Steering device-quality perovskites with unique substrate tolerance

Polycrystalline perovskite lms fabricated on exible and textured substrates often are highly defective, leading to poor performance of perovskite devices. Finding substrate-tolerant perovskite fabrication strategies is therefore paramount. Herein, we show that a small amount of Cadmium Acetate (CdAc2) in PbI2 precursor solution results in nano-hole array lms and improves the diffusion of organic salts in PbI2, and promotes favorable crystal orientation and suppresses non-radiative recombination. Polycrystalline perovskite lms on exible substrate with ultra-long carrier lifetimes exceeding 6 μs are achieved. Eventually, a record PCE of 22.78% is obtained for single-junction exible perovskite solar cells (FPSCs). Furthermore, we nd this strategy is also demonstrated to be applicable for textured tandem solar cells. Our perovskite/silicon tandem solar cells (TSCs) with CdAc2 showed an eciency of 29.25% (0.5003 cm2), with a certied value being 23.07% (11.8792 cm2), respectively, the highest value reported for solution-based TSCs. Un-encapsulated TSCs based on CdAc2 maintain 109.78% of their initial eciency after operation at maximum power point under continuous one-sun illumination for 300 hours at 45 oC in N2 atmosphere. This strategy will provide facile access to high-eciency perovskite-based solar cells. anti-reection lm was attached on top of tandem solar cells. Fabrication and characterization of perovskite/textured silicon tandem solar cells. Silicon heterojunction (SHJ) bottom cells were fabricated on oat-zone double-side textured n-doped wafers. The ITO recombination junction (20 nm) was sputtered. Nanocrystalline NiOx (20 nm) is deposited by DC sputtering as the HTL on top of the recombination junction. The perovskite layer was deposited using the single-junction device deposition recipe, except a 1.82 M of the solution was spin-coated on the textured surface. 70 mL of perovskite precursor solution was dropped on top of Si substrate and deposited by one consecutive spin-coating step of 5,000 rpm for 50 s, respectively. During the spin-coating step, 150 mL solvent of ethyl acetate was quickly poured onto the substrate after 20 s. The perovskite-deposited lms were annealed on a hotplate at 100 °C for 15 min and 130 °C for 15 min. 20 nm of C60 was thermally evaporated on top of the perovskite layer. The SnO 2 buffer layer (25 nm) was deposited by the thermal atomic layer deposition (ALD) technique. Tetrakis(dimethylamino)tin (TDMASn) and H 2 O 2 were used as precursors for the ALD deposition at 100 °C and with 200 cycles. 85 nm of IZO top electrode (50 ohm/sq) was sputtered on the buffer layer with RF power of 30 W over a 2-inch target (90% In 2 O 3 /10% ZnO, 99.9% Plasmaterials). The top Ag and bottom Ag/Al contacts were evaporated through a shadow mask using an Angstrom thermal evaporator. Finally, PDMS antireection lms were added to complete the device. champion devices based on Control and CdAc 2 perovskite thin films.

Flexible substrates are not the only challenging surfaces for the growth of perovskite lms. The perovskite/silicon tandem solar cells (TSCs) are promising candidates to surpass the single-junction solar cells e ciency limit [42][43][44][45] . One of the biggest challenges for enabling the high-e ciency of TSCs is the compatibility of perovskite crystal growth with micron-sized pyramidal textured substrates 46,47 .
Currently, the highest PCE of TSCs based on the textured substrate has reached up to 27.4% 48 , but with rational manipulation of the perovskite crystal growth and passivation of deep level traps, more e cient tandem solar cells can be achieved.
In this work, we introduce Cadmium Acetate (CdAc 2 ) as a perovskite precursor solution additive to manipulate the perovskite crystal growth, achieving excellent optoelectronic properties on the exible substrate and textured silicon substrate. Using density functional theory (DFT) calculations, we con rm that CdAc 2 coordinates with lead iodide (PbI 2 ), forming an intermediate phase and slowing down the perovskite crystal growth rate, which eventually results in a better crystallinity of the nal perovskite lm.
The grain boundaries are also passivated by the crosslinking structure of Ac -, resulting in the excellent thermal durability of the FAPbI 3 lm. As a result, we demonstrate that with the synergistic effect of Cd ion doping and Ac-ion-assisted crystallization, the intrinsic defect density of perovskite polycrystalline thin lm is vastly reduced, and the carrier lifetime is increased to over 6 μs. We achieved PCEs of 24.46% and 22.78% for single-junction PSCs on the rigid and exible substrates. The latter is the highest reported for exible PSCs to date. We also applied the CdAc 2 in perovskite/silicon tandem solar cells and obtained an e ciency of 29.25% and a certi ed e ciency of 23.07% with aperture areas of 0.5003 and 11.8792 cm 2 , respectively. Operational stability showed that un-encapsulated TSCs based on CdAc 2 maintain 109.78% of their initial e ciency after continuous one-sun illumination for 300 hours at 45 o C in the N 2 atmosphere.
Formation of nano-hole array PbI 2 and FAPbI 3 lms The FPSCs in this work are based on FAPbI 3 and are fabricated by using a two-step solution method. In the two-step sequential deposition, the quality of the nal perovskite lms strongly depends on the PbI 2 lms. In general, a compact crystalline PbI 2 lm tends to form on a exible substrate. In this work, we developed an effective method to create PbI 2 lms with arrays of nano-sized pinholes at step one by introducing CdAc 2 into PbI 2 precursor solution. These pinholes allow for improved diffusion and intercalation of the organic ammonium cations in the second step of the deposition and nally produce perovskites with large and uniform grains.
To gain insights into how CdAc 2 may in uence the microscopic structure of the perovskites, we carried out several density functional theory (DFT) calculations (details in SI notes 1,2). Our results show that Cd 2+ can be incorporated in the PbI 2 lattice, substituting Pb 2+ , and Accan bind at the surfaces of PbI 2 , replacing I -. Because Acis a larger and more asymmetric ion, the direct consequence of the latter substitution is an increased distance between the PbI 2 layers and weakened binding between each layer by 0.05 eV/formula unit (Fig. 1a). As shown in supplementary Fig. 1, the introduction of Ac leads to a signi cant decrease in the peak intensity of PbI 2 . This nding suggests that less densely packed PbI 2 lms are more likely to form when Acis added to PbI 2 precursor solution, which may stimulate the formation of pinholes.
The presence of pinholes can potentially facilitate the diffusion and intercalation of FAI in the PbI 2 lm, where the perovskite is formed, resulting in more uniform perovskite lms, as illustrated in Figs. 1b and c.
Our calculations show that during the reactions of FAI with Ac --doped PbI 2 lms the Acis readily replaced by Ito produce a clean PbI 2 surface. At the next stage, the perovskite formation, the role of the Cd becomes important, as the transformation from a face-sharing structure in PbI 2 to a corner-sharing structure in perovskite requires the breaking of some of the metal-iodine bonds in the MI 2 layer and the formation of new metal-iodine bonds for the 3D perovskite. The Cd incorporated in the PbI 2 lattice structure (in the rst step) has direct consequences for the kinetics of the overall transformation. In the Cd-doped PbI 2 structure, creating an I vacancy near a Cd site (to facilitate ion transport) costs 0.02 eV/formula unit more energy than near a Pb site, whereas in the Cd-doped perovskite structure, it costs 0.03 eV/formula unit less energy. This indicates that the incorporation of Cd in the lattice increases the activation barriers for the reaction of converting the layered structure to a 3D perovskite structure. This can lead to a slower crystallization process and result in larger and more uniform perovskites grains.
To further elucidate how CdAc 2 affects the crystallization of the perovskite lms, grazing incidence X-ray diffraction (GIXRD) with various X-ray incident angles was employed, as shown in Figs. 1b and c. From the GIXRD, it is be found that the crystal orientation is signi cantly different at the surface and bottom of the perovskite lm with CdAc 2 . It is clearly shown that the perovskite lm with CdAc exhibits a prominent (111) orientation at the bottom, relative to other crystal planes' peaks, which indicates that the (111) crystals grew faster than differently oriented crystals with the presence of CdAc 2 . The (111) plane orientation is parallel to the substrate and favorable perovskite lattices 49   To gain insight into the properties of our perovskite lms, we carried out confocal photoluminescence (PL) mapping measurements to investigate the morphologies and optoelectronic properties of the perovskite lms with and without CdAc 2 (Fig. 2a). We observed an enhanced photoluminescence performance with better uniformity and stronger intensity for the perovskite with CdAc 2 . As shown in Fig. 2b, the PL peak wavelength distribution of the CdAc 2 treated perovskite lm is obviously narrower than the reference one, indicating a more uniform lm. Moreover, the CdAc 2 -treated perovskite lm exhibits higher PL mapping intensity than the control lm, showing better optical properties (Fig. 2c). This can also be con rmed by time-resolved photoluminescence (TRPL) measurements (Fig. 2d), in which the CdAc 2 -treated perovskite lm exhibits a much longer carrier lifetime (6514 ns) than the control lm (Supplementary Table S1), underlining the bene t of CdAc 2 on the perovskite lm by suppressed nonradiative recombination. To further investigate the charge transfer properties of perovskite lms, steadystate PL and absorption spectra were also conducted. As shown in Fig. 2e, the emission intensity of the perovskite lm is enhanced dramatically with the incorporation of CdAc 2 , indicating a decrease of nonradiative recombination, consistent with PL mapping and TRPL results. The ultraviolet-visible (UV-Vis) absorption spectra are slightly enhanced. We attribute this to the larger grain size and the decreased defect density.
The low defect density can be explained by two factors: (i) the better intercalation of FAI in the pinholerich PbI 2 lms in the rst step and slower crystallization (effect of Cd -) in the second step have overall led to large and high-quality perovskites and (ii) the Acspecies on the grain boundaries passivate the undercoordinated Pb 2+ , preventing further degradation, i.e. the formation and diffusion of I vacancy defects. Such passivation effect is veri ed by additional DFT calculations, where the relative binding strength of Ac-(compared to I-) to Pb is calculated by moving the target anion on the PbI 2 -terminated surface away at a distance of ~4 Å ( Supplementary Fig. 4). The energies show that Ac-has high binding energy equally to recently reported HCOO-2 , and is about 1 eV higher than that of I-( Supplementary Fig.  5).
To further investigate the in uence of the CdAc 2 on the properties of perovskite lm, we carried out the Xray Photoelectron Spectroscopy (XPS). In the Pb 4f spectra of both perovskite lms (Fig. 2f), we can observe two additional peaks (located at 136.4 and 141.3 eV) for the control perovskite lm, which correspond to metallic lead and originate from the unintended losses of iodine, leading to deeplevel defects and non-radiative recombination. Fortunately, it seems that the formation of metallic Pb can be suppressed by the addition of CdAc 2 . Besides, it can be estimated from Pb. We also con rm the effect of CdAc 2 on PbI 2 by 13C nuclear magnetic resonance (NMR) spectra ( Fig. 2g). With the addition of CdAc 2 , the peak of C=O moved from 167.2 ppm to 167.6 ppm, which indicates a strong interaction between CdAc 2 (C=O group) and Pb 2+ . Fig. 2h presents the TGA spectra of the PbI 2 solution and the different adducts. The dimethyl sulfoxide (DMSO) adduct of PbI 2 /DMF decomposes rapidly at 80 °C, while PbI 2 /DMF-DMSO-CdAc 2 decomposes at 139°C. It indicates that the PbI 2 /DMF-DMSO-CdAc 2 is more stable than the PbI 2 /DMF-DMSO due to the strong molecular interaction between CdAc 2 and PbI 2 .
Photovoltaic performance and stability of exible perovskite solar cells Based on the above results, we conducted a series of experiments fabricating rigid and exible perovskite solar cells with the structure of ITO/SnO 2 /Perovskite/Spiro-OMeTAD/Au (Fig. 3a). The  Fig. 14).
To provide more solid evidence on the charge recombination process, we measured the dependence of the V OC on the light intensity for the control and CdAc 2 -treated FPSCs. The slope deviation of V OC versus the natural logarithm of light intensity from kBT/q indicates the occurrence of non-radiative trap assisted recombination (monomolecular recombination), which has been reported to be a dominant recombination pathway in PSCs. As shown in Fig. 3g, the control FPSCs exhibited a slope of 1.91 KBT/e, while the CdAc 2 -treated FPSCs showed a much smaller slope of 1.32 KBT/e. This indicates that defect-assisted non-radiative recombination was effectively suppressed by using CdAc 2 , which is in accordance with the enhanced V OC and FF. The lower leakage dark current (Supplementary Fig. 15) also improved charge transport and decreased recombination loss in the CdAc 2 -treated FPSCs. To quantify the trap density in perovskite lms, we performed space-charge-limited-current (SCLC) of corresponding perovskite lms (Fig. 3h). As a result, the perovskite lm with CdAc 2 has a smaller VTFL than that of the control lm, indicating that lms with CdAc 2 can suppress the formation of defects.
To evaluate the carrier transport characteristics and output voltage in FPSCs, capacitance-voltage (C-V) measurements were further performed, and the relevant Mott-Schottky curves were presented in Fig. 3i. The built-in potential (1.166 V) of the CdAc 2 -treated FPSCs is extracted to be much larger than that of the control device (1.054 V), which means increased separation of photo-generated carriers and enhanced output voltage of FPSCs. A larger recombination resistance is also bene cial to suppress non-radiative recombination ( Supplementary Fig. 16).
Although tremendous progress has been made, long-term stability is still the critical hurdle to commercialization, especially the mechanical performance of FPSCs for their stretchable applications. On that account, we primarily investigated the mechanical stability of un-encapsulated FPSCs with and without CdAc 2 in this study. As depicted in supplementary Fig. 17, after 10000 cycles bend test with a bending radium 3 mm, the performance of FPSCs can still maintain about 80% of the initial value. However, the PCE for the control FPSCs decreased signi cantly, only providing about 11% of the original PCE after the 1800 cycles bend test. Furthermore, we also conducted cold and hot cycle tests. After 24 cold and hot cycles, the CdAc 2 treated FPSCs provide 87.28% of their initial e ciency, while the control device only maintains 21.59% of its initial e ciency after 13 cold and hot cycles. Based on these results, both bending stability and thermal stability were enhanced by the bene cial effect of the CdAc 2 .

E cient monolithic perovskite/silicon tandem solar cells based on a textured substrate
To further highlight the positive effect of this strategy for perovskite crystallization improvement, we fabricated perovskite/silicon TSCs based on a textured substrate, where the crystal growth would be more complicated than the polished one. The perovskite used here was and full coverage on the textured silicon wafer (Fig. 4d). But the CdAc 2 treated perovskite lm exhibits a much larger grain size. Fig. 4e shows the J-V characteristic of the champion perovskite/silicon TSCs with 29.25%, which is the highest e ciency reported for perovskite/silicon TSCs based on textured silicon by spin-coating method (Supplementary Table 3). The statistics of the PCE of TSCs without and with CdAc 2 demonstrated that the CdAc 2 -based TSCs are more reproducible and have a higher Voc and FF ( Supplementary Fig. 22). In Fig. 4f, the EQE of the tandem device is presented. The extracted current density of the top and bottom cells are 20.05 mA/cm 2 and 19.85 mA/cm 2 , respectively, which match well with J-V curve. In addition, a large-area perovskite/silicon TSC of the same design with an active area of 11.8792 cm 2 was successfully fabricated. A PCE of 23.23% was achieved with the CdAc 2 device (Certi ed with a PCE of 23.07% in supplementary Fig. 23 and Table 4), whereas that of the control device was 19.05% in Fig. 4g. We also examined the long-term operational stability of un-encapsulated TSCs with CdAc 2 , which was evaluated under one sun continuous irradiation at maximum power point (MPP) tracking in the N 2 atmosphere. As shown in Fig. 4h, the CdAc 2 treated TSCs maintain 109.78% of the initial e ciency after 300 h. We ascribe the high stability of the TSCs to two main reasons: i) the substitution of Cd indeed slow down the crystallization rate and suppresses the formation of I vacancy and phase segregation 53 . ii) Moreover, as discussed earlier that our DFT calculations show that the Ac anions coordinate strongly with Pb 2+ and passivates I vacancy on the grain boundaries of the perovskite lms. This evident coordination may also help slow down the perovskite growth process, thus resulting in a larger grain size of the perovskite lm with much less trap density in TSCs.

Conclusion
In summary, we have demonstrated a simple and effective approach by using the multifunctional additive CdAc 2 to obtain high-performance FPSCs and TSCs. The strong interactions between CdAc 2 and perovskite components could slow down the grain growth rate, facilitating the growth of device-quality perovskite lms with larger grain size, preferred orientation, reduced defects density, and superior electronic properties. Meanwhile, the residual Ac in the perovskite lms can passivate ionic defects and act as a molecular lock to prevent thermal degradation. Finally, the residual Acin the nal perovskite lms can also passivate I vacancy defects and act as a molecular lock to prevent thermal degradation. As a result, we achieved a champion PCE of 22.78% based on a exible substrate, with excellent thermal and mechanical stability. An enhanced PCE of 29.25% can be obtained on textured perovskite/silicon TSCs. We also achieved a certi ed e ciency of TSCs of 23.07% on large area 11.8792 cm 2 . The CdAc 2 treated TSCs exhibited super stability and retained 109.78% of their initial e ciency under one sun continuous irradiation for 300 hours in an N 2 atmosphere without encapsulation. This work paves the way for the large-scale fabrication of stable and exible perovskite solar cells.

Methods
Materials. All materials were purchased from either Sigma-Aldrich or Alfa Aesar and used without further puri cation unless otherwise stated.
FPSCs device fabrication. The pre-patterned ITO-coated PET substrates were cleaned by ultra-sonication with detergent, deionized water, acetone, and isopropanol sequentially, then dried with N 2 . After that, the ITO substrates were treated by UV zone for 20 min. A 20-nm-thick SnO 2 electron transport layer was Device characterizations. Scanning electron microscopy results were obtained using eld emission scanning electron microscopy (SEM) (Jeol JSM-6700F). The transmittance, re ectance, and absorbance were characterized with a UV-vis-NIR spectrophotometer (Cary 5000, VARIAN). The surface AFM and KPFM were carried out using an ICON2-SYS. The charge transfer kinetics of perovskite lms were measured by steady-state photoluminescence (PL) and time-resolved PL (TRPL) with the laser wavelength of 475 nm and power of 0.2 mW, respectively (Edinburgh FS5). X-ray photoelectron spectroscopy (XPS) and ultraviolet photoelectron spectroscopy (UPS) data were obtained by a multifunctional photoelectron spectrometer (Escalab 250Xi, U.K.      Supplementary Files