that within crystals and (2) the severe ion migration in low crystallinity region especially at grain boundaries facilitates degradation (Fig. 6b) 65 . When further comparing film A with film B, film A shows better crystallinity, larger grain size, and also better stability ( Supplementary Figs. 26 to 30). In addition, better stability of films A and B than thinner films C and D also verifies that thicker films are more advantageous in achieving better stability because the extra top layer of the compact high-quality film in the thicker films naturally helps as a protection layer to protect the film underlying the top layer. Besides, previous reports have demonstrated that often a tiny amount of I 2 existing in perovskite films can cause severe degradation of perovskite 13,66 . While, CH 3 NH 2 used here provides an alkaline condition to suppress the formation of I 2 during the perovskite formation process and/or reduce the I 2 amount during the initial degradation of iodine containing perovskite films ( Fig. 6b and Supplementary Fig. 28f), which may contribute to the better stability of films A, B and C than film D 13,66 .

Discussion
A simple perovskite formation method based on fast reaction of chlorine-incorporated HPbI 3 and CH 3 NH 2 has been developed to fabricate high-quality, over 1 μm thick and stable CH 3 NH 3 PbI 3 (Cl) perovskite films in ambient condition. The resultant films can not only enable high efficiencies of 19.1 ± 0.4% (champion PCE = 20.0%) for small size PSCs and 13.6 ± 0.8% (champion active area PCE = 15.3%) for 5 cm × 5 cm perovskite solar modules, but also deliver excellent device reproducibility. In addition, the resultant un-encapsulated small size PSCs exhibit an excellent T 80 lifetime exceeding 1600 h under continuous light illumination with MPPT in dry N 2 environment. The excellent device stability is mainly a result of the excellent films stability. Our study not only provides a highly reproducible method to fabricate PSCs and modules with enhanced efficiency and stability, but also offers the in-depth understanding for the underlying mechanisms responsible for device stability improvement. The encouraging device performance and excellent stability in this work points out to a promising direction, i.e., use of thick absorber films to realize PSCs and modules with high efficiency, reproducibility, and stability.
Synthesis of HPbI 3 : The HPbI 3 crystals were prepared using an antisolvent vapor-assisted crystallization method as shown in Supplementary Fig. 2. Briefly, 11 g PbI 2 and 7 g of HI were mixed and dissolved in 17 g of DMF to form an HPbI 3 solution. The HPbI 3 solution was heated at 100°C in chlorobenzene vapor environment for 24 h. During this heat-treatment, chlorobenzene diffuses into the HPbI 3 /DMF solution and reduces the solubility of HPbI 3 . Light yellow needle-like HPbI 3 crystals were formed, which were collected and washed with chlorobenzene/ DMF (volume ratio = 3:1) for three times and then washed with ethanol for three times. Finally, the HPbI 3 crystals were dried at 60°C for 24 h under vacuum.
Fabrication of perovskite films and solar cells. To study the dependence of the film thickness on preparation conditions, the HPbI 3 films are obtained by spincoating HPbI 3 precursor (60 wt% HPbI 3 in DMF) at 5000 rpm for 30 s on substrates under different temperatures (room temperature, 60, 70, 80, 90, 100°C) in ambient air condition. For chlorine-incorporated perovskite deposition, the precursor solutions containing different molar ratios of MACl/HPbI 3 (0, 0.05, 0.10, 0.15, 0.20, 0.40, 0.70, 1.00) are deposited on substrates preheated at 90°C at 5000 rpm for 30 s. After thermal annealing at 100°C for 5 min to remove the solvent, part of the MACl and also part or all of the formed HI during this process, the obtained films react with CH 3 NH 2 gas to form perovskite 20,22 . Then the obtained perovskite films are annealed at 100°C for 5 min in ambient air condition before device fabrication or other characterization. A 30 μL of spiro-OMeTAD was spin coated on the perovskite film at 3000 rpm for 30 s, where a spiro-OMeTAD/ chlorobenzene (72.3 mg mL −1 ) solution was employed with the addition of 17.5 μL Li-TFSI/acetonitrile (520 mg mL −1 ), and 28.8 μL 4-tert-butylpyridine. Finally, a gold layer with a thickness of 100 nm was deposited as the counter electrode on the top of spiro-OMeTAD layer through shadow masks via thermal evaporation under high vacuum (5×10 −5 Torr).
Characterization. Current-voltage (J-V) characteristics of PSCs are measured under one sun illumination (AM 1.5 G, 100 mW cm −2 , calibrated using a Newport reference Si-cell, Oriel Instruments Model Number 90026564, 2 × 2 cm 2 ) using a solar simulator (Newport Oriel Sol 1A, Xenon-lamp, USHIO, UXL-150SO) and a Keithley 2400 source meter in ambient air at about 25 o C and a relative humidity of about 40−60%. The small size PSCs were measured using a 0.1 cm 2 metal mask. No mask is used for measurement of perovskite modules. The active area of perovskite modules is 12.0 cm 2 , determined by the overlap areas of top and bottom electrodes described in our previous publication 46 . All the J−V curves are measured under reverse scan with a scan rate of 0.25 V s −1 without preconditioning unless otherwise specified. For stability measurements, the cells were subsequently loaded in our home-designed environmental chamber coupled with a solar simulator (Peccell PEC-L01, AM1.5G) and source meter (Keithley 2401) controlled by a LabView program allowing automatic sequential measurements on the devices with adjustable acquisition time intervals. To simulate continuous solar cell operation an active bias voltage was applied to the cells maintaining the solar cell operation at the maximum power point. The devices were kept at the maximum power output voltage during the intervals between consecutive measurements. No UV-filters were used, i.e., the UV component is included in illumination. The stability measurement was performed under nitrogen box with a relative humidity below 5%. EQE measurements were performed on an Oriel IQE 200 in DC mode.
SEM measurements was carried out in scanning electron microscope (Helios NanoLab G3 UC, FEI). XRD measurements were carried out in a Bruker D8 Discover instrument (Bruker AXS GmbH, Karlsruhe, Germany) equipped with Cu wavelength λ = 1.54 Å X-ray source operated at 1600 W and Goebel mirror. Data were collected from 5 to 60 two theta degrees with a 0.02 degrees step. Experimental data were fitted to obtain unit cell phase parameters using the Profex/ BGMN software (v.3.12.0, http://profex.doebelin.org/). TRPL was acquired using the time-correlated, single-photon counting technique (Hamamatsu, C10627), and excitation was provided by a femtosecond mode-locked Ti:sapphire laser (Spectra-Physics, MAITAI XF-IMW) at 450 nm with an average power at 8 MHz of 0.74 mW. In TRPL, perovskite films were covered with PMMA on the top of the perovskite layer to exclude/minimize the influence of ambient air (especially the influence from H 2 O and O 2 ) [67][68][69][70] . SIMS (Kratos Axis ULTRA) equipped with quadrupole mass spectrometer (HAL 7, Hiden Analytical) was used to collect the elemental signal in positive ion detection mode (PID). For sputtering in SIMS, 3 keV Ar + primary beam with a 10 mA current and diameter of 100 µm were utilized. The beam was set at an angle of 45°with respect to the sample surface normal. The spectrometer was operated at a pressure of 10 −8 Torr. The c-AFM measurement of the films was performed in contact-mode (MFP-3D series, Asylum Research) in ambient conditions (35% RH, 22°C). Top side illumination was generated with a halogen lamp. Chromium/platinum-coated cantilevers with a nominal spring constant of 0.2 N m −1 were used to collect the photo-generated currents. The SCLC data were collected with a semiconductor characterization system in N 2 (4200-SCS, Keithley). Absorbance was measured using a UV−Vis spectrometer (JASCO Inc., V-670). Impedance Spectroscopy measurements had been recorded with an Autolab PGSTAT204 potentiostat equipped with a frequency response analyzer module FRA32, AC perturbation was set to 10 mV and frequency ranged between 1 MHz to 0.5 Hz. The measurements were done in N 2 environment with less than 0.1% relative humidity, and Oriel VeraSol-2class AAALED light source simulating AM 1.5 spectra.

Data availability
The data that support the findings of this study are available from the corresponding author on reasonable request.