Efficient and stable perovskite-silicon tandem solar cells with copper thiocyanate-embedded perovskite on textured silicon

Abstract

Monolithic perovskite/silicon tandem solar cells have achieved promising performance. However, hole transport layers that are commonly used for the perovskite top cell suffer from defects, non-conformal deposition or de-wetting of the overlying perovskite on the textured silicon bottom cells. These issues detrimentally affect device reproducibility and scalability, and thus commercialization. Here we address these challenges through the co-deposition of copper(I) thiocyanate and perovskite, where effective perovskite grain boundary passivation and efficient hole collection are simultaneously achieved by the embedded copper(I) thiocyanate, which creates local hole-collecting contacts. Fabricated monolithic perovskite/silicon tandem devices achieve a certified power conversion efficiency of 31.46% for 1 cm² area devices. Aside from good reproducibility and scalability, our tandem cells exhibit excellent stability, maintaining 93.8% of their initial power conversion efficiency after about 1,200 h of maximum power point tracking at 45 °C, and 90.2% after over 1,000 h of damp-heat testing at 85 °C and 85% relative humidity.

You have full access to this article via your institution.

Download PDF

Main

Monolithic perovskite/silicon tandem solar cells have demonstrated power conversion efficiencies (PCEs) of above 33%, underlining their promise as a future high-performance photovoltaic technology^1,2. State-of-the-art tandems are usually prepared with perovskite top cells in the classic p–i–n architecture, with a continuous hole transport layer (HTL) between the indium tin oxide (ITO) recombination and perovskite layers, implying that electron collection occurs on their sunward side^3,4. Meanwhile, the application of double-sided textured silicon bottom cells is the goal for industrialization of these tandem cells^5,6.

Considerable efforts have focused on perovskite deposition on textured silicon cells in the past few years, either by a two-step sequential deposition enabling conformal coverage⁵ or a one-step solution process (via spin, blade or slot-die coating)^7,8,9,10 where relatively thicker perovskite layers are required to completely cover the micrometre-sized pyramids. Moreover, the impact of the HTLs such as nickel oxide (NiO_x), poly(triarylamine) (PTAA), 2,2′,7,7′-tetra(N,N-di-p-tolyl)amino-9,9-spirobifluorene (Spiro-TTB) and self-assembled monolayers (SAMs) at the silicon surface towards the fabrication and performance of the overlying perovskite top cell needs further elucidation as they all come with challenges that hinder commercial success. Consider NiO_x for example, although it obtains excellent conformality on textured substrates, it is often defective and causes degradation at the interfaces with perovskite layers, which in turn brings device performance and stability losses^11,12, mandating interface passivation¹³. Poly(triarylamine) is a common HTL in single-junction perovskite solar cells (PSCs); however, in tandem solar cells it is rare due to its non-conformality on textured substrates, which impedes efficient hole transport, and the hydrophobicity of its aromatic rings, which impedes wetting of perovskite ink^14,15,16. Thermally evaporated spiro-TTB is another choice, but it prefers aggregating at the ITO-coated pyramid valleys of the silicon bottom cell⁵. Self-assembled monolayers such as [2-(9H-carbazol-9-yl)ethyl]phosphonic acid (2PACz) and [4-(3,6-dimethyl-9H-carbazol-9-yl)butyl]phosphonic acid (Me-4PACz)—usually anchored on ITO—can address several of these shortcomings. However, the issue of de-wetting of the subsequently deposited perovskite precursor inks on these layers¹⁷ jeopardizes device reproducibility and scalability, and this needs to be resolved for scaled fabrication. Strategies such as additive engineering, spray coating or thermal evaporation^18,19,20 for perovskite deposition might work to some degree, yet they unavoidably complicate the fabrication procedure and bring compatibility issues. Alternatively, perovskite deposition directly on transparent conductive substrates has been proven as a reproducible and scalable approach that might even be compatible with rough surfaces^21,22. However, the absence of an HTL causes misaligned energy levels, impeding charge transfer and deteriorating device performance²³. These arguments motivate the search for a simplified device fabrication methodology for textured tandems with efficient hole-collecting contacts that can simultaneously maintain high performance and processing yield.

Here we have systematically screened a few materials that can be dissolved in perovskite ink, and found that copper(I) thiocyanate (CuSCN) (Supplementary Fig. 1)—known to feature well-aligned energy levels, high hole mobility and good stability^24,25—is of particular promise. We co-deposited the perovskite precursor ink with CuSCN directly on the ITO recombination layer of industrial Czochralski (Cz) silicon heterojunction (SHJ) bottom cells and obtained an improved fabrication yield. We compared the radiative recombination losses and charge carrier transport dynamics of co-deposited films with those of pristine perovskite films in contact with different HTLs. Next we verified that CuSCN aggregates at grain boundaries of the buried perovskite surface at a certain depth, forming local hole-collecting contacts for efficient hole transfer and grain boundary passivation (Supplementary Fig. 2). We integrated such CuSCN-embedded PSCs with textured SHJ bottom cells, achieving monolithic tandems with a certified PCE of 31.46% in 1 cm² (steady-state efficiency of 31.02%). Tandems were further scaled up to 4 cm², delivering 28.14% and 25.23% efficiencies through spin- and blade-coating, respectively. Moreover, these tandems exhibited excellent operational and damp-heat stability.

Results

CuSCN-embedded perovskites

We first explored the feasibility of CuSCN-embedded perovskite films deposited on the top ITO of the bifacially textured SHJ bottom cells. The morphology of the silicon surface was evaluated by scanning electron microscopy (SEM) and atomic force microscopy (AFM) (Supplementary Figs. 3 and 4). Various HTLs were employed to compare their efficacy with that of the CuSCN-embedded perovskite films. However, the hydrophobicity of 2PACz and PTAA^19,26 can lead to frequent fractures and notches in the perovskite films due to the diamond-wire-sawing grooves on the silicon surface, abating perovskite coverage (Fig. 1a and Supplementary Fig. 5). This issue is probably related to the fact that de-wetting of the precursor perovskite films on 2PACz or PTAA at the grooves would make them prone to breakage under the processing conditions, including anti-solvent extraction and the spin-coating centrifugal force. By contrast, the CuSCN-embedded perovskites on ITO have satisfactory coverage as the pristine perovskites on sputtered-NiO_x. We counted the macroscopic flaws and achieved visual integrities out of 38 CuSCN-embedded and 41 NiO_x-underlying perovskite films on silicon bottom cells from different batches (Fig. 1b). Noticeably, only 45% (37%) out of the 47 (54) 2PACz (PTAA)-underlying pristine perovskite films were deposited without visible fractures or notches. This suggests a higher device yield for the CuSCN-embedded perovskites, which can be ascribed to better absorber coverage. The SEM cross-sectional micrograph shows the non-uniform thickness of PTAA, which may hinder hole transport⁸ (Supplementary Fig. 6). Meanwhile, the CuSCN-embedded perovskites exhibit grain sizes exceeding 1 μm with good crystallinity (Supplementary Fig. 7).

**Fig. 1: Properties of perovskite films.**

Hereafter we performed steady-state photoluminescence measurements for the perovskite films on silicon bottom cells. The spectral mappings (Fig. 1c) demonstrate stronger luminescence of CuSCN-embedded and 2PACz-underlying perovskite films compared with that of NiO_x- and PTAA-underlying ones, which is indicative of the diminished non-radiative recombination. Correspondingly, the transient photoluminescence (TrPL) results²⁷ in Fig. 1d present a mono-exponential Shockley–Read–Hall-dominated decay at later times²⁸ for the CuSCN-embedded perovskite films, with an emission lifetime of over 2 μs, which is comparable with that of the 2PACz-underlying case; inferior lifetimes are obtained for the NiO_x- and PTAA-underlying films. To elucidate the charge transfer process, we interpreted TrPL decay with differential lifetime τ_TrPL(t) = –{d ln[Φ_TrPL(t)]/d(t)}⁻¹ (Fig. 1e and Supplementary Fig. 8)^27,28. The more rapid rise of τ_TrPL (until ~500 ns) at earlier times indicates a faster hole transfer in the CuSCN-embedded perovskite film.

We performed experimental and theoretical analyses to understand the distribution and role of CuSCN in perovskite films. Perovskite ink with or without CuSCN was deposited on ITO glasses to obtain the peeled-off buried film surfaces (Supplementary Fig. 9). Atomic force microscopy-based infrared spectrocopy (AFM-IR; Supplementary Figs. 10 and 11) and X-ray photoelectron spectroscopy (XPS; Supplementary Fig. 12) results reveal no perovskite remaining on the peeled-off ITO substrates. Scanning elecron microscopy images (Fig. 2a) demonstrate that substantial second phases (marked by pink dashed ovals) are embedded at the perovskite grain boundaries of the buried surface for the co-deposited layer, which is a marked difference from the pristine perovskite film. Infrared signals of the buried surface show a chacterized CuSCN peak²⁹ at ~2,174 cm⁻¹ (Supplementary Fig. 10). The CuSCN phase was found to locally enrich at perovskite grain boundaries without forming a continuous film (Fig. 2b and Supplementary Fig. 13). However, no detectable CuSCN signal is found at the upper surface of the CuSCN-embedded perovskite film (Supplementary Fig. 14). This should be attributed to the preferential crystallization of perovskite in advance of CuSCN at the upper surface of the film and the downward crystallization growth mechanism³⁰ (Supplementary Fig. 15). The X-ray diffraction (XRD), XPS and high-resolution transmission electron microscopy (TEM) results (Supplementary Figs. 16–18) further prove the existence of CuSCN phase. To capture the CuSCN in films, we performed cross-sectional TEM combined with energy-dispersive X-ray spectroscopy (EDS) analysis. Figure 2c clearly shows that the copper-rich phases (green) mainly distribute at the grain boundaries of the buried perovskite surface. Similarly, the SEM–EDS results (Supplementary Fig. 19) also support the CuSCN distribution. We also performed time-of-flight secondary ion mass spectrometry (TOF-SIMS) measurements to probe the longitudinal penetration of CuSCN in the film (Fig. 2d and Supplementary Fig. 20). As the depth profiles depicted, the intensity of SCN⁻ (indicative of CuSCN) declines gradually from the buried surface to the upper surface of the film.

**Fig. 2: Distribution and role of CuSCN in perovskites.**

We then explored the impact of CuSCN addition on perovskite film crystallization. We found that the CuSCN-embedded perovskite films exhibit a larger average grain size than the pristine perovskite films (Supplementary Figs. 7 and 21), which is mainly ascribed to the coordination between SCN⁻ and Pb²⁺ that facilitates the enlargement of the precursor aggregates³¹ (Supplementary Figs. 22 and 23). These aggregates act as the nucleus for promoting the nucleation and growth of perovskites³². Interfacial models of density functional theory (DFT) calculations were built by placing CuSCN on the top of formamidinium iodide (FAI)- and PbI₂-rich FAPbI₃(001) surfaces, respectively (Supplementary Figs. 24 and 25)^33,34,35. From the optimized geometries (Fig. 2e), in the case of the FAI-rich surface, the copper atoms from CuSCN are inclined to bond with surface-dangling iodide, giving a large binding energy of –6.47 eV. On the other hand, for the PbI₂-rich surface, the sulfur atoms from CuSCN would bond with surface-undercoordinated Pb²⁺ with a binding energy of –1.02 eV. The shifts of Pb 4f and I 3d peaks for the buried surface of the CuSCN-embedded perovskite indicate that the uncoordinated I and/or Pb can bond with copper and SCN from CuSCN (Supplementary Fig. 26). We therefore performed photoluminescence measurements and confirmed the stronger and more homogeneous photoluminescence emission for CuSCN-embedded perovskite (Supplementary Fig. 27), implying a declined non-radiative recombination and a local hole-collecting contact structure with good passivation.

Accordingly, a 3D drift-diffusion simulation model was constructed to investigate the hole-collecting probability of the locally embedded CuSCN in single-junction PSCs (Supplementary Note 1). Thanks to a relatively higher contact area ranging from 1.10 to 1.20 (calculated from Fig. 2b), the CuSCN-embedded devices can deliver comparable efficiencies to that of the classic p–i–n PSCs and exhibit improved performance with increased relative contact area (Fig. 2f and Supplementary Fig. 28). We therefore fabricated PSCs featuring the local hole-collecting contact structure on ITO glasses with different CuSCN concentrations in perovskite ink, where the CuSCN contact fractions were estimated (Supplementary Fig. 29). The results show an obviously decreased device efficiency with the reduction of CuSCN concentration (less than 10 mg ml⁻¹; Fig. 2g and Supplementary Table 1). The relatively high CuSCN concentration in perovskite ink different from SAMs co-deposited technique suggests the importance of a large relative contact area in realizing highly efficient CuSCN-embedded devices³⁶. We also explored the potential of CuSCN-embedded PSCs compared with PSCs with layered CuSCN (Supplementary Fig. 30 and Supplementary Table 2) and found that the directly deposited CuSCN layer on the planar ITO causes poor device performance due to the de-wetting of perovskite solution mainly associated with the diethyl sulfide solvent.

To better understand the working mechanism of the CuSCN-embedded PSCs, we performed ultraviolet photo-electron spectroscopy tests to determine the energy diagram (Supplementary Figs. 31 and 32). The results clarify that there is no hole transport barrier at the perovskite/CuSCN interface. A lower average recombination velocity at the ITO/CuSCN-embedded perovskite interface than the ITO/perovskite interface verifies the passivation effect of CuSCN (Supplementary Fig. 33), implying a promising configuration applied to tandem solar cells. This helps to obtain high PCEs of 20.27% (20.01%) with an open circuit voltage (V_OC) of 1.254 V (1.246 V), short circuit current density (J_SC) of 20.62 mA cm⁻² (20.65 mA cm⁻²), fill factor of 78.38% (77.74%) in forward (reverse) voltage scanning direction for the CuSCN-embedded PSCs on the textured silicon substrates (Fig. 2h; see Methods and Supplementary Fig. 34 for device fabrication). The above results display the prominent photoelectric property of the CuSCN-embedded perovskite films and the impressive efficiency in related single-junction PSCs.

Integration into monolithic perovskite/silicon tandems

After confirming the effectivity of the CuSCN-embedded local hole-collecting contact structure in single-junction devices, we integrated it into monolithic tandems (Fig. 3a, right). The CuSCN phase embedded at perovskite grain boundaries simultaneously works as an efficient local hole-collecting contact and an effective defect passivator (Fig. 3a, left). The cross-sectional SEM image and layout of our tandem device are shown in Fig. 3b and Supplementary Fig. 35, respectively. Here, a 15 nm ITO recombination layer is employed for carrier recombination with minimized parasitic absorption⁷. The pseudo J–V curve of one silicon sub-cell was measured by Suns-V_OC (Supplementary Fig. 36a). Meanwhile, a PCE of 24.42% is obtained in a SHJ bottom cell (Supplementary Fig. 36b).

**Fig. 3: Integration and characteristics of CuSCN-embedded tandems.**

As a result, a certified PCE of 31.46% with a steady-state efficiency of 31.02% is achieved for the CuSCN-embedded tandem (Fig. 3c and Supplementary Fig. 37), which, so far, is the highest efficiency for tandems based on inorganic hole transport materials. Typical J–V characteristics also exhibit a laboratory PCE of 31.67% and a negligible hysteresis (Supplementary Fig. 38 and Supplementary Table 3). In contrast, the PCEs of 28.84%, 29.60% and 31.55% are obtained for the NiO_x-, PTAA- and 2PACz-underlying tandems, indicating the effectiveness of the CuSCN-embedded approach. It is worth mentioning that the tandems with stacked NiO_x/2PACz exhibit the best efficiency of 30.38% (Supplementary Fig. 39), which is even lower than the device with only 2PACz. The tandems with co-deposited 2PACz and perovskite³⁶ exhibit a best PCE of only 24.43% (Supplementary Fig. 40), which might be associated with the incomplete coverage of SAMs on textured substrates³⁷. Figure 3d plots the external quantum efficiency (EQE) and corresponding integrated current curves of the CuSCN-embedded tandem. The steady-state tracking PCE of 31.35 ± 0.29% over 15 min is demonstrated for the CuSCN-embedded tandem (Fig. 3e), while unsteady currents are observed in NiO_x- and PTAA-underlying tandems, resulting in the rapidly declined efficiency (Supplementary Fig. 41).

Statistics shown in Fig. 3f compare the efficiency consistencies of the above tandems. The CuSCN-embedded tandems exhibit high reproducibility, with an average PCE of 30.05%. NiO_x-underlying tandems also maintain high reproducibility, but with a much lower average PCE of 26.99% owing to the inferior V_OC and J_SC (Supplementary Fig. 42) that stems from the defective NiO_x/perovskite interfaces¹² and poor conductivity of NiO_x (ref. ³⁸). By contrast, the tandems with stacked NiO_x/2PACz exhibit scattered PCEs (Supplementary Fig. 39b). Similarly, the reproducibility of the 2PACz- and PTAA-underlying tandems is much poorer due to the large leakage currents (Supplementary Fig. 43) resulting from incomplete perovskite absorber layers. These results strongly verify the advantages of tandem solar cells with CuSCN-embedded perovskites.

We fabricated larger sized tandems to explore the scalability of the CuSCN-embedded routine (Fig. 3g and Supplementary Tables 4 and 5). An aperture PCE of 28.14% is achieved with a V_OC of 1.897 V, a fill factor of 77.66% and a J_SC of 19.10 mA cm⁻². The EQE spectra (Fig. 3h) corresponding to the three marked spots in Fig. 3g demonstrate a uniformity of the tandem with a minor integrated-current fluctuation within the perovskite top cell. The large absolute fill factor (1.73%) and J_SC (1.61 mA cm⁻²) deficits in the 4 cm² tandem compared with the 1 cm² tandem might be ascribed to the increased shunt leakage current³⁹ and the shading of silver electrode fingers. We also tried the blade-coating method to fabricate tandem devices based on the co-deposition strategy, resulting in the device with an aperture area of 4 cm² derlivering an efficiency of 25.23% (Fig. 3i). These results illustrate that our co-deposition strategy provides a new route to achieving large-scaled perovskite/silicon tandem solar cells.

Durability was also investigated to further evaluate device performance. We performed time-dependent photoluminescence measurements under a 470 nm laser illumination in N₂ atmosphere at room temperature (Fig. 4a) to compare the luminescence response of perovskite films based on different HTLs. The phase segregation peak at 780 nm can be ascribed to the light-induced I-rich narrow-band-gap perovskite due to ion migration via defects⁴⁰. The time evolution of photoluminescence intensity ratio of 780 nm/740 nm (Supplementary Fig. 44) exhibits a minimum variation of the peak position for the CuSCN-embedded perovskite film, implying the reinforced phase stability. The normalized photoluminescence intensity at 740 nm of all samples (Fig. 4b) exhibits a tendency of increasing in the early times and decreasing at late times, indicative of trap-healing and sample degradation, respectively^41,42.

**Fig. 4: Stability of the tandem devices.**

For operational tandem devices that are affected not only by incident light and radiant heat, but also by Joule heat, electric field, ion migration and other factors, it is important that their long-term stability is investigated^43,44. Specifically, we monitored the operational stability of encapsulated tandem devices under continual xenon-lamp illumination at the maximum power point (MPP). The CuSCN-embedded and 2PACz-underlying tandems maintain eminent durability with 93.8% and 92.0% of their initial efficiencies after ~1,200 h tracking, respectively (Fig. 4c). In contrast, the NiO_x- and PTAA-underlying devices exhibit about 42% and 18% PCE losses after only 350 h and 78 h operation, respectively. Notably, the 2PACz-underlying tandem device benefits from the light-soaking effect⁴⁵ in the first 4 h, with its efficiency increasing from the minimum 96% to nearly 100%. However, even though the performance drops to 98% at the beginning, the CuSCN-embedded tandem device still holds the optimum durability over the whole operational period. For the thermal stability, we tested devices at 85 °C (Fig. 4d). The CuSCN-embedded tandem shows the best stability, retaining 91% of its initial PCE after 400 h, whereas the NiO_x-underlying tandem exhibits the worst stability, with only 70% PCE after 40 h. Meanwhile, the 2PACz- and PTAA-underlying tandems retain 85% and 78% PCEs after 400 h and 100 h, respectively. This result indicates that the tandem thermal durability depends on not only material stability, but also interfacial defect passivation, which interprets the better stability of tandems with CuSCN and 2PACz passivators.

Furthermore, to appraise the resilience of our devices under elevated temperatures and high humidity, we performed a damp-heat test according to International Electrotechnical Commission (IEC) protocol 61215:2021 (85 °C and 85% relative humidity)⁴⁶. Five tandem devices with CuSCN and 2PACz hold, on average, 90.2% and 87.4% of their initial efficiencies, respectively, with almost negligible V_OC and J_SC deterioration for the most stable tandems (Fig. 4e,f and Supplementary Fig. 45). The performance degradation of both tandems is caused by the drop in fill factor, which mainly stems from the shunt resistance decrease for CuSCN-embedded tandem and series resistance increase for 2PACz-underlying tandem^47,48, possibly due to the diffusion of ions and the degradation of contact electrodes⁴⁵. Unfortunately, the NiO_x- and PTAA-underlying tandems show rapidly decreased performance in the first 500 h, with all photovoltaic parameters notably dropping.

Discussion

Overall, the co-deposited CuSCN-embedded perovskite provides a convenient and feasible method towards commercialization of perovskite/silicon tandems requiring high performance, excellent stability, superior reproducibility and scalability with negligible adding cost.

Methods

Materials

Anhydrous dimethylformamide (DMF, anhydrous, 99.8%), anhydrous dimethyl sulfoxide (DMSO, anhydrous, ≥99.9%), anhydrous chlorobenzene (anhydrous, 99.8%), isopropanol (anhydrous, 99.5%) and CuSCN were purchased from Sigma-Aldrich. Indium tin oxide (7–9 Ω sq^–1) patterned glasses, FAI and C₆₀ (~99.9% purity) were purchased from Advanced Election Technology. Poly(triarylamine) (average M_n ≈ 6,000–15,000), caesium iodide (CsI, 99.999%), methylammonium bromide (MABr, 99.5%), lead bromide (PbBr₂, 99.99%) and bathocuproine (BCP) were purchased from Xi’an Polymer Light Technology. Lead(II) iodide (PbI₂, 99.999%) and 2PACz were purchased from Tokyo Chemical Industry. Nickle oxide (NiO_x, 99.9%), ITO (In₂O₃/SnO₂: 90/10 wt%, 99.99%) and indium zinc oxide (IZO, In₂O₃/ZnO; 90/10 wt%, 99.99%) were purchased from ZhongNuo Advanced Material Technology. Magnesium fluoride (MgF_x, 99.99%) and lithium fluoride (LiF, 99.99%) were purchased from Hebei Luohong Technology. Tetrakis(dimethylamino) tin(IV) (TDMASn) (99.9999%) as one precursor of SnO₂ was purchased from Nanjing Ai Mou Yuan Scientific Equipment. Toluene was purchased from Tedia. Ethanol was purchased from Shanghai Macklin Biochemical Technology.

Silicon bottom cell fabrication

N-type Cz silicon wafers with a thickness of ~120 μm were used for fabrication of a SHJ bottom cell. Wet-chemical processes such as saw-damage removal, texturing and RCA (Radio Corporation of America) cleaning were applied to the as-cut wafers. An intrinsic a-Si:H passivation layer (5 nm) was first deposited by plasma-enhanced chemical vapour deposition (PECVD) on both sides, and then phosphorus-doped nanocrystalline silicon oxide (nc-SiO_x:H(n)) layer (20 nm) was deposited as the front emitter, followed by 30 nm boron-doped nanocrystalline silicon (nc-Si:H(p)) deposition at the rear side. All of the a-Si:H(i), nc-SiO_x:H(n) and nc-Si:H(p) layers were deposited using the latest PECVD system produced by Suzhou Maxwell Technologies. After PECVD, ITO were sputtered on the rear (80 nm) and front (15 nm) sides through a shadow mask, followed by silver sputtering at the rear side. An annealing step was performed at 200 °C for 15 min to recover sputtering damage. The wafers were then laser-cut to 2.5 × 2.5 or 4 × 4 cm² for tandem fabrication.

Perovskite top cell fabrication

The silicon wafers were washed by drip-coating ethanol on them at a rotation procedure of 4,000 r.p.m. for 60 s. After that, they were treated with UV–ozone for 15 min and then immediately transferred into a glovebox with N₂ atmosphere or magnetron sputtering equipment for NiO_x sputtering. For CuSCN-embedded perovskite, 1.7 M Cs_0.05FA_0.8MA_0.15PbI_2.25Br_0.75 perovskite precursor solution was prepared by dissolving CsI, MABr, PbBr₂, FAI and PbI₂ into a mixed-solvent system comprising anhydrous DMF/DMSO of 4:1 v/v and then stirring for over 4 h; 12 mg ml⁻¹ CuSCN powder was added into the fully dissolved perovskite precursor solution and then shaken for 10–15 s to get the limpid solution. The CuSCN-embedded perovskite films were deposited by spin-coating precursor solution at a consecutive program of 300 r.p.m., 1,500 r.p.m. and 5,000 r.p.m. for 5 s, 60 s and 13 s, respectively. Chlorobenzene of 300 μl was quickly dropped onto the center of the substrates 15 s before the end of the rotation process. The intermediate phase films were annealed on a hotplate of 80 °C for 15 s and then immediately transferred onto the other one of 100 °C for 15 min to acquire the perovskite films. For other HTLs, 20 nm NiO_x was prepared by radiofrequency sputtering (SCIENS, SI-RF500D) on the 15 nm ITO recombination layer under a 1 × 1 cm² shadow mask. 2PACz (1 mg ml⁻¹ in isopropanol) and PTAA (6 mg ml⁻¹ in toluene) were spin-coated on the ITO recombination layers at 5,000 r.p.m. for 30 s, followed by annealing at 100 °C for 10 min. The deposition and heating of perovskite films on NiO_x, 2PACz and PTAA were the same as that of the CuSCN-embedded perovskite films. MgF_x (1 nm) was thermally evaporated onto perovskite films, which can lead to an energy band bending at the perovskite surface for better electron extraction and further mitigate carrier recombination at the interface⁴⁵. After 10 min of interval in the same chamber, 10 nm C₆₀ was deposited. SnO₂ sputtering buffer layer (12 nm) was then deposited by atomic layer deposition (ALD, KE-MICRO, PE ALD-F50R) with 100 cycles (chamber at 100 °C, TDMASn source at 70 °C with 1.2 s pulse and 5 s purge, H₂O at 20 °C with 1 s pulse and 5 s purge, 90 sccm carrier gas of N₂); 60 nm IZO was sputtered (~2.7 W cm⁻²) on buffer layer under a 1.1 × 1.1 cm² shadow mask. Finally, 400 nm Ag and 60 nm MgF_x were thermally evaporated to accomplish the tandem fabrication.

Large-area perovskite/silicon tandem solar cell fabrication

For silicon bottom cells, a 80 nm ITO back electrode and 15 nm ITO recombination junction were sputtered through a 2 × 2 cm² shadow mask. The silicon bottom cells were then laser-cut into 4 × 4 cm² substrates for tandem fabrication. For perovskite top cells, the anti-solvent chlorobenzene was adjusted to 600 μl for each drop. The IZO electrodes were sputtered with a 2.2 × 2.2 cm² shadow mask. The other layers were fabricated in the same way, but onto 2.5 × 2.5 cm² wafers.

Blade-coated perovskite on textured silicon bottom solar cell

The Cs_0.05FA_0.8MA_0.15PbI_2.25Br_0.75 precursor, containing 5.65 mg ml⁻¹ CuSCN, had a concentration of 0.8 M in DMF, with 25 mol% of DMSO and 5% ethanol. The blade-coating speed was set to 5 mm s⁻¹, the blade-to-substrate gap was 200 μm, and the speed of the N₂ knife was adjusted with a pressure-reducing valve to obtain dry and smooth perovskite precursor film. The coating of the perovskite and drying with N₂ were performed in two steps. The perovskite films were then annealed at 100 °C for 15 min in the ambient with 30–40% relative humidity.

Single-junction perovskite solar cell fabrication on silicon wafers

The bottom electrodes (sputtered ITO under a 1 × 1 cm² shadow mask, ~3 W cm⁻² power density, annealed at 200 °C for 10 min) of single-junction perovskite solar cells on silicon wafers were deliberately thickened to ~150 nm to sufficiently collect charge carriers. The IZO top electrodes were scrapped after being sputtered (Supplementary Fig. 22) to expose the ITO bottom electrodes. The patterned silver electrodes that are slightly different from that of tandems (Supplementary Fig. 22) were thermally evaporated to 400 nm thickness. The fabrications of other layers were the same as in perovskite top cells.

Single-junction CuSCN-embedded perovskite solar cell fabrication on ITO glasses

Indium tin oxide glasses were treated with UV–ozone for 20 min and then immediately transferred into glovebox with N₂ atmosphere. Then, 1.4 M Cs_0.05FA_0.8MA_0.15PbI_2.25Br_0.75 perovskite in DMF/DMSO (4/1 v/v) with or without different contents of CuSCN were spin-coated on ITO substrates at 4,000 r.p.m. for 36 s; 150 μl chlorobenzene was dropped onto the substrates at 30 s of the rotation procedure. After that, the samples were annealed on a hotplate of 80 °C for 15 s and then immediately transferred onto the other one of 100 °C for 15 min to acquire the CuSCN-embedded perovskite films. Finally, 1 nm MgF_x, 15 nm C₆₀, 5 nm BCP and 120 nm silver were thermally evaporated onto the perovskite films to obtain our devices.

Device encapsulation

The front silver grid and back silver electrodes of tandems are dot-coated with silver paste (Dycotec Materials) and lead out with conductive copper tapes. After natural drying for one day, make sure that the silver has good contact with the conductive copper tapes, which are extended out of the sealing part and used as electrodes for stability tests. Take two pieces of 1.1-mm-thick glass and dry them. The tandems are then sandwiched between two pieces of glasses and sealed with epoxy (Ossila, no contact with the devices) at the edge of the covered glasses. The glass seal also requires one day to dry. Note that all of the steps should be performed in ambient nitrogen .

J–V characterizations

The current-voltage (J–V) measurements of the perovskite/silicon tandem solar cells were performed by using a digital source meter (Keithley 2400) and a solar simulator (94022 A, Newport) with standard air mass 1.5 G illumination, which was calibrated by a standard silicon solar cell (PVM937, Newport) with a KG5 filter calibrated by the Newport Corporation TAC-PV Laboratory. The spectral mismatch correction factor is M = 0.994 ± 0.001. The curves were achieved in both the reverse (2.0 V to −0.1 V) and forward (−0.1 V to 2.0 V) voltage scanning modes with 200 data points and a time delay of 0.5 s. The aperture areas of masks are 0.9972 cm² and 4.0026 cm². The J–V curves of single-junction perovskite solar cells were obtained in the reverse (1.3 V to −0.1 V) and forward (−0.1 V to 1.3 V) modes. A light-soaking treatment of approximately 5 to 10 s is required for all of the solar cells with perovskites. For the tandems, all areas except the active area determined by the mask should be shaded during the J–V measurement.

Stability tests

For MPP tracking, the encapsulated devices were tested in air under continual xenon-lamp-based solar simulator (Enli Tech, SS-F5-3A) illumination with a calibrated 100 mW cm⁻² light intensity. The irradiation area was determined by a 1 cm² shadow mask. The operation temperature of the tandems was measured at ~45 °C using an infrared thermometer. For the damp-heat test (85 °C/85% relative humidity), the sealed devices were aged on a hotplate at 85 °C; the ambient humidity was maintained at 85% relaitive humidity, as controlled by an ultrasonic atomizer with adjustable fog output. When testing J–V, the tandems were transferred to another room (<30% relative humidity, room-temperature), and resumed aging after the tests.

External quantum efficiency/steady-state efficiency output

External quantum efficiency measurements were performed using a QE system (QEX10, PV measurement). The chopped monochromatic light beam was fully focused on the active area of the tandems. For perovskite top cell measurement, the silicon bottom cells were saturated by an infrared light-bias LED with 850 nm peak emission. A bias voltage of 0.6 V was used to realize the almost short-circuit conditions. When measuring the silicon bottom cells, a blue light-bias LED with 455 nm peak emission was used to saturated the perovskite subcells and a 1 V bias voltage was applied to maintain the short circuit conditions. For steady-state efficiency tracking, J–V curves in the forward voltage scanning direction were measured with 18 s intervals at 100 mV s⁻¹ from 1.52 V to 1.62 V to find the maximum power output. Between measurements, the devices were held at the voltage of the maximum power point as determined by the most recent J–V test.

XRD/XPS/SEM/TOF-SIMS/AFM-IR

X-ray diffraction patterns were collected by a D8 Discover (Bruker) with a copper Kα X-ray source to characterize the crystal structure and quality of perovskite films. X-ray photoelectron spectroscopy measurements were performed using an X-ray photoelectron spectrometer (AXIS Supra, Kratos) with a monochromatic aluminium Kα X-ray (1486.6 eV) to explore the elements and their valence states in the CuSCN-embedded perovskite films. The top-view and cross-sectional SEM images were collected using a field-emission scanning electron microscope (S-4800, Hitachi). The preparation of the buried surfaces of perovskite films is shown in Supplementary Fig. 9. A thin layer of ultraviolet glue was coated onto the perovskite upper surface and then covered by a clean glass. The sandwiched sample was ultraviolet-soaked for 1–3 min to solidify the glue. The perovskite layer can then be peeled off from the ITO substrate as shown in Supplementary Fig. 9. The buried perovskite surface exposed and the ITO side also was obtained. Time-of-flight secondary ion mass spectrometry analysis was performed using a time-of-flight secondary ion mass spectrometer (Ulvac-Phi, PHI nanoTOF II). Atomic force microscopy-based infrared spectrocopy analysis was performed using an Infrared Scanning Near-field Optical Microscopy (Anasys nanoIR2-s).

Photoluminescence measurements

We performed steady-state photoluminescence and TrPL measurements using a spectrometer (Edinburgh Instruments, FLS920) with a 470 nm diode laser and a long-pass 600 nm filter. The light intensity was controlled with a tunable attenuator and monitored with a power meter. For steady-state photoluminescence, emission spectra with wavelengths between 600 nm and 850 nm were collected. For TrPL, the measurements were implemented using a time-correlated single-photon counting system equipped with a photomultiplier tube detector. The 738 nm light was probed. Photoluminescence mapping measurements were performed using a laser scanning confocal microscope (Olympus, FV1000) with a 470 nm laser⁴⁹.

Transient absorption measurements

For femtosecond transient absorption spectroscopy, the fundamental light beam, with a 1,030 nm output wavelength by an Yb:KGW laser (Pharos, Light Conversion) operating at 100 KHz, was separated into several light beams. The pump beam was generated by a non-collinear optical parametric amplifier. The probe beam was generated by focusing the 1,030 nm beam within a yttrium aluminium garnet crystal. Both pump and probe pulses were focused and spatially overlapped in the sample space, with the temporal delay between them given by a high-resolution delay stage (Newport). The pumping light is 515 nm at the power of 0.4 mW. The area of the laser spot is ~0.4 mm².

Transmission electron microscopy

The cross-sectional sample of the CuSCN-embedded perovskite on the textured silicon bottom cell was prepared using modified focus ions beam technique in a FEI Helios G4 system, including amorphous carbon- and platinum-protecting layers deposition using a gas deposition system and sample thinning. To reduce sample damage by ion beam, low-voltage thinning at 5 kV to slice the sample at first. The voltage was then reduced to 2 kV to make a precision polishing. Note that even though a low-dose cut condition was applied, perovskite surface damage may be inevitable during the focus ions beam process; however, this will not disturb our experimental conclusions. The high-resolution TEM characterization was performed on the aberration correction transmission electron microscopes (FEI Titan G2 60-300) operated at an acceleration voltage of 300 kV. The EDS mapping was captured from probe-side spherical aberration corrected scanning TEM (FEI-Titan ChemiSTEM G2) operated at 200 kV.

Density functional theory calculations

We performed the density functional theory calculations using the projector-augmented wave method as implemented in the Vienna ab initio simulation package code^33,34. The generalized gradient approximation together with Perdew–Burke–Ernzerhof exchange-correlation functional was used. Van der Waals interactions were also included in the calculations using zero-damping DFT-D3 method of Grimme. The uniform grid of 6 × 6 × 2 and 6 × 6 × 6 k-mesh in the Brillouin zone was employed to optimize the crystal structures of trigonal-phase CuSCN and cubic-phase FAPbI₃ bulk, 4 × 4 × 1 k-mesh for CuSCN and FAPbI₃ slabs, and 2 × 2 × 1 k-mesh for CuSCN/FAPbI₃ interfaces. The energy cutoffs of wave functions were set at 500 eV for the bulk and 400 eV for slabs and interfaces. The atomic positions of all structures were fully relaxed until the Hellman–Feynman forces on each atom is less than 0.015 eV Å⁻¹. The binding energy between CuSCN and FAPbI₃ was calculated by \(E_{\rm{binding}} = E_{\rm{CuSCN}/\rm{FAPbI}_3} - E_{\rm{CuSCN}} - E_{\rm{FAPbI}_3}\), where \(E_{\rm{CuSCN}/\rm{FAPbI}_3}\), E_CuSCN and \(E_{\rm{FAPbI}_3}\) are the total energy of CuSCN/FAPbI₃ interface, CuSCN and FAPbI₃, respectively.