A polymer scaffold for self-healing perovskite solar cells

Advancing of the lead halide perovskite solar cells towards photovoltaic market demands large-scale devices of high-power conversion efficiency, high reproducibility and stability via low-cost fabrication technology, and in particular resistance to humid environment for long-time operation. Here we achieve uniform perovskite film based on a novel polymer-scaffold architecture via a mild-temperature process. These solar cells exhibit efficiency of up to ∼16% with small variation. The unencapsulated devices retain high output for up to 300 h in highly humid environment (70% relative humidity). Moreover, they show strong humidity resistant and self-healing behaviour, recovering rapidly after removing from water vapour. Not only the film can self-heal in this case, but the corresponding devices can present power conversion efficiency recovery after the water vapour is removed. Our work demonstrates the value of cheap, long chain and hygroscopic polymer scaffold in perovskite solar cells towards commercialization.

T he rapid rising of the lead halide perovskite solar cells (PSCs), typically based on CH 3 NH 3 PbI 3 (MAPbI 3 ) or its analogues, demonstrate great potential for market applications [1][2][3][4][5][6][7][8][9][10][11][12][13] . The worldwide power conversion efficiency (PCE) race for the perovskite solar cells has stimulated the highest record of 420% update via solvent engineering 8 , interface engineering 5 and composition engineering 10 , which approaches the best of polycrystalline silicon modules. The excellent performance of PSCs is attributed to the high absorption coefficient, weak exciton binding energy, and long diffusion length of the perovskite phase 12,[14][15][16][17] . Two dominant device architectures, namely mesoporous scaffold 1-3 and planar heterojunction structures 4,5 have been developed. The PCE in mesoporous scaffold architecture depends on the pore size, porosity and morphology of the metal oxide nanoparticles, which dominantly determine the coverage, morphology variation of the perovskite layer and the carrier lifetime. Moreover, the fabrication of inorganic metal oxide mesoporous scaffold is complicated and needs sintering at high temperature of 4450°C (refs 2,3), which makes the fabrication more expensive. The planar heterojunction PSCs fabricated layer by layer is either fabricated via high vacuum deposition 4 or solution-based method 5,18 . However, the low-cost solution process is rather challenging to form homogeneous film due to the dewetting process and sensitivity to the atmosphere [18][19][20] . Recently, progress has been reported to control the nucleation rate by changing dissolvent or temperature of the substrate 21,22 . Besides high efficiency and high reproducibility, long-term stability is also a crucial barrier to commercialize the PSCs, because the lead halide perovskite materials are very sensitive to ambient humidity and easy to dissolve and degrade in humid environment 5,23,24 . Therefore, it is very challenging and pressing to develop ultra stable and high PCE devices resistant to hostile operation environment.
Here we report a novel perovskite solar cell architecture based on an insulating polymer scaffold structure. The polymer-scaffold perovskite layer is simply fabricated via a single one-step process under mild temperature of 105°C. The as-made polymer-scaffold perovskite solar cell (PPSC) devices demonstrate high PCE up to B16% under standard AM1.5 illumination and very small efficiency variation compared with that without the insulating polymer. Remarkably, unsealed PPSCs exhibit excellent performance in stability test in highly humid environment (70% relative humidity), which can endure the moisture for over 300 h. Most strikingly, the PPSCs show strong self-healing or water resistant effect under vapour spray. The perovskite film, as well as the corresponding devices can recover rapidly to its original phase and performance after being vapour sprayed and dried in ambient air, respectively.

Results
Design, fabrication and characterization of PPSCs. The PPSC architecture inherits the basic structure of PSCs with inorganic mesoporous scaffold 1-3 , instead, the inorganic nanocrystal scaffolds are replaced with long-and flexible-polymer chain networks, which can be processed under mild temperature (Fig. 1a-c). The fabrication of PPSC is similar to that of planar heterojunction PSCs, except for the precursor solution preparation (Fig. 1d). Long-chain insulating polymers, dissolved in precursor solution, assemble the polymer scaffold simultaneously as the precursor solution is spin coated on the substrate. The selected polymer should be chemically inert to ingredients for perovskite synthesis; it should also act as an electrical insulator to guarantee the effective transport of photo-generated carriers. Here polyethylene glycol (PEG), which meets the requirements, is chosen to be added in the precursor solution of PbCl 2 and CH 3 NH 3 I (MAI; 1:3 molar ratio) in dimethylformamide. The molar ratio of PEG monomers (C 2 H 4 O) to ultimate product MAPbI 3 is close to 1:1 (PEG in 40 mg ml À 1 ). The PEG scaffold perovskite absorber layer was prepared on the TiO 2 (B40 nm thickness)-coated fluorine doped tin oxide (FTO) glass by spin coating followed by annealing at 105°C for 70 min. 2,29,7,79-tetrakis-(N,N-di-p-methoxyphenylamine) 9,99-spirobifluorene (Spiro-OMeTAD) and gold were used as the hole transport layer and electrode, respectively.
Figure 1e,f present top-view scanning electron microscope (SEM) images of the perovskite films without and with polymer scaffold. Those without scaffold exhibited large pinholes and bare conducting substrates, increasing the risk of short circuits. Those with scaffold demonstrated continuous and complete coverage. The cross-section SEM image showed the perovskite layer without polymer to have considerable variation in thickness and many voids (Fig. 1g). The film with polymer was uniform in thickness over large area (Fig. 1h). The PEG scaffold can also work in another precursor solution PbI 2 :MAI (1:1) ( Supplementary Fig. 1). PEG plays an important role in improvement of the film quality for the following reasons. First, the simultaneously formed PEG scaffold, which is formed by entangled long-chain molecules, acts as a three-dimensional skeleton to support perovskite crystals, causing it to cover the substrate more uniformly (Fig. 1c), demonstrating an advantage over planar heterojunction PSCs. Uniformly distributed PEG scaffolds function as Al 2 O 3 scaffolds 3 , and they can be produced in a much simpler and less costly manner. Second, the wetting properties of the precursor solution have been improved after introducing PEG ( Supplementary Fig. 2), which can improve film morphology 19 . Third, PEG molecules in precursor solution can slow down the crystallization of perovskite from the X-ray diffraction characterization by tracking the intermediate phase in the crystallization process, which is shown below.
To confirm that PEG acts as scaffold in the architecture, detailed energy dispersive spectroscopy (EDS) analysis in a Cs-corrected scanning transmission electron microscope (STEM) was conducted for element mapping investigation (Fig. 2a-c). The high-angle annular dark field (HAADF) STEM image ( Fig. 2a) reveals that the perovskite crystalline grains are surrounded by polymer material. The profiles of perovskite grains are highlighted with dashed lines according to the EDS mapping of Pb component (Fig. 2b). The amorphous material surrounding the perovskite grains is PEG in accordance with the EDS mapping of O element because only PEG contains O. The O mapping demonstrates that PEG not only covers the perovskite grains, but also cross-links the neighbouring perovskite grains via self-assembly forming a polymer-scaffold network to support the perovskite grains (Fig. 2c). Above conclusion is also evidenced by the fact that O mapping covers larger area than Pb and extends from crystal boundaries, indicating that the perovskite grains are imbedded in the PEG networks, corresponding well to the HAADF-STEM image. In addition, the O element mapping on the cross-sectional area of the PPSC film confirms a homogeneous distribution of PEG molecules in the perovskite film ( Supplementary Fig. 3). High-resolution transmission electron microscope image of perovskite grain demonstrates good crystal lattice ( Supplementary Fig. 4), suggesting that PEG molecules do not damage the perovskite crystal structure.
The phase evolution of PEG scaffold perovskite film during annealing was characterized by X-ray diffraction analysis (Fig. 2d). Over the course of 30-50 min, the film turned from orange to yellow, the (110) and (220) peaks of perovskite phase 25,26 became more pronounced and some new peaks appeared between 10 and 20°. The intermediate phase disappeared after annealing for 60 min, and the sample became deep black at 105°C after 70 min of annealing, with stronger (110) and (220) peaks. X-ray diffraction patterns of perovskite film without PEG and with PEG under the same 100°C sintering temperature are also shown in Supplementary Fig. 5 Photoluminescence (PL) spectra were measured to verify whether excess recombination sites and defect states are induced in the PEG scaffold perovskite film 27,28 . Compared with the PL spectrum of pristine perovskite film, the intensity is not influenced after being coated with PEG ( Fig. 3b), indicating the insulating nature of PEG, as shown in Fig. 3a. Figure 3c shows the time-resolved PL spectra with and without PEG polymer scaffold, with the similar carrier lifetime by using single exponential fitting (110.8 ns with PEG, 120.1 ns without PEG). This result indicates B1 mm diffusion length if diffusion coefficient is supposed to be 0.1 cm 2 s À 1 , which is sufficient to guarantee effective collection efficiency. On the other hand, no visible shift (peak position of 760 nm) and broadening (half width of 40 nm) of peaks associated with perovskite layers with and without PEG scaffold is observed, indicating that PEG does not induce any excessive recombination ( Supplementary Fig. 6).
Photovoltaic performance of PPSCs. The photovoltaic properties of PPSCs are characterized systematically. The PEG molecular weight and its concentration in precursor solution were found to play an important role in the performance of PPSCs. Figure 4a shows the current density-voltage (J-V) curves of samples with PEG in different molecule weights (12,000/20,000/ 100,000 Da) measured under simulated AM1.5 illumination of 100 mW cm À 2 . The molecular weight of PEG, proportional to the length of polymer, has a significant effect on the performance of PPSCs, and the optimal molecular weight was found to be 420,000 Da because longer PEG molecules facilitate the formation of three-dimensional molecular network in the perovskite film, and guarantee uniform morphology. PEG concentration is also critical to devices because it is related to the series resistance. As demonstrated in Fig. 4b, the device shows much better performance with 20 and 40 mg ml À 1 PEG concentration, and its PCE decreases at higher concentrations (80 mg ml À 1 ) because the excess PEG in the film increases the series resistance and lowers the efficiency. Note that viscosity of the solution can be strongly influenced by PEG concentration and molecule weight, so the r.p.m. value of spin coating is optimized at different PEG concentration and molecule weight (see Methods for details) to guarantee B400 nm film thickness. The improved efficiency was also consistent with the increased recombination time constant 28 in PPSCs, compared with that of control devices, as is shown in Fig. 4c. Owing to the improved coverage, the recombination time is almost one order of magnitude larger than that without polymer scaffold at zero bias. The highest PCE of 16% was obtained for PPSCs with optimal PEG molecule weight of 20,000 Da and concentration of 20 mg ml À 1 , and showed an open voltage (V oc ) of 0.98 V, short-circuit current (J sc ) of 22.5 mA cm À 2 , and fill factor of 0.72 ( Supplementary Fig. 7a). Another representative device shows a steady-state PCE of 15.4% at first 10 s, and the PCE gradually drops down and eventually stabilized at 13.5% in 1,000 s (0.7 V bias, Supplementary Fig. 7b). Parallel devices present H value from 10 to 30% under 500 mV s À 1 if hysteresis index is defined H ¼ (PCE reverse À PCE forward )/PCE reverse Â 100% to characterize the hysteresis effect. PCE reverse and PCE forward represent PCE obtained from J to V measurement in reverse scan and forward scan, respectively. The statistical PCE distributions of 4100 samples without and with (for two concentrations) polymer scaffolds are shown in Fig. 4d. The PPSCs exhibited a narrow PCE distribution for 20 and 40 mg ml À 1 , with an average efficiency of 14 and 12.5%, respectively, but the PCE distribution of the PSCs without polymer scaffold was much broader, varying from 2 to 14%, averaging at B8%. The significant increase in PCE and high reproducibility of the PPSC devices are direct reflections of the uniform thickness and homogeneous morphology of the polymer-scaffold perovskite layer. Improved stability in highly humid environment. More importantly, the insulating PEG scaffold structure can stabilize the PSCs in humid environment due to the strong hygroscopicity property of PEG. Here three groups of unsealed devices were prepared without and with PEG (two concentrations), respectively. Figure 5a shows the PCE evolution as a function of time of the three groups of devices in the dark under highly humid environment (relative humidity 70%) without any sealing. In contrast to the fast PCE decay from 12 to 0.5% within 50 h of the pristine PSCs, the PPSCs retained 65% of its PCE after 300 h aging. Similarly, the J sc , V oc , and fill factor of the pristine sample decreased quickly during the first 50 h, while those of the PPSCs retained relatively high values for up to 300 h ( Supplementary  Fig. 8a,b). Note that the device with PEG concentration of 40 mg ml À 1 showed better performance than the one with 20 mg ml À 1 because more PEG can form a more compact layer, enhancing the moisture resistance of devices. We also studied the stability under continuous light illumination in 70% humidity for unsealed devices (Supplementary Fig. 8c,d). Photovoltaic performance degraded quickly within 10 h in the device without PEG scaffold, while PPSC retained 83% of its PCE during the same aging period. Similarly, V oc , J sc , and fill factor suffered substantial decrease in the first 10 h for pristine sample while they remained relatively high values in the presence of PEG. X-ray diffraction patterns of the perovskite film with PEG scaffold showed no structure destruction after 72 h continuous light irradiation, while the samples without PEG degraded into PbI 2 quickly within 10 h (Supplementary Fig. 9). Evidence above strongly suggests that PEG can greatly protect perovskite film from decomposition under moisture attack and significantly improve the stability of PPSCs at very humid environment.
Self-healing property of PPSCs after vapour spray. In addition to its improved stability, the PPSC devices can demonstrate a self-healing behaviour and humidity resistance effect on exposure to water vapour. Figure 5b shows a comparison of the changes in colour of the perovskite layer with PEG (top) and the perovskite layer without PEG (bottom) after both were sprayed with water vapour for 60 s. The perovskite film without PEG decomposed into PbI 2 and turned yellow irreversibly. In contrast, the PEG scaffold perovskite film showed yellow in colour at first and recovered to black in B45 s after removing from the spray (Fig. 5b). This amazing self-healing behaviour is further vividly demonstrated in 120 s videos ( Supplementary Movies 1 and 2). The self-healing capability of the PPSC device was also manifested in the J-V characterization (Fig. 5c). The PCE degrades with smaller V oc and J sc after exposure to water vapour for 60 s. Surprisingly, results showed that the J-V curve can almost regain its original value in 45 s, indicating that the PPSCs can heal itself when it is returned to ambient air after being sprayed with water. This unique humidity-resist characteristic is very suitable for practical applications because once the devices are exposed to very humid environment; they can self-heal to high PCE again in a short time when they return to sunlight again. Moreover, to further prove the self-healing property, perovskite film with PEG was sprayed with water vapour for 60 s, then after self-healing, it was coated by Spiro-MeOTAD and Au electrode to fabricate into corresponding device, which showed photovoltaic performance comparable to its original PPSC devices without water spray ( Supplementary Fig. 10). The self-healing process was recorded by X-ray diffraction analysis (Fig. 5d). The peak at 12.7°reveals that PbI 2 phase formed when water vapour was sprayed on the film. After selfhealing, the PbI 2 peak disappeared and the perovskite film fully recovered to its original crystal phase. The self-healing can also be demonstrated by the absorption spectra (Fig. 5e), which presented the same absorbance and similar Urbach energy (B40 meV; ref. 29). The decomposition of perovskite phase is caused by the chemical reaction with water molecules [30][31][32] , which is manifested by the X-ray diffraction analysis of the perovskite film without PEG after water spray ( Supplementary Fig. 11).
According to the observations in Fig. 2a-c, we conjecture that these entangled PEG molecules are anchored on the surface of perovskite grains by hydrogen bonding, and the formation is illustrated in Fig. 6a. The -OH I-interaction had been evidenced in the work by Li et al 32 . We use liquid state 1 H NMR measurements to prove such interactions. The proton NMR spectra of three samples were compared (Fig. 6b): deuterated DMSO solution with MAPbI 3 (sample 1), with MAPbI 3 þ PEG (sample 2) and with PEG (sample 3). In the signals of sample 3, double methylene group linked to oxygen on both ends (-[O-CH 2 -CH 2 ] n -) is characterized by the peak at d ¼ 3.48 p.p.m. (Fig. 6b). An upfield chemical shift of Dd B À 0.18 p.p.m. with several splitting peaks is observed in sample 2. Such chemical shift can be explained by the hydrogen bond between MA þ and O in PEG chain, which weakens the influence of O on protons of methylene in -[O-CH 2 -CH 2 ] n -. Furthermore, the proton resonance signals of -NH 3 þ in sample 1 (peak at d ¼ 7.33 p.p.m.) shift towards upfield with DdB À 0.13 p.p.m. in sample 2 (Fig. 6b), which can also be attributed to the hydrogen bonds discussed above. Such interaction is also evidenced by a dissolution experiment devised by us ( Supplementary Fig. 12).
As presented in Fig. 6c, the improved stability and self-healing effect of the PPSC devices can be ascribed to the excellent hygroscopicity of the PEG molecules and their strong interaction with the perovskite. On one hand, the omnipresent PEG molecules can absorb water efficiently to form a compact moisture barrier around perovskite crystal grains with little water penetrating into the film. On the other hand, on water spray, the black perovskite film with PEG turned light yellow (PbI 2 forms) at first. However, due to the strong interaction between MAI and PEG, (NMR measurements and a dissolution experiment in Supplementary  Fig. 12), the MAI molecules were anchored by the nearby PEG molecules rather than escape away (process 1, 2 in Fig. 6c). After being kept away from water vapour, PbI 2 in the film reacted with nearby MAI to regenerate the perovskite MAPbI 3 phase, very similar to the two-step synthesis 2 (process 3 in Fig. 6c). The instant decomposition-regeneration mechanism explains the fast self-healing process in the PEG scaffold perovskite film.

Discussion
The hygroscopic PEG scaffold can stabilize the perovskite film, rendering the devices resistant to moisture with strong self-healing property. Hygroscopic polymer-scaffold architecture paved a new effective way in perovskite solar cell, to solve the hydrolysis problem in ambient air with greatly decreased package cost. The polymer-scaffold perovskite layer reinforces the architecture and improves strain tolerance. It may also facilitate production of flexible, wearable devices in the future. Future work will be focused on looking for cheap long-chain polymers which have stronger hygroscopicity and binding effect with MAI to protect the perovskite from decomposition in ambient environment. Moreover, by comparing the stability test result under light illumination ( Supplementary Fig. 8) and in dark (Fig. 5a), we found that light has a strong destructive influence on stability performance. One would expect that this influence comes from decomposition of perovskite film accelerated by light with the aid of hydrolysis by water molecules under illumination (reactive equation in Supplementary Fig. 11). However, after 72 h light soaking, the perovskite film with PEG was not destructed compared with that without PEG from X-ray diffraction spectra ( Supplementary Fig. 9). This indicated another important factor influencing the efficiency in addition to material decomposition. Hence, we speculate that I À ion migration 33-36 may be induced or enhanced by light. Ion migration may answer for the poor stability under illumination, no matter in the traditional architecture without PEG scaffold or in our PPSCs, which needs further investigation. Further improvement needs be focused on revealing the mechanism of ion migration in the long-term stability issue, as well as how to inhibit it in PPSCs, which would lead to much better performance in stability test under illumination with high humidity.

Methods
Device fabrication and device characterization. All chemicals were purchased from Sigma-Aldrich or J&K Scientific Ltd. unless otherwise stated. The photovoltaic devices were fabricated on FTO-coated glass (Pilkington, Nippon Sheet Glass). First, laser-patterned, FTO-coated glass substrates were cleaned by ultrasonic cleaned in deionized water, acetone and ethanol, followed by an ultraviolet treatment for 5 min. Compact layers were deposited on the substrates by spin coating titanium diisopropoxide bis(acetylacetonate) solution (75% in 2-propanol) diluted in ethanol (1:20, volume ratio) for two times and annealed at 450°C for 30 min. After cooling to room temperature, the substrates were transferred to a hot plate at 90°C before spin coating. CH 3 NH 3 I was synthesized according to the reported procedure 1   For NMR measurement, it is examined by Bruker 600 UltraShield using pulse signal. We use deuterated DMSO as solvent. The CH 3 NH 3 PbI 3 solution is prepared from CH 3 NH 3 I, PbI 2 (Aldrich; 1:1) for 53 and 153 mg ml À 1 solution and stirred in DMSO at 60°C for 12 h. CH 3 NH 3 PbI 3 solution mixed with PEG (Sigma-Aldrich) is prepared by adding 40 mg PEG (molecule weight: 20,000 Da) to 1 ml CH 3 NH 3 PbI 3 solution.
For X-ray diffraction measurement, flat PbI 2 and TiO 2 /PbI 2 nanocomposites were deposited on glass slides using the above-mentioned procedures. X-ray powder diagrams were recorded on an X'PertMPD PRO from PANalytical equipped with a ceramic tube (Cu anode, l ¼ 1.5406 Å), a secondary graphite (002) monochromator and a RTMS X'Celerator detector, and operated in BRAGG-BRENTANO geometry. The samples were mounted without further modification, and the automatic divergence slit and beam mask were adjusted to the dimensions of the thin films. A step size of 0.008 deg was chosen and an acquisition time of up to 7.5 min per deg. A baseline correction was applied to all X-ray powder diagrams to remove the broad diffraction peak arising from the amorphous glass slide.
Steady-state PL spectra were measured using a He-Cd laser (325 nm in wavelength) and green laser (514 nm in wavelength) guided by a micro-zone confocal Raman spectroscope (Renishaw inVia microRaman system) as the laser beam with a spot size diameter of 2 mm. The collected duration is B5 ms. The time-resolved fluorescence spectra were recorded with a high-resolution streak camera system (Hamamatsu C10910). We used an amplified mode-lock Ti: Sapphire femtosecond laser system (Legend, Coherent) and a two-stage optical parametric amplifier (OperA Solo, Coherent) to generate the pump beam with a repetition rate of 1 KHz. All the samples were excited by 517 nm at room temperature with 110 nJ cm À 2 .
The electrochemical impedance spectrum and cyclic voltammograms were measured using a potentiostat/galvanostat (SP-150, Bio-Logic, France). The frequency can be tuned from 0.1 Hz to 1 MHz. All the samples were measured under 10 mW cm À 2 .
The J-V characteristics were obtained using an Agilent B2900 Series precision source/measure unit, and the cell was illuminated by a solar simulator (Solar IV-150A, Zolix) under AM1.5 irradiation (100 mW cm À 2 ). Light intensity was calibrated with a Newport calibrated KG5-filtered Si reference cell. We use black mask to define the cells' area, and the masking effect is confirmed by testing the J sc with and without it, which has a 5% J sc difference. The J-V curves are tested from 1.5 to À 0.2 V with a scan velocity 500 mV s À 1 . The masked active area is 9 mm 2 .
Stability test. Unsealed PPSCs and PSCs were put near a humidifier to control its relative humidity B70%. For the stability test under dark, the cells were stored in a black box with 70% relative humidity, and were tested every 8 h. For the stability tests under continuous light illumination, light source ranges from 300 to 800 nm wavelength with 70 mW cm À 2 was used. J-V curves were recorded under AM1.5 light irradiation. The temperature is controlled B30°C. Three devices were measured in this way and demonstrate excellent repeatability. For the self-healing behaviour, we sprayed the water vapour for 60 s onto the resulting devices by humidifier. J-V curves were collected before and right after water vapour exposure under standard illumination AM1.5.