Crystal-facet-directed all-vacuum-deposited perovskite solar cells

MainHybrid metal halide perovskites have redefined the limits of low-temperature-processed photovoltaics (PVs), achieving certified power conversion efficiencies of over 27% in single-junction cells and close to 35% in perovskite-on-silicon tandem cells1,2. Their low-temperature processability3 and bandgap (Eg) tunability4 position them as ideal candidates for next-generation applications, from lightweight flexible modules5,6 to ultrahigh-efficiency multijunction architectures7,8,9,10. However, the dominant solution-based fabrication methods, which are responsible for most record-breaking devices, face scalability and manufacturing challenges, including the use of (toxic) solvents11, batch-to-batch variability12, spatial inhomogeneity13 and incompatibility with highly textured substrates—for instance, industry-standard pyramidal silicon14.Vacuum thermal evaporation (VTE), a solvent-free, industry-proven deposition technique widely used in organic light-emitting diode (OLED) displays15, functional glass coatings16 and inorganic thin-film PVs17, offers a compelling pathway to overcome these limitations. Its inherent ability to conformally coat textured substrates makes it particularly suitable for perovskite-on-silicon tandems18. Although single-component layers are no doubt simpler, high-volume manufacturing exists for co-evaporation of multiple components—for instance, with copper–indium–gallium–selenide PVs19 and doped organic semiconductors in OLEDs20. Unlike the solution-processed counterparts that are typically formed from a mixture of the required precursors21,22,23, the VTE co-deposition of perovskite layers heavily relies on the fine control of each evaporation source24. Two hurdles that pose a critical challenge to forming homogeneous perovskite layers via VTE are the deposition of low-sublimation-temperature materials, such as methylammonium or formamidinium iodide (MAI or FAI)25, and the control of stoichiometry for target bandgaps26. Therefore, translating this potential into high-performance evaporated ‘baseline’ perovskite solar cells (PSCs) will be crucial for establishing industrially compatible VTE manufacturing paradigms and requires tremendous efforts on crystallization control27, halide segregation mitigation28 and grain orientation engineering to render high operational stability for perovskite absorbers10,21.Here we demonstrate a multisource co-evaporation strategy that decouples stoichiometric precision from crystallization kinetics, enabling phase-stable 1.67-eV wide-bandgap (WBG) perovskites with dominant (100) ‘face-up’ orientation. By independently tuning six thermal sources, we achieve a maximum power point tracking (MPPT) efficiency (ηMPPT) of 19.3% (18.35% certified, 0.25 cm2) and 18.5% (1 cm2) for ‘all-vacuum-deposited’ WBG cells. These cells retain 100% peak efficiency after 20,000 h of dark storage in nitrogen. Under the stringent ISOS-L-2 stressing protocols (ISOS, International Summit on Organic Photovoltaic Stability; full-spectrum simulated sunlight, open circuit (OC)) at 75 ± 5 °C and 65 ± 5 °C, our devices maintain 80% of peak ηMPPT after 1,080 h and 1,250 h, respectively, rivalling state-of-the-art solution-processed counterparts29,30,31,32,33,34. Operando hyperspectral imaging reveals suppressed compositional heterogeneity and non-radiative losses, directly linking facet-engineered microstructure to operational resilience. Critically, we integrate these advances into perovskite-on-silicon tandems that achieve substantial efficiency (27.2%, 1 cm2 with a solution-processed self-assembled monolayer (SAM) hole transport layer (HTL), and 24.3% with an all-vacuum-deposited device stack) and field-lifetime stability for the all-vacuum-deposited tandem, retaining ~80% of initial performance after eight months of continuous operation in Italy.Crystal-facet-directed co-evaporated WBG perovskitesCompared with dual-source-deposited MAPbI3 (Eg = 1.55 eV; ref. 35) or FAPbI3 (Eg = 1.48 eV; ref. 36), there are only a handful of reports on WBG perovskites fabricated via VTE37,38,39,40,41. One effective approach is to employ ‘seed’ layers to guide the crystal growth and improve perovskite film quality during VTE37,42. We screened seed-layer-assisted37 and seed-free co-evaporation routes43 to form ~1.67-eV perovskites suitable for perovskite-on-silicon tandems (Fig. 1a); the highest phase purity and performance were obtained for seed-free co-evaporation with a small PbCl2 co-source (5 mol%) (Supplementary Note 1 and Supplementary Figs. 1 and 2). We conduct X-ray diffraction (XRD) measurements on these evaporated perovskite films (Fig. 1b), in which we observe a preferential orientation along the (100) planes (2θ = 14.2–14.4°). Upon introducing Cl− from either seed layers or a co-evaporated source, we detect a subtle shift of the (100) peak to higher reflection angles, implying Cl incorporation44. Furthermore, we identify PbI2 phase, which is photo-unstable and detrimental to device stability45,46, in the reference perovskite (that is, seed free with no PbCl2) and all the perovskites deposited on seed layers, but not in the PbCl2-added, seed-free co-evaporated perovskites (Fig. 1b). Moreover, the latter shows highly oriented, uniaxial domains along the (100) planes with negligible (110) peak (Fig. 1c). The full-width at half-maximum (FWHM) of the (100) peak is narrowed twofold compared with the films prepared using other protocols. To link the crystallographic properties and device performance, we fabricate all-vacuum-deposited PSCs with the architecture shown in Fig. 1d. While other devices suffer from substantial hysteresis and s-kinked behaviour, our PbCl2-added, seed-free co-evaporated cells deliver hysteresis-free current density–voltage (J–V) curves with the highest efficiency (Fig. 1e). Importantly, our seed-free co-evaporation enables straightforward ‘in situ’ bandgap tuning (~1.65–1.72 eV) without seed-layer-induced stoichiometric perturbations (Supplementary Fig. 3 and Supplementary Note 2)37.Fig. 1: Crystal-facet-directed co-evaporated WBG perovskites.a, Schematic of a co-evaporation system with multiple VTE sources available for separate precursors. QCM, quartz-crystal microbalance. b, XRD patterns of 500-nm co-evaporated perovskite films of Cs0.17FA0.83Pb(I0.80Br0.20)3 (denoted as ‘Ref.’), with 10-nm seed layers of CsCl, PbCl2 and CsPbCl3 separately, to compare with a seed-free co-evaporated perovskite of Cs0.17FA0.83Pb(I0.75Br0.20Cl0.05)3 (denoted as ‘Seed free with PbCl2’). Certain diffraction angles are enlarged to show the PbI2 and (110) phases. c, FWHM and (110)/(100) peak ratios of the corresponding 500-nm co-evaporated perovskite films shown in b. d, Schematic of device architecture for all-vacuum-deposited WBG PSCs. The absorber layers are the corresponding 500-nm co-evaporated perovskites shown in b. The seed layers are only involved in the device stack if specified. Spiro-TTB, 2,2′,7,7′-tetra(N,N-di-p-tolyl)amino-9,9-spirobifluorene. e J–V characteristics of all-vacuum-deposited WBG PSCs using the device architecture shown in d. The absorber layers are the corresponding 500-nm co-evaporated perovskite films shown in b. f, XRD patterns of 500-nm seed-free, co-evaporated Cs0.17FA0.83Pb(I0.75Br0.20Cl0.05)3 perovskite films that were annealed at 135 °C in air (40–50% RH) for 0–30 min. g, Photos and top-view SEM images of the corresponding films shown in f, taken immediately after XRD measurements (~1.5 h of exposure to ambient conditions). h, GIWAXS pattern for the in situ annealing of an as-deposited co-evaporated FA0.83Cs0.17Pb(I0.75Br0.20Cl0.05)3 perovskite material at 135 °C in air (40–50% RH in the laboratory), which resembles the device fabrication conditions. i, Ex situ 2D GIWAXS intensity mapping from an incident angle of 1.0° for a post-annealed (for 30 min) co-evaporated Cs0.17FA0.83Pb(I0.75Br0.20Cl0.05)3 perovskite film of 500 nm. j, A schematic illustration of the face-up and random orientations of (100) crystal facets with respect to the ITO substrate.Full size imageWe selected our ‘seed-layer-free’ process for further optimization and investigation. Hereafter, unless we specify otherwise, the targeted perovskite composition is FA0.83Cs0.17Pb(I0.75Br0.20Cl0.05)3. Taking advantage of humid air44,47,48, we adopt an ambient annealing procedure (135 °C in air, 40–50% relative humidity (RH)) and examine different post-annealing durations (0–30 min) for the perovskite films. The pre-annealed perovskite presents a (100)-dominant orientation (Fig. 1f), with no change in the orientation preference upon thermal treatment, as indicated by the relative intensities of (100) and (200) peaks, while the counts gradually increase with annealing, suggesting increased overall crystallinity. Compared with the unannealed case, 30-min annealing results in the (100) intensity increasing ~30-fold and the FWHM reducing by a factor of ~2, giving an overall integrated peak area increase of ~12.5-fold. For annealing times shorter than 10 min, we observe pinhole formation and decolouration of the perovskite films following ~1.5 h of ambient exposure, indicative of material degradation (Fig. 1g and Supplementary Fig. 4). The films with elongated annealing (that is, 20 and 30 min) show no decolouration.To track crystallographic evolution during annealing, we perform in situ grazing-incidence wide-angle X-ray scattering (GIWAXS) measurements on as-deposited co-evaporated perovskites during slow heating to 135 °C in air (40–50% RH) (Fig. 1h). As the temperature T approaches 100 °C, scattering halos emerge, and upon reaching 135 °C the (100) peak intensity increases approximately fourfold. The dominant scattering in the Qz direction (Qxy = 0) is consistent with the final perovskites adopting a ‘face-up’ orientation, where the (100) planes lie along the out-of-plane direction, whose intensity is ~3.6 times that of the in-plane direction (Fig. 1i,j and Supplementary Figs. 5c and 6). These results confirm that extended annealing above the crystallization temperature is essential for obtaining crystallographically desirable, ambient-stable perovskite films49.Relative to solution-processed analogues, our co-evaporated films show higher crystallinity and larger, flatter grains in scanning electron microscopy (SEM) images, spanning the entire perovskite film thickness (Supplementary Figs. 5–8 and Supplementary Note 3). Thus, our modified co-evaporation recipe, which differs from convention only in composition and trace PbCl2 addition, successfully overcomes the historic challenge that co-evaporated perovskites exhibit tiny, misorientated polycrystalline grains39.Ageing of perovskite filmsGiven their improved crystallographic and morphological properties21, we investigate the stability of our evaporated perovskite films and devices. We age ‘bare’ perovskite films coated on indium tin oxide (ITO) substrates, with a thin poly(methyl methacrylate) coverage as rudimentary encapsulation, under the ISOS-L-2 procedure (75 ± 5 °C, 0.76-sun, in ambient (50–60% RH in the laboratory))50. Optical microscopy (Fig. 2a) reveals intense ‘surface wrinkling’ in fresh solution-processed films, a feature linked to local halide heterogeneity that accelerates light-induced degradation51. Such morphology is absent from evaporated films. Upon ageing, solution-processed films degrade rapidly, forming pinholes within 48 h and discolouring by 96 h, while evaporated films show no visible degradation until 196 h. XRD (Fig. 2b and Supplementary Fig. 9) confirms faster degradation of the cubic perovskite phase in solution-processed films. A PbI2 peak appears immediately in solution-processed films but only after 144 h for evaporated films. Conversely, aged evaporated perovskites exhibit minor additional peaks (2θ = 14.8° and 29.6°), probably corresponding to Br-rich perovskite phases from light- and heat-induced halide segregation52, which are absent from solution-processed perovskites, suggesting distinctive material degradation pathways. UV–Vis (Fig. 2c) shows a rapid absorbance drop at the band edge in solution-processed films within 48–96 h, signalling near-total loss of the photoactive phase, contrasted by a slow decline in evaporated films. These results collectively demonstrate the superior stability of evaporated perovskites, which probably stems from eliminating the use of highly coordinating solvents53,54 and improving the facet orientation and crystallinity (Supplementary Note 4 and Supplementary Figs. 10 and 11)55,56.Fig. 2: Perovskite film ageing characterization.a, Optical microscopic images of unencapsulated evaporated and solution-processed perovskite films using an absorber composition of FA0.83Cs0.17Pb(I0.75Br0.20Cl0.05)3, aged under the ISOS-L-2 protocol (full-spectrum simulated sunlight, 75 ± 5 °C, ambient air (50–60% RH in the laboratory)). A thin layer of poly(methyl methacrylate) was deposited on the films to protect them from dust and scratches. Photographs shown here were taken with an optical microscope with a ×20 magnification rate. Scale bar: 0.1 mm. b,c, Evolution of the XRD patterns (b) and the absorbance spectra (c) of the corresponding perovskite films shown in a. Inset: photographs of the perovskite films taken during ageing. The XRD patterns (b) in the logarithmic scale are shown in Supplementary Fig. 9 to reveal any impurity phases formed during ageing.Full size imagePerformance and operational stability of co-evaporated PSCsBefore we move on to assess the long-term stability in complete devices, we fabricate all-vacuum-deposited cells (Fig. 1d) and assess their performance. We also fabricate solution-processed PSCs of the same targeted composition on a [4-(3,6-dimethyl-9H-carbazol-9-yl)butyl]phosphonic acid (Me-4PACz) HTL with the same stacks, to compare the device performance and stability (see Supplementary Fig. 12 for the reason for the choice of HTL).Our 0.25-cm2 evaporated cells with a SnO2 ‘buffer layer’ exhibit open-circuit voltage (VOC) up to 1.20 V (Fig. 3a,b and Supplementary Table 1), and excellent ηMPPT of 19.3% (Supplementary Fig. 13), certified as 18.35% (Supplementary Figs. 14 and 15). Our 1-cm2 evaporated cells (Supplementary Figs. 16 and 17 and Supplementary Table 1) demonstrate a minimum performance deficit (ηMPPT = 18.5%) compared with small-area cells, indicating high film uniformity, promising for large-area deposition. These efficiencies are among the highest reported for all-vacuum-deposited WBG PSCs (Supplementary Fig. 18), and are certified. A detailed VOC loss analysis is given in Supplementary Note 5 and Supplementary Fig. 19. The external quantum efficiency (EQE)-derived PV bandgaps (EgPV) for the evaporated and solution-processed PSCs are 1.67 and 1.685 eV, respectively (Supplementary Figs. 20 and 21). Several unencapsulated evaporated cells demonstrate long ‘shelf lifetimes’, maintaining their peak performance (ηMPPT \(\approx\) 18%) after ~20,000 h of storage under N2 (>2 yr) (Supplementary Fig. 22).Fig. 3: PSC characterization.a, J–V characteristics of our representative all-vacuum-deposited and solution-processed PSCs with an aperture size of 0.25 cm2 using an absorber composition of FA0.83Cs0.17Pb(I0.75Br0.20Cl0.05)3, with either atomic layer deposition (ALD) SnO2 or evaporated BCP buffer layers. R, reverse scan. F, forward scan. Anti-reflective foils were applied on the glass side of the cells to minimize the optical loss. Detailed PV metrics can be found in Supplementary Table 1. b, The ηMPPT of our representative evaporated or solution-processed 0.25-cm2 PSCs using FA0.83Cs0.17Pb(I0.75Br0.20Cl0.05)3. c, Evolution of the ηMPPT of encapsulated all-vacuum-deposited and solution-processed PSCs under the ISOS-L-2 protocol in ambient air (50–60% RH in the laboratory). The error bands represent the s.d. of independent cells (n = 5, 6, 8 and 8 for evaporation SnO2, evaporation BCP, solution SnO2 and solution BCP, respectively), and the centres represent the median values. d, Stability of individual champion cells from c. Dashed lines: time taken to drop to 80% of peak ηMPPT (T80, Champ). e, Literature stability data under different ageing protocols for passivation-free WBG PSCs with ideal bandgaps (1.65–1.72 eV) for perovskite-on-silicon tandems. Black spheres, PSCs containing solution-processed layers. Red triangle, all-vacuum-deposited PSCs. The references are listed in Supplementary Table 2. f, Photos of a laminated all-vacuum-deposited perovskite-on-silicon tandem before and after five months of outdoor ageing in Italy. g, Top panel: normalized PRtc for an all-vacuum-deposited perovskite-on-silicon tandem cell (EvapTex1) and a reference HJT Si cell (RefHJT) under outdoor ageing conditions at MPPT for over eight months. The sample size is 1 for each condition. The error bands around PRtc indicate the 10% to 90% percentile range weighted by power. Bottom panel: Tcell,wt (left-hand axis) and GPOA (right-hand axis) over the entire ageing period.Full size imageThe rarely reported stability of WBG PSCs under harsh conditions, and the much shorter reported operational lifetime of all-vacuum-deposited PSCs39,41 relative to solution-processed counterparts, urge accelerated indoor stability measurements mimicking the real solar spectrum and elevated temperature to reliably predict field lifetime (Supplementary Note 6)57,58,59. Therefore, we conduct an ageing study on complete PSCs under the stringent ISOS-L-2 protocol combining light (full-spectrum simulated sunlight (0.76 sun) with no UV filter, Supplementary Fig. 23) and heat stressors (75 ± 5 °C) at OC (Supplementary Fig. 24). Our evaporated PSCs retain 80% of peak performance (T80) after ~840 and 330 h for the SnO2- and bathocuproine (BCP)-based devices, respectively, representing outstanding stability for WBG PSCs (Fig. 3c and Supplementary Figs. 25–29). Of particular note, our champion 0.25-cm2 evaporated cell reaches a T80 over 1,080 h (T80,Champ = 1,080 h) with no change in visual appearance (Fig. 3d and Supplementary Fig. 24b). Similar stability (T80,Champ = 900 h) is realized on a 1-cm2 champion cell. In contrast, the solution-processed cells only achieve T80 30% (Fig. 5e,f and Supplementary Fig. 50). The increases in nOC and nSC upon ageing are probably due to an increased fraction of Shockley–Reed–Hall recombination. This may partially arise from deteriorating interface as evidenced by the substantial drop in device VOC. In addition, the fast decay in JSC indicates deterioration of the optoelectronic properties for the solar cell, which can also contribute to higher n. Similarly, using nOC = 1.011, nSC = 1.012 at 1,150 h and nOC = 1.12, nSC = 1.19 at 3,000 h substantially reduces the discrepancies between ex-JSC and real JSC for the evaporated cell to 1,250 h, Supplementary Fig. 49). Interestingly, we observe a second population peak with lower QCE in the fresh and aged evaporated cell (Fig. 5j), which arises from a ‘defective’ area with slightly lower initial IPL at OC and degrades faster than the other regions (Fig. 5c,d). However, the device performance demonstrates a strong tolerance to this defective area because most of the mapping areas retain high QCE. The origin of this defective area is unclear, but a possible interpretation would be charge-extraction loss due to non-uniform charge transport layers.In addition to a less considerable shift towards lower QCE in the co-evaporated perovskites, the population distribution roughly maintains its featured ‘two-peak’ shape after ageing (Fig. 5j). By contrast, the population distribution for solution-processed perovskites substantially deviates from its initial shape upon ageing (Fig. 5i). The current density losses that we observe in our solution-processed devices most likely originate from a drop in the QCE, which we postulate to be induced by the photogeneration and spatial redistribution of mobile ionic species70,80. Our crystal-facet-directed co-evaporated perovskites effectively mitigate charge-extraction loss, thus highlighting the significance of forming highly oriented, highly crystalline, phase-pure perovskites for stable WBG PSCs and perovskite-on-silicon tandems.DiscussionOur findings demonstrate that our recipe for growing WBG (1.67-eV) perovskites via co-evaporation results in films with large grains of high crystallinity with ‘face-up’ orientation, high phase purity and homogeneity. The high material quality leads to good material and device stability under stringent heat and light stressors. Notably, our evaporated PSCs achieve excellent long-term stability on par with state-of-the-art solution-processed PSCs, which contain stabilizing additives. We further provide spectral insight into halide segregation and trap-mediated recombination, correlating microscopic luminescence features with macroscopic device performance, revealing that sustaining high charge carrier extraction efficiency is the root cause for the enhanced stability of our evaporated PSCs. Furthermore, we demonstrate all-vacuum-deposited perovskite-on-silicon tandems with robust operational longevity under outdoor conditions, highlighting the great promise of realizing efficient, durable PSCs via solvent-free fabrication protocols.MethodsPrecursor material preparationITO-coated glass substrates (15 Ω cm−2, Biotain), lead(II) iodide (PbI2, 99.999%, trace metal basis, Tokyo Chemical Industries), lead(II) bromide (PbBr2, 99.999%, Alfa-Aesar), lead(II) chloride (PbCl2, 99.999%, trace metal basis, Sigma Aldrich), FAI (>99.999%, Dynamo), caesium iodide (CsI, 99.9%, metal basis, Alfa-Aesar), Me-4PACz (>99.0%, Tokyo Chemical Industry), (2-(3,6-dimethoxy-9H-carbazol-9-yl)ethyl)phosphonic acid (MeO-2PACz, >99.0%, Tokyo Chemical Industry), (4-(3,6-diiodo-9H-carbazol-9-yl)butyl)phosphonic acid (I-4PACz, >99.0%, Lumtec), aluminium oxide nanoparticles (Al2O3, 99%, Sigma Aldrich), silver pellets (Ag, 99.999%, Kurt J. Lesker Company), gold pellets (Ag, 99.999%, Kurt J. Lesker Company), chromium bar (Cr, 99.999%, Kurt J. Lesker Company).Ethanol (anhydrous, ≥99.9%, VWR), 2-propanol (IPA, anhydrous, 99.5%, Sigma Aldrich), N,N-dimethylformamide (DMF, anhydrous, 98%, Sigma Aldrich), dimethyl sulfoxide (DMSO, anhydrous, 98%, Sigma Aldrich), anisole (anhydrous, 99.7%, Sigma Aldrich), chlorobenzene (CB, anhydrous, 99.5%, Sigma Aldrich).In this work, all the materials were used as received without further purification and weighed in a nitrogen-filled glovebox without exposure to light. The perovskite precursor solutions were stirred overnight in a nitrogen-filled glovebox and used without any further treatment.Film deposition and device fabricationITO–glass substrates were cleaned by sonication in Decon90/water (2 vol.%), scrubbed and rinsed. They were then sequentially sonicated in deionized water, acetone and IPA (15 min each), dried with N2 and treated with UV–ozone for 15–30 min before being transferred into a N2 glovebox.Hole transport layersFor co-evaporated perovskites, a 5-nm Spiro-TTB HTL was thermally evaporated at 0.1 Å s−1 (chamber pressure ~1 × 10−7 mbar). For solution-processed perovskites, two HTLs were prepared: Spiro-TTB (5 mg ml−1 in CB) was spin-coated at 5,000 r.p.m. for 30 s and annealed at 120 °C for 10 min. Me-4PACz (0.3 mg ml−1 in EtOH) was spin-coated at 3,000 r.p.m. for 30 s and annealed at 100 °C for 10 min. Both solution-derived HTLs were hydrophobic. Therefore, an Al2O3 nanoparticle wetting layer (1:150 vol.% in IPA) was spin-coated at 2,000 r.p.m. for 20 s on HTLs before perovskite deposition.Co-evaporated FA0.83Cs0.17Pb(IxBryCl1 −x−y)3 perovskitesTo deposit the reference WBG perovskite FA0.83Cs0.17Pb(I0.80Br0.20)3, five precursor sources (2× FAI, PbI2, CsI and PbBr2) were co-evaporated in a perovskite deposition chamber (details shown below) that is integrated as part of the National Thin-Film Cluster Facility at Oxford, under high vacuum (<1 × 10−6 mbar), using quartz-crystal microbalances for precise rate control. The deposition rates were set to 0.2 Å s−1 (FAI, 165 °C), 0.397 Å s−1 (PbI2, 310 °C), 0.074 Å s−1 (CsI, 435 °C) and 0.125 Å s−1 (PbBr2, 285 °C) to achieve the target stoichiometry, yielding a final thickness of 500 nm. In some preliminary experiments, 10-nm seed layers of PbCl2, CsCl or CsPbCl3 were deposited before the perovskite.For the facet-engineered, seed-free growth, a six-source co-evaporation approach was adopted. Here, 5 mol% of PbI2 was replaced with PbCl2, leading to a nominal composition of FA0.83Cs0.17Pb(I0.75Br0.20Cl0.05)3, with PbI2 and PbCl2 deposited at 0.354 Å s−1 (300 °C) and 0.027 Å s−1 (335 °C), respectively. The bandgap was tuned between 1.65 and 1.72 eV by varying the PbBr2 fraction (x = 0.17–0.28) while keeping other rates unchanged. Before each run, crucibles were replenished to maintain consistent charge levels, and all depositions were performed on unheated substrates (~27 °C). Unless otherwise noted, as-deposited films were subsequently annealed at 135 °C for 30 min in ambient air (40–50% RH).Solution-processed FA0.83Cs0.17Pb(I0.75Br0.20Cl0.05)3 perovskitesPrecursor solutions (1.4 M) were prepared in a DMF:DMSO mixture (4:1 v/v) from stoichiometric amounts of PbI2, PbBr2, FAI, CsI and PbCl2 in a N2-filled glovebox. After full dissolution, 170 µl of the solution was dynamically dispensed onto a substrate spinning at 1,000 r.p.m. The spin-coating programme was accelerated to 5,000 r.p.m. over 5 s, held for 35 s and quenched with 325 µl of anisole 5 s before the end. The as-cast films were transferred to ambient air (40–50% RH) and annealed at 135 °C for 30 min. Once cooled, the substrates were returned to the glovebox for subsequent processing steps.Electron-transport layers (ETLs)C60 (20-nm) and BCP (5-nm) layers were deposited on top of the perovskite films via thermal evaporation under vacuum (~1 × 10−7 mbar) with active-area masks applied. For devices with SnO2 buffer layers, 20 nm of SnO2 was grown in the ALD chamber using tetrakis(dimethylamido)tin(IV) (TDMASn) and deionized water precursors with 140 pulses. The reactor temperature during pulsing was stabilized at 100 °C while TDMASn was kept at 75 °C.Metal contactsSilver (100 nm) was thermally evaporated through shadow masks (active area defined as 0.25 or 1 cm2) at an initial rate of 0.2 Å s−1 (ramped up to 1 Å s−1 in 20 min) under high vacuum (~10−7 mbar) in a thermal evaporator (Nano36, Kurt J. Lesker). The devices for stability tests used 3.5-nm Cr and 100-nm Au gold electrodes instead of silver.Perovskite deposition chamberThe perovskite thin-film layer is fabricated in an Angstrom Engineering deposition chamber under high vacuum, achieved through a two-step process combining an Ebara EV-A10-2 dry pump and a CTI Cryo Torr 8F cryogenic pump. Chamber pressure is monitored in real time using an INFICON Gemini MPG500 gauge, and the system is vented with high-purity nitrogen (99.999%).Inside the chamber, eight identical thermal evaporation sources (Luxel RADAK II type) are arranged to minimize cross-contamination, each equipped with a 10-cm3 alumina crucible and individually shielded. Source shutters are actuated by oil-free compressed dry air and controlled via software. Evaporation rates from each source are precisely measured using water-cooled quartz-crystal microbalances that are also shielded to prevent signal interference.A rotatable substrate stage (0–30 r.p.m.) enables combinatorial deposition through a programmable shutter, allowing gradients in thickness and composition for high-throughput screening. The stage can be resistively heated to 400 °C with ±2.5 °C stability. System integrity is monitored using a Stanford Research Systems RGA-300 residual gas analyser, which detects impurities and interlocks the gate valve to prevent contamination of connected cluster chambers.Sputter systemThe sputtering system (Angstrom Engineering) is integrated into the cluster tool, enabling automated, mask-based deposition sequences for full PV device fabrication. Capable of coating substrates up to M2 size with 96.6% uniformity, the chamber is evacuated by an Ebara backing pump and a Pfeiffer turbo pump to a base pressure of 1 × 10−8 mbar. Process gases (Ar, N2, O2, CF₄ and H2) are supplied from BOC or a hydrogen generator and regulated through mass-flow controllers via the AERES software.An indium zinc oxide (IZO, (In2O3)90(ZnO)10) target bonded to a copper backplate was used as a critical window/interlayer for tandem devices. The target was mounted at a tilt angle of 32° and a target-to-substrate distance of 25 mm to optimize uniformity. Before deposition, the target was plasma-cleaned for 180 s with the substrate shutter closed. IZO layers (100 nm) were sputtered at room temperature using a gas mixture of Ar (18.1 sccm) and O2 (0.3 sccm) under a process pressure of 3 × 10−3 mbar, with the substrate rotating at 10 r.p.m. Deposition parameters were 410 V, 289 mA and 118.4 W, yielding a rate of 1.14 Å s−1 over 175 s. After deposition, the substrate was transferred under high vacuum back to the load lock.Evaporated perovskite-on-silicon tandem solar cellEvaporated perovskite-on-silicon tandem solar cells were fabricated on industry-standard M2-sized n-type Czochralski silicon wafers. The silicon-bottom subcells were prepared using HJT technology: wafers were KOH-textured to form random pyramids (1–5 μm), cleaned with an ozone- and HF-based process and passivated with a final HF dip. Intrinsic and doped hydrogenated amorphous silicon (a-Si:H) layers were deposited by plasma-enhanced chemical vapour deposition. For the rear contact, 70 nm of ITO (97/3 wt% In2O3/SnO2) was sputtered; a 12-nm ITO layer was deposited on the front side to serve as the recombination junction. After laser dicing into 3 × 3 cm2 substrates, the wafers were annealed at 180 °C for 20 min in ambient air to heal sputter-induced damage.Before perovskite deposition, substrates were treated with UV–ozone for 15 min. For devices with an evaporated HTL, 5 nm of Spiro-TTB was thermally evaporated. For tandems employing a solution-processed SAM HTL, a 1:1 mixture of MeO-2PACz and I-4PACz (each 0.25 mg ml−1 in IPA) was spin-coated at 4,000 r.p.m. and annealed at 110 °C. The 1.67-eV WBG perovskite absorber (FA0.83Cs0.17Pb(I0.75Br0.20Cl0.05)3) was deposited via the six-source co-evaporation method described earlier, with the thickness increased to 800 nm to enhance light harvesting. After deposition, the stack was annealed at 135 °C for 30 min in ambient air. For SAM HTL devices, a post-deposition passivation layer (0.5 mg ethane-1,2-diammonium iodide (EDAI2)+ 0.25 mg phenethylammonium chloride (PEACl) in 1 ml of 3:1 IPA:CB) was applied81,82,83, followed by spin-coating and annealing at 100 °C for 3 min.The electron-transport stack consists of 15 nm of thermally evaporated C60, 20 nm of ALD SnO2 and 70 nm of sputtered IZO as the semi-transparent top contact. Front metal contacts were formed by evaporating 250 nm of Ag through an active-area mask, followed by 700 nm of Ag fingers. The rear contact was completed with 200 nm of evaporated Ag. The device active area is 1 cm2.Film and device characterizationUltraviolet–visible (UV–Vis) absorption spectroscopyAbsorbance spectra were measured with a Varian Cary 300 Bio UV–Vis spectrophotometer with a 50 × 50 mm2 reflective neutral-density filter with an optical density of 3.0 (made out of UV-fused silica). Tauc plot analysis was used to estimate the material bandgap assuming a direct bandgap.X-ray diffractionThe one-dimensional XRD patterns were obtained with a Panalytical X’Pert Pro X-ray diffractometer with a Cu Kα1 (1.54060-Å) source.Ex situ grazing-incidence wide-angle X-ray scatteringEx situ GIWAXS measurements were performed on a Rigaku SmartLab diffractometer equipped with a HyPix-3000 2D detector and a rotating Cu Kα source (8.048 keV). The sample-to-detector distance was 65 mm, and incidence angles of 0.5° and 1° were used. The X-ray beam was shaped using a parallel-beam optic, a 0.5° in-plane collimator and a 0.1-mm incident slit. Each scan was acquired for 60 min. Detector images were integrated into Q-space and processed into azimuthally averaged one-dimensional profiles using custom scripts based on the pyFAI and pygix libraries. Variable-angle GIWAXS scans (αᵢ = 0.5°–1°) were acquired for 60 min each using parallel-beam optics, a 0.5° in-plane collimator, a 0.1-mm incident slit and a 10-mm length-limiting slit. Data were processed as described above.In situ grazing-incidence wide-angle X-ray scatteringIn situ GIWAXS measurements were performed at beamline I07 of the Diamond Light Source. Perovskite films were annealed on a temperature-controlled hot plate under a nitrogen atmosphere. A 10-keV synchrotron beam struck the sample at a grazing incidence angle of 0.5°, and scattering patterns were recorded with a Pilatus 2M detector (calibrated using silver behenate). Solution-processed films were spin-coated on site within a fume hood, while evaporated films were prepared in Oxford and transferred via a vacuum-sealed tube. All data (experiment SI39532) were integrated into Q-space and azimuthally processed using pyFAI and pygix.Scanning electron microscopyAn FEI Quanta 600 FEG environmental scanning electron microscope was employed to investigate perovskite layer morphology. Accelerating voltages between 4 and 15 kV were employed for various analyses.Nuclear magnetic resonance spectroscopyA two-channel Bruker AVANCE III HD NanoBay 400-MHz instrument running TOPSPIN 3 equipped with a 5-mm z-gradient broadband/fluorine observation probe is used. The signal from deuterated ethanol solvent is used for reference.Photoluminescence quantum efficiency (PLQE) measurementPLQE measurements were acquired using a custom-built PLQE set-up in an integrating sphere. Samples were photoexcited using a 450-nm laser (laser power = 0.665 mW). The PL was collected using a high-resolution monochromator and hybrid photomultiplier detector assembly (PMA Hybrid 40, PicoQuant GmbH). The PLQE was extracted from the photon energy (hf) and photon numbers of the excitation and emission obtained from numerical integration using Python.Time-of-flight secondary ion mass spectrometry (ToF-SIMS)Perovskite samples for ToF-SIMS measurements were prepared following the same procedure as used for solar cells. Measurements were conducted using a hybrid SIMS instrument (IONTOF M6), which was equipped with a bismuth primary ion source and an O2+ sputter source to probe the positive atomic and/or fragment ions. The ToF-SIMS data were acquired over an area of 150 × 150 μm2 using a 30-keV Bi+ primary ion beam. This was followed by a sputtering process, with each cycle lasting for 3 s. During each sputtering cycle, a 400 × 400 μm2 area of the sample was bombarded with 1-keV O2+ ion beams in an interlaced mode. The sample holder, containing the samples, was sealed in a container filled with argon gas and was transferred into the instrument immediately before testing.Capacitance–voltage (C–V) profilingThe C–V characteristics were taken at 20 kHz with AC perturbation of 20 mV using an LCR meter (E4980A, Keysight). The scan frequency was determined from the plateau region of the capacitance–frequency profile at zero bias. The cells were kept in the dark at room temperature to reach equilibrium. The Mott–Schottky plot was generated from the C–V profile data by calculating (A/C)2, where A is the cell unit area (0.25 cm2) and C is the capacitance. The electrically charged defect density profile N(x) = −2[dC(x)−2/dV]−1/qε0εrA2, where q is the elementary charge, ε0 is the vacuum permittivity and εr is the relative permittivity of the material, can be calculated as a function of profiling distance x = ε0εrA/C from the Mott–Schottky plot.Characterization of solar cellsJ–V and MPPT data were acquired in ambient air using a Keithley 2400 series source meter under simulated AM1.5G illumination (WAVELABS SINUS-220 simulator) and in the dark. The active area was defined by a black-anodized metal aperture (0.25 or 1.00 cm2) in a light-tight holder. Each device was characterized sequentially: first, VOC was recorded after 6 s of steady-state illumination; then, reverse (OC to SC) and forward (SC to forward-bias) J–V scans were performed at 245 mV s−1; this was followed by 30 s of active MPPT using a gradient-descent algorithm to extract the stabilized power output (ηMPPT); finally, JSC was measured after 3 s under steady-state short-circuit conditions. The simulator intensity was calibrated before each measurement session using a KG3-filtered Si reference diode (Fraunhofer ISE) to match its certified 1-sun JSC. Additionally, the internal spectrometer of the solar simulator provided a real-time spectral check; the ratio of this internal reading to the value recorded during calibration yielded an equivalent irradiance factor (here 0.985–1.005 suns), which was applied to correct the reported power conversion efficiencies.EQE measurementEQE was measured using a custom system built around a Bruker Vertex 80v Fourier-transform interferometer. A 250-W quartz–tungsten halogen lamp provided the illumination, which was monochromated, mechanically chopped at 280 Hz and focused onto the masked device (0.25 or 1 cm2). The resulting AC photocurrent was converted to a voltage across a 50-Ω series resistor and recorded with a lock-in amplifier. The EQE spectrum was obtained by dividing the device’s photocurrent response by that of a calibrated silicon reference cell (Thorlabs FDS100-CAL).Optical microscopyOptical microscope images were taken on a Nikon Eclipse LV100ND microscope with Nikon TU Plan Fluor lenses (10×/0.30 A, 20×/0.45 A, 50×/0.60 B, 100×/0.90A). The images were taken with an attached Nikon digital camera D6.10.Photo- and thermal stability testPSCs designated for stability testing were first sputter-coated with a 300-nm SiOₓ layer through an active-area mask to shield the device from the UV-curable encapsulant. Encapsulation was performed in a nitrogen-filled glovebox using a cover glass (28.0 × 21.5 × 1.1 mm3) and UV-activated adhesive (Eversolar AB-341). The adhesive was uniformly applied across the substrate and cured under UV light for 3 min. For accelerated ageing, encapsulated devices were placed in an Atlas SUNTEST CPS+ chamber equipped with a 1,500-W air-cooled xenon lamp, which provided full-spectrum AM1.5G-equivalent illumination at an irradiance of approximately 76.5 mW cm−2. No UV filter was employed (Supplementary Fig. 23a). All ageing tests were conducted under OC conditions, in accordance with the ISOS-L-2 protocol.To monitor the actual device temperature during ageing, we used a black standard thermocouple supplied with the chamber (setpoint 85 °C) alongside custom ‘mock-device’ thermocouples. The latter were bonded to black-anodized aluminium foil attached to ITO-coated glass slides, replicating the optical and thermal properties of the test solar cells. These mock devices, positioned identically to the test cells, recorded a stabilized temperature of 75 ± 5 °C. We hence moderate our estimation of our cell temperature in the ageing boxes to be 75 ± 5 °C. We note that in our previous publications where we stated 85 °C ageing the temperature was recorded on the Atlas-supplied black temperature standard, and hence likely to be 75 ± 5 °C, rather than 85 °C. The chamber was air-cooled, and the laboratory ambient RH (at ~21 °C) varied between 50% and 60% throughout the tests. Periodically, devices were removed for J–V characterization using the standard measurement protocol described previously.For stand-alone perovskite thin-film stability tests, a protective poly(methyl methacrylate) layer (10 mg ml−1) was spin-coated onto the films to prevent physical damage, and to act as a rudimentary encapsulation.Outdoor stability testThe perovskite-on-silicon tandem cells were prepared at Oxford and sent to CEA, France, for lamination. The laminated cells were then sent to a test facility that was deployed at Eurac Research in Bolzano, Italy (coordinates: 46° 28′ 31.7″ N 11° 19′ 49.7″ E) that allows for the measurement of perovskite-on-silicon tandem devices. The samples were installed on a frame with a tilt of 30° and an orientation of 190° from the north. The set-up was designed to measure electrical parameters of up to 24 small-scale encapsulated PV cells, with voltages up to 2.0 V and currents up to 100 mA at resolutions of 1 mV and 2 mA respectively (Fig. 3f). It monitors current, voltage and power at the MPPT using micro-MPPTs from Ljubljana University at a time resolution of 1 min. PT100 sensors concurrently measure cell temperature at the outer backside. All data were collected using a Python script. In addition to the device measurements, we collected plane-of-array irradiance data using a pyranometer and two crystalline-silicon-based reference cells, and weather data including ambient temperature, RH and precipitation intensity and type.Hyperspectral operando microscopy characterizationHyperspectral microscopy was performed using a Photon etc. IMA-VIS inverted system with 5× (Olympus MPLFLN 5×) and 20× (Olympus LCPLFLN 20× LCD) semi-apochromat objectives. Samples were transferred from a nitrogen glovebox to a motorized stage for in situ PL mapping during ageing. A 532-nm continuous-wave laser was used for optical excitation. For the WBG sample used in characterization, the equivalent 1 sun is equal to 59 mW cm−2. For hyperspectral image acquisition, a laser power equivalent to 1.71 sun was used. For ex-JV, a laser power equivalent to 1 sun was used so that the measured JSC under the microscope matched the JSC of the device measured under our standard solar simulator. Illumination and detection were conducted through the same objective using a beam-splitter, enabling PL collection between 547 and 1,000 nm.For spectral mapping, emitted light was dispersed by a volume Bragg grating with a spectral step of 2 nm over the range 650–850 nm; broadband mapping was acquired without the grating. A cooled Hamamatsu ORCA-Flash4.0 V3 camera (2,048 × 2,048 pixels, maintained at ~10 °C) recorded the signal. Absolute photon counts were obtained via a two-step calibration procedure described previously68. The technique was applied to evaporated and solution-processed perovskite films with different electron-transport layers, and during accelerated ageing experiments.FTPS measurementFTPS was performed using a Bruker Vertex 80v interferometer, an AM1.5-filtered xenon lamp and a trans-impedance amplifier. Unencapsulated PSCs (0.25 cm2) were held at OC in ambient air (~45% RH) and intermittently illuminated at 1-sun intensity. To acquire each EQE spectrum, the lamp power was briefly reduced to 0.1 sun to accommodate the amplifier’s gain limits. A long-pass filter (715 nm for evaporated devices, 780 nm for solution-processed ones) blocked above-bandgap light, extending the subgap sensitivity by four orders of magnitude. All spectra were calibrated against a certified silicon reference cell.To extract the Urbach energy (EU), each EQE spectrum was fitted in the Urbach-tail spectral region, near the absorption edge, with the equation$$\mathrm{EQE}=\exp \left(\frac{{hv}-{E}_{0}}{{E}_{{\rm{U}}}}\right)$$ (1) where E0 is a constant.Accelerated ageing for the sample used in hyperspectral microscopy characterizationFor the hyperspectral microscopy ageing study, samples were aged under the ISOS-L-2 protocol: 65 ± 5 °C, 1-sun AM1.5G illumination (no UV filter, Supplementary Fig. 23b), OC bias and 80–90% RH. The tests were conducted in a Sunirad A-22 chamber (Lumartix SA) equipped with a plasma lamp. Encapsulated samples were placed face down (glass side up) and masked to expose only the 0.25- or 1-cm2 active area. During ageing, samples remained inside the chamber and were periodically removed for J–V characterization and hyperspectral microscopy.Charge-extraction pseudo-JV During ageing, PL mappings were acquired at 0, 1,150 and 3,000 h. For each measurement, the sample was mounted with its electrode facing upward and contacted via a probe connected to a Keithley 2400. A ×5 objective uniformly illuminated the entire 0.25-cm2 active area, and broadband PL was collected without spectral filtering. Mappings were recorded every second while the bias was stepped at 0.01 V s−1, collecting 140 consecutive maps in both forward and reverse scans from VOC to 0 V (VSC). QCE was calculated per pixel (Supplementary Note 9). From these maps, area-averaged external J–V curves and spatially resolved QCE at VOC, VMPP and VSC were extracted. Bright and dark regions were distinguished by applying a threshold filter at VSC.Reporting summaryFurther information on research design is available in the Nature Portfolio Reporting Summary linked to this article.