a,b, The crystal structure of PbI₂ (a) and SnI₂ (b). c,d, The evolution routes of Pb-based (c) and Sn-based (d) perovskites from the intermediate to perovskite phase. Stages i–iv represent the intermediate phase, metastable phase, transition state and perovskite phase, respectively. e, The desorption and transition barriers from the intermediate to perovskite phase. f, The desorption barrier E₁, transition barrier E₂ and total crystallization barrier E_tot of Sn- and Pb-based perovskites.

Subsequently, the free solvent is removed by solvent extraction methods, and the intermediate phase forms (Fig. 1c–e, stage i). Clearly, both PbI₂ and SnI₂ are strongly coordinated by DMSO molecule at this stage. PbI₂ is coordinated with one DMSO molecule, whereas we found that SnI₂ coordinates with three DMSO molecules (Supplementary Figs. 4 and 5)^35,39. The Sn–O bond length (2.18–2.32 Å; Supplementary Fig. 4) in the Sn-based intermediate phase is shorter than that of Pb–O in the Pb-based intermediate phase (2.49 Å)³⁵. The higher coordination number and shorter coordination bond suggest that it is more difficult for DMSO to desorb from the Sn-based intermediate phase, which can also be further proved by the higher desorption temperature of DMSO in the Sn-based intermediate phase (154 °C versus 95 °C; Supplementary Fig. 5) and the downshift of the S=O stretching vibrational mode (Supplementary Fig. 6) compared with the Pb-based intermediate phase. Understandably, the generation rate from the intermediate phase should be strongly influenced by the desorption rate of DMSO, which can be described by the desorption barrier (E₁; Fig. 1e) and the Arrhenius equation. We obtained the desorption rate at different temperatures on the basis of thermogravimetric analysis (TGA), as shown in Supplementary Text 1, fitted the desorption rate using the Arrhenius equation and obtained an E₁ value of 2.12 eV and 0.99 eV for the Sn-and Pb-based intermediate phase (Supplementary Fig. 7), respectively. The E₁ value of the Sn-based intermediate phase is 1.13 eV higher, indicating that DMSO desorption is more difficult, which echoes our previous analysis.

After DMSO desorption, the Pb(Sn)–I framework reacts with the organic halide to form perovskite, and we call the unreacted state the metastable phase (Fig. 1c–e, stage ii). This phase must undergo structural reorganization to establish a corner-sharing octahedral framework. For Pb-based perovskite, the metastable phase (Fig. 1c, stage ii) would inherit the edge-sharing feature of [PbI₆]⁴⁻ in the intermediated phase after DMSO desorption. Theoretically, the transition from edge-sharing units to a corner-sharing framework should include a Pb–I bond breakage process and a formation process for an isolated trigonal bipyramid (denoted as the transition state; Fig. 1c, stage iii)⁴⁰. We define this transition barrier as E₂. Subsequently, the free I⁻ ions from the organic halide insert into the isolated [PbI₅]³⁻ to form a corner-sharing [PbI₆]⁴⁻ framework. Finally, the organic cations occupy the space enclosed by the [PbI₆]⁴⁻ framework to form the perovskite phase (Fig. 1c, stage iv). For Sn-based perovskites, the metastable phase (Fig. 1d, stage ii) inherits the no-sharing feature of [SnI₅]³⁻ in the intermediated phase. The no-sharing [SnI₅]³⁻ framework could spontaneously transfer into a corner-sharing [SnI₆]⁴⁻ framework through electrostatic attraction, and we believe that the barrier for this transition is negligible. Finally, the organic cations enter the corner-sharing [SnI₆]⁴⁻ framework to complete the perovskite phase (Fig. 1d, stage iv). Solid-state reactions reveal that SnI₂·3DMSO and Cs_0.2FA_0.8I react immediately to form the perovskite phase once they come into contact at room temperature (Supplementary Fig. 8), confirming that the barrier for the transition from the no-sharing [SnI₅]³⁻ framework to the corner-sharing [SnI₆]⁴⁻ framework is negligible. Conclusively, the total crystallization barrier (E_tot) for Sn-based perovskite is determined only by E₁, while the E_tot of Pb-based perovskite is determined by both E₁ and E₂, which accounts for the different crystallization rate in mixed Sn–Pb perovskite. Since the metastable phase is very difficult to harvest, we failed to measure E₂ experimentally. We thus roughly evaluated E₂ as the bonding energy of Pb–I (1.86 eV)⁴¹ because the formation of the transition state needs to break one Pb–I bond.

Clearly, the formation rates of Sn- and Pb-based perovskites depend on the collective rates of both DMSO desorption and phase transition from the metastable to perovskite phase. The total formation rate R_tot can be expressed by the Arrhenius equation as⁴²

where A_tot is the pre-exponential factor, k_B is the Boltzmann constant and T is the absolute temperature. The E_tot values for Sn- and Pb-based perovskites are E₁ (2.12 eV) and E₁ + E₂ (2.85 eV; Fig. 1f), respectively. Therefore, for Pb-based perovskite, significantly more energy is required to overcome E_tot, resulting in a much slower crystallization rate compared with Sn-based perovskites. Conclusively, maintaining nearly identical E_tot values for the Sn- and Pb-based perovskites is the key to achieving synchronous crystallization.

Crystallization barrier control for synchronized growth

The vacuum-assisted method, one of the most promising techniques for future mass production of perovskite photovoltaics, was applied for the deposition of Sn–Pb mixed NBG PSCs. To achieve fast nucleation and a compact film, the vacuum should be as low as possible, and a threshold pumping time of 20 s is required to fully dry our film under a vacuum degree of ~3 Pa (Supplementary Figs. 9 and 10). The Sn–Pb mixed perovskite composition in this study is Cs_0.1FA_0.6MA_0.3Pb_0.5Sn_0.5I₃, which has an optimal bandgap of 1.22 eV for the bottom sub-cell¹⁵.

To obtain a uniform Sn–Pb mixed perovskite, we managed to fine-tune the E_tot value to guarantee a balanced crystallization rate. Since the E₂ value of the Pb-based perovskite relies on intrinsic atomic rearrangement and would stay relatively constant, we sought to tune the E₁ value to balance E_tot. Here, instead of additives, we simply set a series of ratios of DMSO/[Pb + Sn] to tune the coordination of Sn²⁺ and Pb²⁺ with DMSO and target synchronous crystallization (Supplementary Figs. 11 and 12). According to the above-mentioned method and Supplementary Text 1, we obtained the E₁ value of the Sn- and Pb-based perovskites as a function of DMSO/[Pb + Sn] with a fixed Sn:Pb ratio of 1:1 (Fig. 2a). Both of them exhibited the nature of a smoothed step function. In mixed Sn–Pb perovskite precursor, because the coordination ability of Sn²⁺ is much larger than that of Pb²⁺, when the DMSO/[Pb + Sn] ratio is 1.5:1, all the Sn²⁺ should be preferentially coordinated with DMSO while the Pb²⁺ is at most partially coordinated with DMSO; when this ratio is 2:1, both Sn²⁺ and Pb²⁺ are prone to be exactly coordinated by DMSO (simplified as SnI₂·3DMSO and PbI₂·1DMSO). When the ratio is increased further to beyond 2:1, both Sn²⁺ and Pb²⁺ are in principle fully coordinated and free DMSO exists in solution. On the basis of the morphology, crystallinity and photoluminescence characterization, it is observed that the film quality initially improves but then deteriorates with increasing DMSO content (Supplementary Figs. 13 and 14). The film quality is optimal when Sn²⁺ is fully coordinated and Pb²⁺ is partially coordinated.

Fig. 2: Balancing the crystallization barrier of Sn- and Pb-based perovskites via precisely tuning the DMSO/[Pb + Sn] ratio.

a, The desorption barrier E₁ of Sn- and Pb-based perovskites as a function of DMSO/[Pb + Sn]. In the grey region, ΔE₁ is close to E₂, nearly balancing the crystallization rate for Sn- and Pb-based perovskites. b, The Sn/Pb ratio of the control and optimized perovskite films measured by energy-dispersive spectroscopy. c, Energy-level schemes of control and optimized perovskite films based on the ultraviolet photoelectron spectroscopy (UPS) spectra. d, ToF-SIMS of control and optimized perovskite films. The grey region denotes the upper surface of the perovskite film. e, Steady-state PL spectra of the control and optimized films excited from the top and bottom surfaces. f, Time-resolved photoluminescence spectra of control and optimized Sn–Pb perovskite films deposited on bare glass. g, The Urbach energy E_u of Sn–Pb NBG perovskite films extracted from absorbance spectra.

We then sought to identify the optimal crystallization window. The dashed red line denotes the reference line of E₂, and the optimal processing window lies where E_1,Sn − E_1,Pb (denoted as ΔE₁; Fig. 2a, solid grey line) is close to E₂, thus enabling a nearly identical E_tot. The grey region meets our requirements, where the DMSO/[Pb + Sn] ratio ranges from approximately 1.76 to 2.15. In this region, ΔE₁ can reach 1.65 eV, which is very close to E₂ (~1.86 eV). Thus, a synchronous crystallization rate is expected. We fabricated a series of solar cells with DMSO/[Pb + Sn] ratio ranging from 1.56:1 to 2.35:1 with ethylenediammonium diiodide (EDAI₂) treatment. The difference in film thickness was relatively small, with a maximum variation of 80 nm (Supplementary Fig. 15), which is unlikely to affect device performance significantly^15,43. According to the PCEs of these solar cells, we found that the optimal ratio window ranges from 1.85:1 to 2.05:1 (Supplementary Fig. 16). We also fabricated devices without EDAI₂ and found that the optimal DMSO/[Pb + Sn] ratio remains the same (Supplementary Fig. 17), while EDAI₂ treatment primarily enhances the device fill factor (FF)²⁰ (Supplementary Fig. 18). Furthermore, the device PCEs with a DMSO/[Pb + Sn] ratio of 1.95:1 exhibited the highest efficiency with the smallest standard deviation. We thus chose this optimal device to compare with a control device fabricated with a ratio of 0.8:1 (a common ratio used in literature for antisolvent-free methods)^4,25,44.

Characterization of mixed Sn–Pb perovskite films

We expect our optimized film to enjoy better composition uniformity and higher film quality thanks to the synchronous formation of Sn- and Pb-based perovskites. The horizontal Sn/Pb ratio revealed significant variation in the control film, while the optimized film exhibited a more uniform lateral Sn/Pb ratio, approximately equal to 1 (Fig. 2b). This indicates a more uniform elemental distribution in the optimized film, in accordance with surface potential results (Supplementary Fig. 19). We further observed an upwards shift in the Fermi level in the optimized film (Fig. 2c and Supplementary Fig. 20), indicating a more intrinsic film. We further evaluated the longitudinal homogeneity of Sn and Pb elements in mixed Sn–Pb perovskite films by using time-of-flight secondary-ion mass spectrometry (ToF-SIMS). For the control film, the Sn element was enriched on the surface while the Pb element gradually increased from the surface towards the bulk (Fig. 2d and Supplementary Fig. 21). This originates from the faster formation rate of Sn-based perovskite because, in our vacuum-assisted film fabrication, perovskite crystallization starts from the surface. For the optimized film, both Sn and Pb elements exhibit uniform distribution without much variation. The improved longitudinal homogeneity was also well confirmed by the identical photoluminescence emission peaks of perovskite film when excited from the top and bottom surface (Fig. 2e). As a result, the optimized film exhibited better uniformity thanks to the synchronous crystallization.

We further performed time-resolved photoluminescence to investigate the carrier dynamics in the films (Fig. 2f). The optimized perovskite film showed an average carrier lifetime that was twice as long (651.2 ns) as that of the control film (339.3 ns), indicating a reduced defect density. The Urbach energy (E_u) of the optimized films decreases from 43.4 ± 0.3 to 32.5 ± 0.26 meV (Fig. 2g), implying reduced structural disorder or secondary phases in the optimized film. We also found that the oxidation of Sn²⁺ was partly reduced in the optimized films owing to the reduced Sn element at the surface (Supplementary Fig. 22), which is also in line with the higher average surface potential and lower defect density near the surface for the optimized film compared with the control film (Fig. 2c and Supplementary Fig. 23).

Photovoltaic characteristics of Sn–Pb NBG PSCs

We then fabricated Sn–Pb NBG PSCs using the control and optimized perovskite films. The device structure was indium tin oxide (ITO)/poly(3,4-ethylenedioxythiophene):poly(styrene sulfonate) (PEDOT:PSS)/perovskite/C₆₀/bathocuproine (BCP)/Ag, where the thickness of the perovskite layer is ~1,000 nm (Supplementary Fig. 24). Compared with control devices, the optimized devices showed improved performance across all photovoltaic parameters (Fig. 3a). The best-performing device has an open-circuit voltage (V_OC) of 0.885 (0.871) V, a short-circuit current density (J_SC) of 32.4 (32.5) mA cm⁻² and a FF of 79.8% (78.2%), yielding a PCE of 22.88% (22.14%) under reverse (forward) voltage scanning (Fig. 3b), which is much better than the control devices (Supplementary Fig. 25 and Supplementary Table 1). The device PCEs with a narrow distribution are shown in Supplementary Fig. 26. The steady-state power output (SPO) efficiency of the device at the maximum power point is 22.53% (Supplementary Fig. 27). The integrated J_SC values derived from the external quantum efficiency (EQE) curves of the control and optimized cells were 28.66 mA cm⁻² and 31.98 mA cm⁻² (Fig. 3c), respectively. The main EQE improvement stems from photons beyond 700 nm, which have a deeper penetration depth and are absorbed by the whole bulk film down to the perovskite–C₆₀ interface. Thanks to the more homogeneous composition and reduced Sn⁴⁺-associated defects, these photogenerated carriers enjoy better collection in the optimized device, leading to the better EQE. Transient photovoltage characterization reveals that the decay lifetime of the optimized device (283 μs) is longer than that of the control device (44 μs; Fig. 3d), indicating that the carrier trapping and non-radiative recombination was substantially suppressed, which can also be further confirmed by the light intensity-dependent photovoltage (Supplementary Fig. 28). In summary, we demonstrate a PCE that compares favourably with reported antisolvent-free NBG PSCs (Fig. 3e and Supplementary Table 2). We further prepared Sn–Pb mixed perovskite films and devices using both gas-assisted and antisolvent methods. At a DMSO/[Pb + Sn] ratio of about 1.95:1, these films showed the highest crystallinity, and the corresponding devices show the best performance (Supplementary Figs. 29 and 30), demonstrating that this optimal DMSO range is effective for both antisolvent-free and antisolvent methods. In addition, this finding is also applicable to Sn–Pb mixed PSCs fabricated by both spin-coating and blade-coating methods (Supplementary Figs. 31 and 32).

Fig. 3: Device performance of optimized Sn–Pb NBG PSCs.

a, A comparison of the photovoltaic performance between control and optimized devices (16 individual devices of each type). The box represents the interquartile range (25–75th percentile) of the dataset, with the central line indicating the median value. Whiskers extend to 1.5 times the interquartile range. b, Current density–voltage (J–V) curves of the best-performing Sn–Pb NBG PSCs. c, EQE spectra of control and optimized Sn–Pb NBG PSCs. d, Transient photovoltage decay curves of control and optimized Sn–Pb NBG PSCs. e, A summary of reported PCEs of Sn–Pb NBG PSCs fabricated by antisolvent and antisolvent-free methods.

Vacuum-assisted all-perovskite TSCs

We then fabricated two-terminal all-perovskite TSCs. The top cell has an absorber composition of DMA_0.1Cs_0.4FA_0.5Pb(I_0.75Br_0.25)₃ (DMA = dimethylammonium; FA = formamidinum) with 5 mol% methylammonium lead chloride (MAPbCl₃) additive. Detailed information about the wide-bandgap (WBG) top cells can be found in the Supplementary Information (Supplementary Fig. 33 and Supplementary Table 3). The tandem devices have a structure of glass/ITO/NiO/4PADCB/DMA_0.1Cs_0.4FA_0.5Pb(I_0.75Br_0.25)₃/C₆₀/SnO₂/Au/PEDOT:PSS/Cs_0.1FA_0.6MA_0.3Pb_0.5Sn_0.5I₃/C₆₀/BCP/Ag, where 4PADCB is a self-assembled monolayer (4-(7H-dibenzo[c,g]carbazol-7-yl)butyl)phosphonic acid (Fig. 4a,b). Ultimately, our best-performing device showed a PCE of 28.95% (28.47%) under reverse (forward) voltage scanning, with a V_OC of 2.134 (2.125) V, a J_SC of 16.05 (16.03) mA cm⁻² and a FF of 84.52% (83.58%). An independent laboratory also certified comparable efficiencies of 28.87% and 28.18% with reverse and forward voltage scanning, respectively (Fig. 4b, Supplementary Fig. 34 and Supplementary Table 4). The integrated J_SC values of the WBG and NBG sub-cells obtained from the EQE spectra (Fig. 4c) are 16.01 and 15.96 mA cm⁻², respectively, in good agreement with the J_SC values determined from current density–voltage (J–V) measurements. We report a high PCE for all-perovskite TSCs fabricated by antisolvent-free techniques, including vacuum-, gas- and radiation-assisted methods and two-step methods (Fig. 4d and Supplementary Table 4). The device demonstrated an SPO efficiency of 28.53% at the maximum power point (Fig. 4e). We further fabricated 1.02 cm² monolithic all-perovskite TSCs, which yielded a V_OC of 2.09 (2.08) V, a J_SC of 15.90 (15.83) mA cm⁻² and an FF of 81.29% (80.84%), corresponding to a PCE of 27.01% (26.62%) under reverse (forward) scanning (Supplementary Fig. 35). The unencapsulated TSC retained 87% of its original PCE after about 450 h of MPP tracking (Fig. 4f), which is comparable to previous reports^4,8,9. When operated at 65 °C, the tandem device maintained 60% of its initial PCE after 100 h of illumination (Supplementary Fig. 36), which is inferior to previous reports^13,45. We attribute the low operational stability to Ag diffusion from electrode to perovskite, the volatile MA composition in NBG perovskites and the unstable PEDOT:PSS hole transport layer. Thus, we suggest several potential perspectives to improve the device stability, including replacement of the BCP with a compact SnO₂ fabricated by atomic layer deposition (ALD), replacement of the PEDOT:PSS with an inorganic hole transporting layer such as NiO and the development of an MA-free NBG perovskite system.

Fig. 4: Device performance of all-perovskite TSCs.

a, A cross-sectional scanning electron microscopy image of a representative tandem device. b, J–V curves of the tandem TSC certified by Shanghai Institute of Microsystem and Information Technology. c, The EQE spectra of the bottom and top sub-cells. d, A summary of typical PCEs of all-perovskite TSCs fabricated by antisolvent and antisolvent-free methods. e, The SPO of a high-efficiency tandem TSC. f, Continuous maximum power point (MPP) tracking of an unencapsulated tandem device.