Low-temperature-processed efficient semi-transparent planar perovskite solar cells for bifacial and tandem applications

Semi-transparent perovskite solar cells are highly attractive for a wide range of applications, such as bifacial and tandem solar cells; however, the power conversion efficiency of semi-transparent devices still lags behind due to missing suitable transparent rear electrode or deposition process. Here we report a low-temperature process for efficient semi-transparent planar perovskite solar cells. A hybrid thermal evaporation–spin coating technique is developed to allow the introduction of PCBM in regular device configuration, which facilitates the growth of high-quality absorber, resulting in hysteresis-free devices. We employ high-mobility hydrogenated indium oxide as transparent rear electrode by room-temperature radio-frequency magnetron sputtering, yielding a semi-transparent solar cell with steady-state efficiency of 14.2% along with 72% average transmittance in the near-infrared region. With such semi-transparent devices, we show a substantial power enhancement when operating as bifacial solar cell, and in combination with low-bandgap copper indium gallium diselenide we further demonstrate 20.5% efficiency in four-terminal tandem configuration.

T he emerging perovskite solar cells have received increasing attention due to the high efficiency, easy processing and potentially low cost [1][2] . Although the power conversion efficiency of perovskite solar cells have soared up to 20.1% (ref. 3), there is still ample room for further efficiency improvement through novel concepts [4][5][6] . An effective approach to increase the efficiency and lower the production cost is to make semi-transparent solar cells that can convert solar energy into electricity from both front and rear side of the device 7 . The bifacial cell concept has been implemented in various kinds of solar cell technologies [8][9][10][11][12] , and up to 50% output power enhancement has been demonstrated in Si wafer bifacial modules by collecting the albedo radiation from surroundings 13 . In addition, the semi-transparent perovskite solar cells hold great promise for applications in tandem solar cells 14 , photon energy upconversion 15 , building-integrated photovoltaics 16 , wearable electronics 17 , powering sensors and electronic gadgets in homes.
There are still many challenges hindering the realization of high-performance semi-transparent perovskite solar cells. The state-of-the-art perovskite solar cells 3,18-22 employ high-temperature-processed (B500°C) TiO 2 (mesoporous and compact) as electron transporting layer (ETL), which is incompatible with monolithic tandem or flexible solar cells on plastics. Thus, it is desirable to develop a planar structure that allows low-temperature processing. It is well known that perovskite solar cells, particularly in planar configuration with TiO 2 as ETL, suffer from pronounced hysteresis in the current density-voltage (J-V) curve, which often results in overestimated device performance 23 . Therefore, it is imperative to eliminate the hysteresis, which is not only important to ensure a stable performance but also to unambiguously evaluate the real device efficiency. [6,6]-phenyl-C 61 -butyric acid methyl ester (PCBM) has been proven to assist in eliminating the J-V hysteresis in inverted device configuration 24 . Several efforts have been made in positioning PCBM in regular device structure; however, the hysteresis phenomenon still remains 25,26 . For bifacial and tandem solar cell applications, it is crucial to replace the commonly used metal contacts in perovskite cells by highly transparent conducting electrodes, which allow sunlight incident from front and rear side of the device and to transmit the photons with energy below the bandgap of the perovskite. Previously, thin layers of Au (ref. 27), carbon nanotubes 28 and PEDOT:PSS 29 have been explored as transparent contact for perovskite solar cells, while the efficiencies are generally below 10% and these contacts featured strong absorption in near infrared (NIR). Efficient semi-transparent devices have been realized with graphene rear electrode; however, all the devices show strong hysteresis in current density-voltage curves 30,31 . Recently, sputtered transparent conductive oxides (TCO), including indium tin oxide (ITO) 32 , ZnO:Al 33 and indium zinc oxide (IZO) 34 , have been reported with highest efficiency of up to 12.1% for a semi-transparent cell and 19.5% in perovskite-CIGS (copper indium gallium diselenide) four-terminal configuration 33 . Bailie et al. 35 developed a 12.7% efficiency semitransparent perovskite solar cell employing Ag nanowires, and demonstrated 18.6 and 17% tandem efficiency by combining with CIGS and low-quality silicon bottom cell, respectively. Despite the high NIR transparency and low sheet resistance of Ag nanowires, the complicated deposition method leads to large efficiency variation and poor reproducibility, and the fast diffusion of Ag into Spiro-OMeTAD and perovskite results in degradation of device stability and performance 36 . Therefore, the implementation of a suitable transparent electrode material and deposition process is the key and challenge to high-efficiency semi-transparent perovskite solar cells. An ideal transparent electrode for perovskite solar cells should meet the following requirements: superior transparency in visible and NIR region, low sheet resistance, robust adhesion, good chemical stability and compatibility with adjoining layers. As the perovskite and Spiro-OMeTAD are temperature and solvent sensitive 35 , a lowtemperature and dry process would be favoured for the deposition of transparent electrodes. In this regard, the roomtemperature sputtering deposited TCO would be an ideal candidate.
Here we report a low-temperature (r50°C)-processed hysteresis-free high-efficiency semi-transparent planar perovskite solar cell. The hysteresis is eliminated through a PCBM interlayer and high-quality perovskite absorber enabled by the hybrid thermal evaporation-spin coating method. High-mobility hydrogenated indium oxide (In 2 O 3 :H) is employed as transparent rear electrode through a room-temperature radio-frequency magnetron sputtering, which enables a semi-transparent device with steady-state efficiency of 14  transmittance between 800 and 1,150 nm. Such a device when operated as a bifacial cell yields an additional 13.5% power output improvement, and 20.5% efficiency is demonstrated using perovskite top cell and CIGS bottom cell in four-terminal tandem configuration.

Results
Low-temperature-processed planar perovskite solar cells. Figure 1 shows a schematic drawing and cross-sectional scanning electron microscopy (SEM) images of the perovskite absorber deposited by hybrid thermal evaporation-spin coating method. The high-temperature-processed TiO 2 is replaced by B100 nm compact ZnO layer deposited through radio-frequency magnetron sputtering at room temperature, followed by spin coating of B50 nm PCBM. The room-temperature-processed ETL is attractive for low-cost, lightweight and flexible plastic substrates, and compatible with monolithic tandem fabrication. As illustrated in Fig. 1a, the perovskite layer is formed through a hybrid evaporation solution approach, where a compact PbI 2 layer is initially deposited by thermal evaporation 37 . This method allows a uniform and compact PbI 2 deposited over large area. Then CH 3 NH 3 I in isopropanol solution is added by spin coating, followed by thermal annealing at 50°C for 2 h on hotplate. Afterwards, B200 nm Spiro-OMeTAD is spin coated on top of the perovskite. Finally, the cell is finished by evaporating 60 nm Au for opaque reference solar cells. For semi-transparent solar cells, thermally evaporated MoO 3 and room-temperaturesputtered In 2 O 3 :H are employed as rear contact. Thus, the highest processing temperature during the opaque and semi-transparent device fabrication is not higher than 50°C. More detailed fabrication process information can be found in the Methods section.
For planar perovskite solar cells, a uniform and compact perovskite layer is essential for achieving high efficiency 38 . As ZnO conformally covers the rough fluorine-doped tin oxide (FTO) surface, the deposition of PCBM smoothens the surface where the perovskite is grown (Fig. 1b). Consequently, the ARTICLE thermally evaporated PbI 2 exhibits a compact and uniform layer over large area (Fig. 1c). Figure 1d,e presents the cross-sectional SEM images of perovskite layer before and after annealing, showing that the perovskite layer retained the desired compact morphology from parent PbI 2 . With thermal annealing, the grain size increases considerably from several tens of nanometres to values comparable to the film thickness (Fig. 1d,e), which might be beneficial to the transport and recombination properties of the absorber.
The herein introduced perovskite preparation method offers several merits. In previously reported two-step methods where PbI 2 is spin coated from solution, it is quite challenging to control the morphology of the PbI 2 film deposited from solution on planar substrates 39 . The low solubility of PbI 2 in the commonly used solvent N,N-dimethylformide 40 could limit the thickness variation and the choice of the underlying substrates where PbI 2 is deposited. We overcome these challenges by using a simple thermal evaporation method, which can deposit a compact and uniform PbI 2 layer with precisely controlled thickness over large area. The room-temperature vapour deposition has no constraints on the underlying substrates, which opens up numerous possibilities for exploring other solvent-sensitive carrier-selective contacts and novel cell structure. For instance, this process allows the introduction of PCBM below the perovskite layer, which is crucial to eliminate the J-V hysteresis 24 , but difficult to realize in solution process as the commonly used polar solvents, such as N,N-dimethylformide, dimethylsulfoxide or g-butyrolactone, for perovskite formation are detrimental to thin layers of PCBM.
Planar perovskite solar cells without J-V hysteresis. The hybrid thermal evaporation-spin coating method enables us to investigate the influences of PCBM on perovskite microstructure and device performance in regular planar configuration. Figure 2a,b presents the cross-sectional SEM images of the planar perovskite solar cell with and without PCBM layer, respectively. Other than PCBM, the devices are fabricated by an identical process with perovskite layer grown from 120 nm PbI 2 (estimated by quartz microbalance) and 40 mg ml À 1 CH 3 NH 3 I solution. It can be seen from the SEM images that the perovskite layer shows considerable surface roughness and thickness non-uniformity when grown directly on ZnO, which leads to low-resistance shunting paths and insufficient light absorption. A uniform and compact perovskite layer with large grain size is obtained when grown on PCBM. The microstructural difference in perovskite layers is mainly attributed to the morphological differences in PbI 2 layers, as shown in Supplementary Fig. 1. If PbI 2 is grown on ZnO directly, porous layers comprising numerous nanoplates are obtained 41 . This could form lots of grain boundaries and defects after the conversion into perovskite. It is important to note that high-efficiency devices produced by the here described process always contain residual PbI 2 as shown in Supplementary Fig. 2. The existence of residual PbI 2 is also observed in many highefficiency devices reported in literature and several beneficial effects, for example, grain boundary passivation, hole-blocking effect and so on, have been proposed 42,43 . Figure 2c shows the J-V characteristics of the corresponding planar perovskite solar cells under simulated AM1.5G irradiation. Owing to the hysteresis in J-V measurements 31 , it is crucial to report both forward (short circuit to forward bias) and backward (forward bias to short circuit) measurements along with the measurement conditions. The photovoltaic parameters are summarized in Supplementary Table 1. The device with PCBM layer shows an open circuit voltage (V oc ) of 1.101 V, short circuit current density (J sc ) of 17.6 mA cm À 2 and fill factor (FF) of 74.9%, resulting in power conversion efficiency (Z) of 14.5% in forward scan. In backward scan a V oc of 1.103 V, J sc of 17.5 mA cm À 2 , FF of 74.6% and Z of 14.4% are obtained. The J-V curves from forward and backward scans coincide well, indicating no J-V hysteresis is observed in the devices. The J sc is confirmed by external quantum efficiency (EQE) measurement ( Fig. 2d) with an integrated J sc of 17.4 mA cm À 2 under AM1.5G spectrum. Furthermore, the efficiency measured at the maximum power point (MPP), that is, the steady-state output of the device, is shown in Fig. 2e. The cell exhibits a steady-state efficiency of 14.4% and current density of 15.7 mA cm À 2 at MPP under continuous illumination. This is consistent with the J-V and EQE measurements. The device without PCBM layer generally shows a lower V oc , J sc and FF with pronounced J-V hysteresis, and the steady-state efficiency lies between forward and backward efficiency. The integrated J sc of 12.1 mA cm À 2 is also lower than that derived from the J-V measurement.
Time-resolved photoluminescence decays are presented in Fig. 2f to probe the effect of PCBM on carrier dynamics in perovskite solar cells. In both cases the decay tails allow to estimate a minority carrier lifetime of B80 ns (ref. 44). Furthermore, it can be seen that the addition of PCBM leads to a nonlinear initial PL transient, which can be explained by charge-separating fields 45 . This indicates that the application of PCBM promotes a working junction without the need of initial biasing. Therefore, the elimination of J-V hysteresis and enhanced photovoltaic performance primarily stem from improved interface and junction formation and from the highquality perovskite layer with enhanced crystallinity, uniform and compact morphology with large grain size.
Semi-transparent planar perovskite solar cells. For semitransparent cells, the opaque Au is substituted by In 2 O 3 :H transparent electrodes. In 2 O 3 :H is a high-mobility TCO material with superior visible and NIR transmittance 46 , and excellent thermal and chemical stability for solar cell applications were reported 47 . Deposition of In 2 O 3 :H by sputtering at room temperature without post-annealing results in layers with reasonably high mobility of 51.3 cm 2 V À 1 s À 1 and sheet resistance of 25.7 O per square for 149 nm thick (determined from profilometer) amorphous layers. The effect of postannealing treatment on the electrical properties (sheet resistance, electron mobility and electron density) and optical properties (transmission and absorption) of In 2 O 3 :H films on glass is presented in Supplementary Fig. 3. The amorphous nature of the as-sputtered In 2 O 3 :H is confirmed by X-ray diffraction ( Supplementary Fig. 4). The radio-frequency sputtering method is a high-throughput and industrial scale method. However, high energetic ion bombardment during sputtering can damage the underlying layers. This is especially the case for organic layers such as Spiro-OMeTAD. Two different TCOs, that is, ZnO:Al and In 2 O 3 :H, are deposited by radio-frequency magnetron sputtering with different extent of ion bombardment and compared as transparent electrode.
The degree of ion bombardment is more severe during the deposition of ZnO:Al than that in In 2 O 3 :H due to the different sputter geometries, lower total pressure and shorter target-tosubstrate distance 48 , as illustrated in Supplementary Fig. 5. In both devices, 8.7 nm MoO 3 is evaporated as buffer layer, and perovskite is grown from 120 nm-thick PbI 2 precursor. As can be seen in Fig. 3a,d, the average V oc , J sc , FF and Z of devices with ZnO:Al are much lower than that of the ones with In 2 O 3 :H, and the s.d. is also larger in case of ZnO:Al. Notably, a perovskite solar cell with 10.1% efficiency (Supplementary Fig. 6) is achieved by directly depositing In 2 O 3 :H on top of Spiro-OMeTAD. This is the highest value with direct deposition of TCO on top of the hole transporting layer (HTL) as back electrode without buffer layer, indicating that direct sputtered TCO might allow to achieve highefficiency if the ion bombardment is suppressed during deposition.
In the current device architecture and deposition procedure, the use of MoO 3 proves still to be beneficial. Figure 3 shows the photovoltaic and optical properties as a function of the MoO 3 thickness as well as the schematic cell structure of semitransparent planar perovskite solar cell. A general increase of all the photovoltaic parameters combined with a substantial increase in uniformity is observed with increasing MoO 3 thickness up to 35 nm. Further, it is shown that the J sc is less sensitive to MoO 3 thickness variation if the PbI 2 precursor thickness is increased to 140 nm while the FF and V oc exhibit the same trend as devices grown from 120 nm PbI 2 precursors. For MoO 3 thicknesses above 35 nm no increase is observed anymore. The highest V oc and FF in semi-transparent cells are comparable to opaque reference ones, indicating a reasonable trade-off between minimized sputtering damage and added series resistance with 35 nm MoO 3 thickness. In addition to the photovoltaic performance, the MoO 3 thickness also affects the transmission and reflection spectra as shown in Fig. 3e. The transmission shows a decreasing trend with increasing MoO 3 thickness between 800 and 940 nm, while an opposite trend with thickness variation between 940 and 1,200 nm. The device without MoO 3 shows a peak transmission of 82%, while device with 35 nm MoO 3 shows the highest average transmission in 800-1,150 nm region. Therefore, an optimal 35 nm-thick MoO 3 layer is deposit on Spiro-OMeTAD before In 2 O 3 :H deposition in terms of photovoltaic and optical properties in the following work. Figure 4 summarizes the microstructure, transmission and photovoltaic performance of a semi-transparent planar perovskite solar cell produced as described above. The cross-sectional SEM image in Fig. 4a shows a cell structure of FTO/ZnO/PCBM/ CH 3 NH 3 PbI 3 /Spiro-OMeTAD/MoO 3 /In 2 O 3 :H. An electron beam-evaporated Ni-Al grid is applied for better charge carrier collection in cells with area above 0.5 cm 2 . The cell area is defined by mechanical scribing down to the FTO layer. The perovskite grown from 140 nm PbI 2 and 45 mg ml À 1 CH 3 NH 3 I solution shows a flat and dense layer with grain size comparable to the film thickness. Figure 4b displays the transmission, reflection and absorption of the semi-transparent cell. The transmission through the whole device shows a peak of 77% at 940 nm and an average of 72% between 800 and 1,150 nm. The high sub-bandgap transmission is attributed to the low free-carrier absorption in high-mobility In 2 O 3 :H (ref. 48). In addition, the photograph of the semi-transparent device in the inset of Fig. 4b shows a decent transmission in visible region as the picture behind the device can be clearly recognized. The hatched area in Fig. 4b indicates optical losses due to insufficient absorption in the perovskite layer. This means there is still much room to increase the current in the semi-transparent cell by optimizing the (optical) thickness of perovskite layer. Figure 4c,d presents the J-V curves and EQE spectra of the semi-transparent planar perovskite solar cell illuminated from front (FTO) and rear (In 2 O 3 :H) side. The forward and backward scans show negligible J-V hysteresis. The corresponding photovoltaic parameters are summarized in Supplementary Table 1. The semi-transparent device shows a V oc of 1.104 V, J sc of 17.4 mA cm À 2 , FF of 73.6% and Z of 14.1% illuminated from FTO, and V oc of 1.105 V, J sc of 12.2 mA cm À 2 , FF of 70.6% and Z of 9.5% illuminated from In 2 O 3 :H, as shown in Fig. 4c. The V oc remains the same regardless of the illumination side, while higher J sc and FF are observed when illuminated from FTO side. The EQE spectra are consistent with the J sc discrepancy in J-V curves, where rear illumination shows a lower EQE, particularly below 430 nm. This distinct difference can be well explained by the strong parasitic absorption of Spiro-OMeTAD ( Supplementary  Fig. 7). Strong interference and the shading effect of metal grid also contribute to the lower current density when illuminated from rear side. The semi-transparent cell shows a high V oc of 1.1 V, comparable to the reference cell. The FF is slightly reduced due to the lower conductivity of the In 2 O 3 :H/Ni-Al grid contact   compared with Au. Generally, the current density in the semitransparent cell is lower than that in the opaque cell when the perovskite layer is grown from same PbI 2 thickness (120 nm) as shown in Supplementary Fig. 8. To verify the efficiency measured from J-V, the steady-state output at MPP is measured and presented in Fig. 4e. On illumination, the current density and efficiency increase within 3 s (ramp up of the MPP tracker) to 16.4 mA cm À 2 and 14.2% for front illumination, and 10.8 mA cm À 2 and 9.6% for rear illumination, respectively. The steady-state power conversion efficiency of 14.2% is the highest reported value for semi-transparent planar perovskite solar cell to date, which is particularly remarkable due to the simultaneously high transparency in NIR region. In addition, the semitransparent cell with In 2 O 3 :H showed improved air stability compared with the opaque cell with Au in a preliminary stability experiment, as shown in Supplementary Fig. 9. In the following, two of many applications of high-efficiency semi-transparent perovskite solar cells are discussed in more detail.
Bifacial planar perovskite solar cell. Bifacial solar cells are capable of converting sunlight incident from the front and the rear side of the cell into electricity, therefore the power output could be improved if highly reflective surrounding is present behind rear side of the cell. To illustrate how the power output gain works, we simply put a white paper from a commercial notebook as reflective surrounding under rear side of the cell during the MPP measurement. As shown in Fig. 4f, the power output quickly increased from original 14.1 to 16.0 mW cm À 2 after adding the white paper behind the solar cell, while it dropped back to 14.1 mW cm À 2 after removing the white paper. Using white paper as reflective surrounding gives 13.5% performance enhancement. By using a more reflective background and proper arrangement of solar modules in a system, the output power gain can be further improved up to 50% as demonstrated in silicon technology by Ceuvas et al. 13 .
Four-terminal perovskite-CIGS tandem solar cell. A fourterminal mechanically stacked tandem device is demonstrated based on perovskite as top subcell and CIGS as bottom subcell, as schematically shown in Supplementary Fig. 10. The CIGS absorber layer is fabricated by a multistage co-evaporation process 49 . In a four-terminal tandem device, the top and bottom subcells are individually processed and mechanically stacked together. As the top cell and bottom subcell are electrically independent, two MPP trackers can be used to ensure optimum operation of each subcell at any condition. This also means that the efficiency of each subcell can be evaluated separately, and the sum of the two gives the efficiency of the device in tandem configuration. In this work, the performance of the bottom CIGS is directly measured with perovskite as a filter under standard illumination ( Supplementary Fig. 10). The cell area of the bottom CIGS is defined by a laser-scribed shadow mask with aperture of 0.213 cm 2 inserted between the top and bottom subcells. The J-V curves and EQE spectra of the top and bottom subcells are displayed in Fig. 5, and the photovoltaic parameters are summarized in Table 1. For comparison, the photovoltaic performance of the CIGS without perovskite filter is also presented. After adding the perovskite top cell as filter, the J sc of the bottom CIGS decreased as expected due to reduced light intensity. The FF and V oc are only marginally reduced, resulting in 6.3% efficiency for the CIGS bottom cell. Together with the 14.2% efficiency of the perovskite top, this result in 20.5% efficiency for the device in tandem configuration, which is a substantial improvement compared with each single-junction efficiency as well as to what was reported earlier for polycrystalline thin-film solar cells in tandem configuration [32][33][34][35][50][51][52] .
Currently, the transmission is still limited by the strong parasitic absorption in the front FTO electrode. The lowtemperature process presented here facilitates the use of more transparent TCOs, such as In 2 O 3 :H or indium zinc oxide. The transmission and absorption spectra of In 2 O 3 :H, ZnO:Al and commercially available FTO and ITO are compared in Supplementary Fig. 11. The In 2 O 3 :H has the lowest absorption among all the TCOs investigated here between 300 and 2,000 nm. This implies there is much room in further lowering the parasitic absorption of semi-transparent perovskite solar cell in NIR region if front FTO is replaced by high-mobility In 2 O 3 :H. Since both the fabrication process for the CIGS solar cell and the for the perovskite cell are compatible to deposition on polymer films, lightweight and bendable polycrystalline tandem devices are in reach. Optimization of the bandgap of top and bottom cells, application of anti-reflection coating and optical coupling media can lead to even higher tandem efficiency in the near future as discussed in more detail by Kranz et al. 33 .

Discussion
In conclusion, a low-temperature-processed high-efficiency semitransparent planar perovskite solar cell without J-V hysteresis is developed by using sputtered hydrogenated indium oxide (In 2 O 3 :H) as transparent rear electrode. The high-temperatureprocessed (B500°C) TiO 2 is replaced by room-temperaturesputtered ZnO, and the perovskite is prepared by a combined thermal evaporation of PbI 2 and spin coating of CH 3 NH 3 I followed by 2-h annealing at 50°C, thus making the maximum processing temperature as low as 50°C. The hybrid thermal evaporation-spin coating method enables the introduction of PCBM in regular device configuration, which results in a highquality perovskite layer and better electron collection, thus leads to a 14.5% hysteresis-free planar perovskite solar cell. Highmobility In 2 O 3 :H is employed as transparent rear electrode deposited by sputtering at room temperature, which enables a semi-transparent planar perovskite solar cell with high steadystate efficiency of 14.2% and simultaneous 72% average transmission between 800 and 1,150 nm. Finally, we show a 13.5% output power enhancement when the semi-transparent cell functioned as bifacial solar cell and demonstrate 20.  Radio-frequency sputtering of aluminium doped zinc oxide. Aluminium-doped zinc oxide (ZnO:Al) layers were deposited in a high-vacuum sputtering system by radio-frequency sputtering of ceramic ZnO (containing 2 wt% Al 2 O 3 ) targets. The sputtering deposition consist of a 5-min deposition ramp up from 0.6 to 2.5 W cm À 2 and followed by 5 Â 3 min at 2.5 W cm À 2 under 20 sccm Ar and 0.29 sccm Ar/O 2 (3 mol% O 2 ). There is a 30-min waiting time during each step to alleviate the temperature effect.
X-ray diffraction measurements. X-ray diffraction patterns were obtained in Bragg-Brentano geometry by using a X'Pert PRO y-2y scan (Cu-K a1 radiation, l ¼ 1.5406 Å) from 10 to 60°(2y) with a step interval of 0.0167°.
Scanning electron microscopy. The cross-sectional images of the samples were investigated with a Hitachi S-4800 using 5 kV voltage and 10 mA current. A thin layer of Pt was coated on top of the sample to avoid charging effect.
Ultraviolet-visible spectroscopy. The transmission (T) and reflection (R) spectra were acquired using a ultraviolet-visible-NIR spectrophotometer (Shimadzu UV-3600) equipped with an integrating sphere. Absorption A is calculated using the following formula: A ¼ 1 À T À R.
Time-resolved photoluminescence. For time-resolved photoluminescence measurements, a 639 nm diode laser with pulse duration of 92 ps was used as excitation source. Pulse repetition rates were around 1 MHz and the number of photons per area and pulse was B1 Â 10 13 photons per cm 2 . The spectrally integrated luminescence decay was measured using a Picoquant PMA-C 192-M photomultiplier time and a Picoquant TimeHarp 260 digitizing system with 50 ps time channel width.
Hall effect measurements. Hall effect measurements of In 2 O 3 :H films were performed with a HMS 3,000 Hall effect measurement system at room temperature in the dark using van der Pauw geometry. The electron density and electron mobility were calculated from Hall measurements.
Solar cell performance characterization. The current density-voltage characteristics of perovskite solar cells were measured under standard simulated AM1.5G illumination using a Keithley 2,400 source meter. The illumination intensity was calibrated to 1,000 W m À 2 using a certified single crystalline silicon solar cell. The J-V measurement is performed in both forward (form À 0.1 V to 1.2 V) and backward (from 1.2 V to À 0.1 V) directions separately without any pretreatment (for example, light soaking, holding at forward bias for certain time and so on). The scan rate and delay time are 0.3 V s À 1 and 10 ms, respectively. The external quantum efficiency of the devices were measured with a lock-in amplifier.
The probing beam was generated by a chopped white source (900 W, halogen lamp, 260 Hz) and a dual grating monochromator. The beam size was adjusted to ensure that the illumination area was fully inside the cell area. A certified single crystalline silicon solar cell was used as the reference cell. White light bias was applied during the measurement with B0.1 sun intensity. The steady-state efficiency as a function of time was recorded using a maximum power point tracker, which adjusts the applied voltage in order to reach the maximum power point (perturb and observe algorithm). The starting voltage is set to be 0.1 V.