Incorporating thermal co-evaporation in current-matched all-perovskite triple-junction solar cells

Incorporating thermal co-evaporation in currentmatched all-perovskite triple-junction solar cells† Terry Chien-Jen Yang, ‡ ab Taeheon Kang,‡ a Melissa Fitzsimmons, a Guadalupe Vega, c Yang Lu, ab Leo Rosado, d Alberto Jim´ enez-Solano, e Linfeng Pan, a Szymon J. Zelewski, abf Jordi Ferrer Orri, abg Yu-Hsien Chiang, ab Dengyang Guo, ab Zher Ying Ooi, a Yutong Han, a Weidong Xu, a Bart Roose, ab Caterina Ducati, g Sol Carretero Palacios, d Miguel Anaya ac and Samuel D. Stranks * ab Thermal co-evaporation of halide perovskites is a solution-free, conformal, scalable, and controllable deposition technique with great potential for commercial applications, particularly in multi-junction solar cells. Monolithic triple-junction perovskite solar cells have garnered significant attention because they can achieve very high efficiencies. Nevertheless, challenges arise in fabricating these devices, as they require multiple layers and precise current matching across complex absorber stacks. Here we demonstrate a current-matched monolithic all-perovskite p–i–n triple-junction solar cell enabled by controlled thermal co-evaporation of various absorber layers in the stack. The top and middle subcells were fabricated by developing optimized thermally co-evaporated Cs 0.3 FA 0.7 Pb(I 0.56 Br 0.44 ) 3 (1.80 eV bandgap) and FAPbI 3 (1.53 eV) perovskites, respectively, while the bottom subcell employed a solutionprocessed Cs 0.25 FA 0.75 Pb 0.5 Sn 0.5 I 3 (1.25 eV) perovskite. By optimising absorber thicknesses and compositions through optical modelling, we achieve excellent current matching between the top (9.6 mA cm −2 ), middle (9.3 mA cm −2 ), and bottom subcells (9.0 mA cm −2 ), achieving an overall efficiency of 15.8%. Optical modelling simulations suggest that current matching and efficiency up to 11.4 mA cm −2 and 37.6% respectively could be attainable using the latest interlayer materials. This work highlights the potential of scalable vapour-based deposition techniques for advancing multi-junction perovskite-based solar cells, paving the way for future developments in this field. Broader context Single-junction crystalline silicon solar cells, the dominant technology in photovoltaics today, are nearing their maximum theoretical efficiency of 29.4%. To continue to reduce the cost of solar energy, either in terms of $ per m 2 or $ per W, higher efficiency is essential. One of the most promising and well-established strategies to exceed the single-junction efficiency limit is by stacking absorbers with varying bandgaps in a multi-junction cell. Bandgap tunable perovskite materials have the potential to be the next mainstream photovoltaic technology. In particular, multi-junction perovskite-based solar cells have garnered signicant attention because they can achieve very high efficiencies. One method of depositing perovskite lms is thermal co-evaporation which is a vacuum technique that is solution-free, conformal, scalable, and highly thickness controllable. Thus, it is very attractive for commercial applications especially in monolithic multi-junction perovskite-based solar cells where current matching between the subcells is critical. We demonstrate a highly current-matched monolithic all-perovskite triple-junction solar cell enabled by controlled thermal co-evaporation of various perovskite absorber layers in the stack and aided by feedback through optical modelling simulations using the refractive index data from these materials. This research paves the way for expanding the use of vacuum-deposition thermal co-evaporation techniques in perovskite-based multi-junction research and development. a Department of Chemical Engineering and Biotechnology, University of Cambridge, Cambridge CB3 0AS, UK b Department of Physics, Cavendish Laboratory, University of Cambridge, Cambridge CB3 0HE, UK. E-mail: [email protected] c Departamento F´ ısica de la Materia Condensada, Instituto de Ciencia de Materiales de Sevilla, Universidad de Sevilla−CSIC, Calle Am´ erico Vespucio 49, 41092 Sevilla, Spain d Instituto de Ciencia de Materiales de Madrid, ICMM-CSIC, 28049 Madrid, Spain e Departamento de F´ ısica, Universidad de C´ ordoba, Edicio Einstein (C2), Campus de Rabanales, 14071 C´ ordoba, Spain f Department of Experimental Physics, Faculty of Fundamental Problems of Technology, Wrocław University of Science and Technology, 50-370 Wrocław, Poland g Department of Materials Science and Metallurgy, University of Cambridge, Cambridge CB3 0FS, UK †Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d4el00012a ‡These authors contributed equally to the work. Cite this: EES Sol.,2025,1,41 Received 27th September 2024 Accepted 27th December 2024 DOI: 10.1039/d4el00012a rsc.li/EESSolar © 2025 The Author(s). Published by the Royal Society of Chemistry EES Sol.,2025,1,41–55 | 41 EES Solar PAPER Open Access Article. Published on 21 January 2025. Downloaded on 8/12/2025 11:56:44 AM. This article is licensed under a Creative Commons Attribution 3.0 Unported Licence. View Article Online View Journal | View Issue Introduction Perovskite solar cells (PSCs) have become one of the most popular research topics in the eld of photovoltaics due to their high-efficiency, bandgap tunability, low cost, and potential for commercialisation. The power conversion efficiency (PCE) of single-junction PSCs has increased from 3.8% (2009) to 26.7% (2024), 1 which can be compared to the theoretical Shockley– Queisser (SQ) single-junction efficiency limit of 33%. 2,3 Multijunction (MJ) devices using various bandgap materials can achieve higher efficiency through better utilization of the solar spectrum by reducing both the above bandgap thermalization and below bandgap non-absorbed losses. Intensive efforts with double-junction monolithic tandems for both perovskite– perovskite 4–15 and perovskite–silicon 16–28 tandem solar cells have been widely demonstrated, with reported efficiency up to 34.6%. 1 However, only a handful of 2-terminal monolithic perovskite-based triple-junction devices, either perovskite– perovskite–perovskite 29–32 (PPP), perovskite–perovskite– silicon, 33–40 and even perovskite–perovskite–organic, 41 have been reported thus far. Hörantner et al. 42 in 2017 reported through modelling studies that perovskite-based triple-junctions could potentially reach practical efficiencies up to 39%. Drawing reference to another existing photovoltaic (PV) technology, the record for the triple-junction III–V solar cell at 1-Sun AM1.5 has already reached 39.5%. 43,44 The record device consists of carefully engineered III–V materials with an ideal bandgap combination of 1.88, 1.33, and 0.92 eV for the top, middle, and bottom subcells respectively, albeit with expensive absorber layers. Nevertheless, experimentally reported efficiencies of 2-terminal monolithic PPP and perovskite–perovskite– silicon triple-junction solar cells are only at 24.3% (ref. 32) and 27.6% (ref. 40) (31.5% (ref. 45) for 4-terminal), respectively, and far from their double-junction counterparts. The rst reason is the complexity in terms of the number of functional layers that need to be deposited. A triple-junction solar cell requires a minimum of about 14 functional layers, which means it takes signicant time and resources to even fabricate a single batch of devices, leading to slower feedback cycles than lower junction number analogues. Furthermore, if there is an issue with any of the layers, for example the uniformity (both vertically and laterally), composition, or thickness, then the nal performance of the triple-junction device would be suboptimal. Initial efforts have been made to allow identication of problematic layers in MJ devices, 46 but more work is needed in this area. The second reason is the stringent current matching criteria necessary for monolithic (series-connected) MJ solar cells. In a monolithic perovskite–perovskite tandem device, bandgap and current matching is required between only two absorbers. In a triplejunction PPP device, three absorbers must be optimally matched in terms of bandgap and current, creating more degrees of freedom (see ESI Note 1†). This makes precise current-matching more challenging, requiring judicious optimisation and a deep understanding of the device stack, such as through advanced optical and electrical modelling. A third challenge involves material compatibility, particularly during the deposition and annealing of subsequent functional layers, where process conditions must align without degrading the performance of underlying layers. These challenges are worth tackling and solving them will ultimately allow all-perovskite triple-junction device efficiencies to surpass their tandem counterparts (see ESI Note 2 and Table 1†for existing devices in literature). Thermal co-evaporation is a conformal, upscalable, and highly thickness-controllable vacuum deposition technique for fabricating perovskite lms. 47–50 Such vapour-based techniques are very promising for the industrial commercialisation of MJ perovskite solar cell technology as seen with other thin lm technologies including (single-junction) CdTe. The multisource co-evaporation method 13,51–54 has the advantage of ne tuning of the material composition and thus bandgap. This technique was pioneered for halide perovskites by the likes of Snaith et al. 55 and Bolink et al., 47,56 but to date far less effort has been made on thermal evaporation than solution processed equivalents. Similarly, the majority of perovskites used in MJ solar cells to date have been solution-processed, with only a few notable exceptions demonstrating co-evaporated perovskite layers in tandem solar cells. 6,13,52,57,58 The current state-of-the-art 24.3% monolithic triple-junction PPP solar cell made by Wang et al. 32 was fabricated using solution-processed perovskites for all three of the subcells. For tandem devices where evaporated perovskites have been used, the record for perovskite–perovskite and perovskite–silicon are 24.1% (ref. 13) and 24.6% (ref. 59), respectively. So far, there has been no demonstration of a triple-junction perovskite solar cell made with thermal coevaporation. In this work, we rst demonstrate an optimized thermally coevaporated FAPbI 3 perovskite composition, which has a suitable bandgap of 1.53 eV as the middle subcell in our triple-junction device. We then present the fabrication of a monolithic triplejunction PPP solar cell using thermally co-evaporated perovskite top and middle layers. This technique enables precise control over bandgap and thicknesses, driven by optical modelling and experimental feedback. By optimising the deposition process, we achieved exceptional current-matching across the top, middle, and bottom subcells, culminating in a short-circuit current (J SC ) of 9.3 mA cm −2 , among the highest reported for triple-junction PPP solar cells. Results and discussion Optimisation of a co-evaporated FAPbI 3 middle cell In our previous work on thermal co-evaporation, 13 we demonstrated wide-bandgap (up to 1.80 eV) Cs 0.3 FA 0.7 Pb(I x Br 1−x ) 3 perovskites which were used as a top cell for a 2-terminal perovskite–perovskite tandem solar cell. Here we selected a nominal perovskite composition of Cs 0.3 FA 0.7 Pb(I 0.56 Br 0.44 ) 3 (1.80 eV) as a reliable base process for the top wide-bandgap absorber for our nal PPP triple-junction solar cell. Thermal coevaporation of the Cs 0.3 FA 0.7 Pb(I x Br 1−x ) 3 perovskite required a 10% PbI 2 excess 60 (with respect to a nominally stoichiometric composition) for efficient devices. In order to attain a suitable mid-bandgap perovskite for use in our PPP triple-junction 42 |EES Sol.,2025,1,41–55 © 2025 The Author(s). Published by the Royal Society of Chemistry EES Solar Paper Open Access Article. Published on 21 January 2025. Downloaded on 8/12/2025 11:56:44 AM. This article is licensed under a Creative Commons Attribution 3.0 Unported Licence. View Article Online devices, we turn to a thermally co-evaporated FAPbI 3 rst demonstrated by Borchert et al. 61 in 2017, which has a suitable bandgap of around 1.5 eV for triple-junction solar cells as shown in previous models. 62 Note, we use a set of controlled checks from Tauc plots, X-ray diffraction (XRD) and external quantum efficiency (EQE) absorption onsets to gauge our nominal perovskite compositions. 13,60 We thermally co-evaporate FAPbI 3 (1.53 eV) by simultaneously evaporating both FAI and PbI 2 powders onto the rotating sample stage, as shown in Fig. 1a. Through this optimisation process, we found that the performance of resulting FAPbI 3 perovskite solar cells is highly sensitive to the ratio between the PbI 2 and FAI rates. We set the PbI 2 rate constant at 0.6 Å s −1 whilst varying, in separate evaporation deposition runs, the FAI rate from 0.6 Å s −1 up to 1.5 Å s −1 . It is noted here that when evaporating sensitive organic components, namely FAI or MAI, there is a “stickingcoefficient”, which means that the rate detected via the quartz crystal monitor (QCM) is much lower than what actually ends up on the substrate surface. Thus, it has been reported by us and others that excess FAI is needed to approach stoichiometric precursor ratios in the deposited lms. 60,63,64 We note that given the poor sticking characteristics of the organic source, FAI, it is difficult to accurately quantify the ratio of the FAI to PbI 2 in the nal thermally co-evaporated lm, thus only the relative rates (in the Å s −1 of the FAI with the PbI 2 xed at 0.6 Å s −1 ) are reported here. Fig. 1b shows the XRD patterns of the various lms deposited on glass/indium tin oxide (ITO)/(2-(3,6-dimethoxy9H-carbazol-9-yl)ethyl)phosphonic acid (MeO-2PACz) substrates (representing the bottom layers of the solar cells) followed by thermal annealing on a hotplate at 150 °C for 20 min in a N 2 Fig. 1 Thermally co-evaporated FAPbI 3 absorber and solar cell characterisation. (a) Schematic of the FAPbI 3 thermal co-evaporation process. (b) XRD patterns of various FAPbI 3 films deposited on glass/ITO/MeO-2PACz. Note, the ratio between the FAI : PbI 2 is represented by the evaporation rate of the FAI source, where the PbI 2 source is fixed at 0.6 Å s −1 , (c) PDS of FAPbI 3 with different ratios deposited on fused silica substrates (inset: Urbach energy values), (d and g) PSC photovoltaic data (based on 2 batches of devices with a minimum of 3 devices (8 cells each) per rate) of the devices with the structure glass/ITO/MeO-2PAC/FAPbI 3 /C60/BCP/Cu, where (d) V OC , (e) J SC , (f) FF, (g) PCE, (h) J–Vstability of four representative devices, (i) EQE with integrated EQE-J SC of four representative devices. © 2025 The Author(s). Published by the Royal Society of Chemistry EES Sol.,2025,1,41–55 | 43 Paper EES Solar Open Access Article. Published on 21 January 2025. Downloaded on 8/12/2025 11:56:44 AM. This article is licensed under a Creative Commons Attribution 3.0 Unported Licence. View Article Online glovebox environment. The three peaks of interest are the black a-FAPbI 3 (100) phase at 14.2°, the PbI 2 phase at 12.9°, and an unwanted yellow non-perovskite d-FAPbI 3 phase at 11.8°. We can see that at an FAI rate of 0.6 Å s −1 there is a large PbI 2 peak as well as characteristics of both d-FAPbI 3 and a-FAPbI 3 phases. For the FAI rate at 0.9 Å s −1 , the PbI 2 peak decreases in intensity together with the d-FAPbI 3 , whilst the a-FAPbI 3 peak increases. The same trend continues as we increase the rate to 1.1 Å s −1 ,as the PbI 2 peak disappears and only the a-FAPbI 3 peak is observable. From 1.2 Å s −1 the (100) peak then starts to decrease until 1.5 Å s −1 where there is very little detectable perovskiterelated peak in the lm, with the lm acquiring a visible orange shade (ESI Fig. 5†). Photothermal deection spectroscopy (PDS) measurements presented in Fig. 1c allowed us to calculate Urbach energy, 65 a proxy for examining energetic disorder, which showed a decreasing trend from 0.9 (18.4 meV) to 1.5 Å s −1 (17.2 meV) with increasing rate of FAI, whereas the bandgap remained the same at 1.53 eV (Urbach energy ts and Tauc plots can be found in ESI Fig. 6†). From the top-view scanning electron microscopy (SEM) images in ESI Fig. 7d–f,†the morphological grain size increased from 170, 180, to 280 nm with increasing FAI from 0.9, 1.2 to 1.5 Å s −1 respectively. This was in line with a slight increase in thickness (0.9, 1.2, and 1.5 Å s −1 were 490, 512, and 552 nm, respectively) as seen from prolometry measurements (ESI Fig. 8†). Note, the thicknesses above vary because the total time for each run was set to be the same, dictated by the xed PbI 2 rate of 0.6 Å s −1 and xed time, and thicknesses could be made to match if the total run time was varied. We note that evaporated FAPbI 3 tends to have smaller grain sizes between 100 and 200 nm compared to solutionprocessed analogues (typically >200 nm). 66 Single-junction FAPbI 3 perovskite solar cells with the structure ITO/MeO-2PACz/FAPbI 3 /C60/BCP/Cu were fabricated and measured under 1-Sun conditions and AM1.5 (see Methods). Fig. 1d–g shows the J–Vparameters for devices with the FAI rates of 0.6, 0.9, 1.2, and 1.5 Å s −1 . As seen in Fig. 1d, a clear drop in open-circuit voltage (V OC ) is observed as the FAI : PbI 2 ratio increases, with the V OC decreasing from an average of 1.08 V at 0.6 Å s −1 to 0.48 V at 1.5 Å s −1 . In contrast, Fig. 1b shows that the highest J SC values are achieved between 0.9 and 1.2 Å s −1 , averaging around 18.8 mA cm −2 (0.9 Å s −1 ) and peaking at 20.9 mA cm −2 (1.2 Å s −1 ). We also performed a thickness comparison to show that there is no signicant drop in J SC over the thickness range of interest for our MJ solar cells (around 500–700 nm) and we only see a drop in J SC at very high thicknesses (1750 nm) (see ESI Fig. 9†). Returning to Fig. 1d–g, it was clear that an optimum solar cell performance can be found close to 0.9 Å s −1 , with an average efficiency of 12.7%. The general trend was that near 0.9 Ås −1 , a slight decrease in FAI to PbI 2 ratio leads to an increase in V OC , whereas a slight increase in FAI to PbI 2 ratio leads to higher J SC . It is well established that a slight excess in PbI 2 suppresses nonradiative charge carrier recombination for solution-based PSCs and leads to performance gains (especially in V OC ) over stoichiometric compositions, 67–69 as we also observe here with thermally co-evaporated samples. However, at some point for the very PbI 2 rich samples (for example FAI <0.9 Å s −1 in this case), the non-perovskite d-FAPbI 3 phases increases, as we see in the XRD (Fig. 1b), which leads to performance losses. 70,71 In addition, excess residual PbI 2 , in the form of amorphous phases that may not be detectable via XRD, can cause other photo-stability issues especially under illumination. 72–76 On the other hand, for the very PbI 2 decient samples, that is, the FAI rich in this case (FAI >1.2 Å s −1 ), the J SC drop could be attributed to the accumulation of organic species at grain boundaries which hinders charge carrier mobility and/or likely carrier injection into the charge transport layers. 69 The average series (R S ) and shunt resistance (R SH ) for each rate are as follows: 0.6 Å s −1 (R S =15.8 Ucm 2 ,R SH =543.8 Ucm 2 ), 0.9 Å s −1 (R S =11.8 Ucm 2 ,R SH =503.4 Ucm 2 ), 1.2 Å s −1 (R S =24.1 Ucm 2 , R SH =59.5 Ucm 2 ), and 1.5 Å s −1 (R S =88.2 Ucm 2 ,R SH =97.0 U cm 2 ) (see ESI Table 2†for statistical analysis). The series resistance reaches an optimum (minimum) at 0.9 Å s −1 FAI rate, whereas shunt resistance continues to decrease in general with increasing FAI rate. Overall, these ndings, together with our previous work on co-evaporated perovskites, 13 show that a slight excess of PbI 2 is important in order to achieve sufficient performance and phase stability, which is in turn consistent with other studies on thermally co-evaporated perovskites. 52,77,78 Devices show stabilized power output over the course of 300 seconds as shown in Fig. 1h. In particular, the 0.9 Å s −1 device is the fastest to reach a stabilized maximum power compared to the other devices tested here. For voltage and current-density tracking, see ESI Fig. 10.†Fig. 1i shows the EQE of representative devices which match well to the J SC measured in the J–V. The EQE integrated J SC (EQE-J SC ) trend in Fig. 1i, follows closely to that of the J SC seen in the J–Vscans (Fig. 1e). A maximum EQE-J SC value is reached at the 1.2 Å s −1 sample with a value of 22.1 mA cm −2 , which is the typical range for a standard solution-processed FAPbI 3 solar cell. 79 Given all the properties shown, from this point forward, we utilise the absorber with an FAI rate of 0.9 Å s −1 (relative to PbI 2 rate of 0.6 Å s −1 ), which can achieve single-junction device efficiency exceeding 14%, which is competitive for an all-vapour FAPbI 3 absorber without additives 59 and thus as a suitable bandgap perovskite absorber for the middle subcell in our PPP triple-junction solar cell conguration. Monolithic PPP triple-junction solar cells –optical modelling vs. real devices We performed optical simulations based on the transfer matrix method (TMM) of our proposed triple-junction device stack shown in Fig. 2a using complex refractive index data acquired from ellipsometry (see ESI Note 3 and Fig. 11†). For the top subcell we used a co-evaporated Cs 0.3 FA 0.7 Pb(I 0.56 Br 0.44 ) 3 perovskite absorber with a bandgap of 1.80 eV from our previous work, 13 which is close to the widest possible with minimal photoinduced halide phase segregation. For the middle subcell, we used the co-evaporated FAPbI 3 perovskite absorber (1.53 eV) as demonstrated above. Finally, for the bottom subcell, we used our solution-processed Cs 0.25 FA 0.75 - Pb 0.5 Sn 0.5 I 3 perovskite absorber (1.25 eV). 13 For reference we show the single-junction equivalents of each type of perovskite absorber in ESI Fig. 12,†which provides a guide for 44 |EES Sol.,2025,1,41–55 © 2025 The Author(s). Published by the Royal Society of Chemistry EES Solar Paper Open Access Article. Published on 21 January 2025. Downloaded on 8/12/2025 11:56:44 AM. This article is licensed under a Creative Commons Attribution 3.0 Unported Licence. View Article Online performance and absorption onset. For the remaining interlayers of the stack such as the glass, ITO, fullerene C60, atomic layer deposited tin oxide (ALD-SnO x ) and spin-coated graphene oxide nanoparticles (GO) we used data acquired either from public databases or measured in-house by ellipsometry (see Methods). Note, the GO layer, which we report elsewhere, 80 is an Fig. 2 Demonstration of monolithic PPP triple-junction solar cells together with optical modelling analysis. (a) Schematic diagram of the entire device stack, (b) optically modelled J SC heatmap of the top and middle subcell perovskite absorber thickness variation based on the stack shown in (a). Note that the optical modelling range for the top and middle perovskite absorber layers are 0–400 nm and 400–800 nm, respectively, with afixed bottom layer thickness of 800 nm, (c) corresponding high-resolution cross-sectional SEM image, (d) EQE showing each of the three subcells including the integrated EQE-J SC and optically modelled EQE with the same subcell absorber thicknesses. In addition, the optically modelled EQE for the maximum point (180 nm, 580 nm) in the heatmap in (c) is shown in ESI Fig. 13b,†(e) photovoltaic characteristics (V OC ,J SC , FF, and PCE) for the batch, and f. Reverse J–Vof the champion device (inset: 100 hours stability measurement). Note, the photovoltaic data consisted of seven substrates (8 cells each). © 2025 The Author(s). Published by the Royal Society of Chemistry EES Sol.,2025,1,41–55 | 45 Paper EES Solar Open Access Article. Published on 21 January 2025. Downloaded on 8/12/2025 11:56:44 AM. This article is licensed under a Creative Commons Attribution 3.0 Unported Licence. View Article Online alternative to the typical ITO or more parasitically absorbing thin Au interconnection layers typically used in MJ perovskitebased solar cells. We rst performed TMM optical simulations to optimize the thicknesses of the co-evaporated top (0–400 nm) and co-evaporated middle (400–800 nm) perovskite absorber layers. The resulting J SC heatmap, shown in Fig. 2b, plots the top absorber layer thickness (y-axis) against the middle absorber layer thickness (x-axis), with the solution-processed Pb–Sn bottom absorber xed at 800 nm. Note, the Pb–Sn bottom absorber should be as thick as possible in terms of current collection in all-perovskite MJs, provided that sufficient charge transport can be maintained. Most work on Pb–Sn PSCs have demonstrated absorber thicknesses of less than 1 mm, 10,13,81–85 likely due to the short carrier diffusion length and/ or difficulty in achieving high quality lms via solution-processing given solvent solubility limits and spin-coating kinetics. Lin et al. 86 demonstrated thickness up to 1.2 mm in ammoniumcation-passivated Pb–Sn perovskites without loss in charge transport, thus achieving higher short-circuit current and record perovskite–perovskite tandem solar cell efficiency. The simulation results in Fig. 2b show that the maximum J SC occurs when the top and middle absorbers, using our chosen compositions, are 180 and 580 nm, respectively. These insights provide valuable guidance for selecting the optimal layer thicknesses in the monolithic PPP triple-junction solar cells. We fabricate a monolithic PPP triple-junction perovskite solar cell based on the stack represented in Fig. 2a with a corresponding high-resolution cross-sectional SEM image shown in Fig. 2c. In terms of the absorber thicknesses, the top thermally co-evaporated perovskite absorber, Cs 0.3 FA 0.7 Pb(I 0.56 - Br 0.44 ) 3 , was 170 nm (blue shade), the thermally co-evaporated middle absorber, FAPbI 3 , was 670 nm (green shade), and nally bottom subcell solution-processed perovskite, Cs 0.15 FA 0.85 - Pb 0.5 Sn 0.5 I 3, was approximately 800 nm (red shade). From Fig. 2c, it was observed that the top and the middle subcells were very conformal driven by the thermal co-evaporation process. Referring back to Fig. 2b, the top and middle absorber thicknesses of 170 and 670 nm respectively places the efficiency reference point within the optimum current-matching region. The EQE of our champion PPP triple-junction solar cell is shown in Fig. 2d, where we can observe that the top, middle, and bottom subcells had an integrated EQE-J SC of 9.6, 9.3, and 9.0 mA cm −2 , respectively. Given the sensitivity of MJ devices with the illumination spectrum, we provide the spectra used in our class AAA J–Vsolar simulator and the EQE setups in ESI Fig. 14a and b.†As detailed in ESI Note 1,†the overall J SC of a tandem or MJ device is constrained by the subcell with the lowest photocurrent. In our case, the limiting factor is the bottom Pb–Sn subcell, which operates at 9.0 mA cm −2 . Note, there is reasonable agreement, although a slight difference, between the J SC measured via J–V versus the J SC obtained from integrated EQE measurements (EQE-J SC ). 87,88 We specically highlight the remarkably small current variation between the subcells, with EQE-J SC values 9.0, 9.3, and 9.6 mA cm −2 showing an absolute spread of less than 4% from the champion J SC value of 9.3 mA cm −2 (6.7% between the highest and lowest EQE-J SC values). In contrast, previous reports on PPP triple-junction solar cells with all solution processed perovskite layers (ESI Fig. 2†) have exhibited larger spreads between their highest and lowest integrated EQE-J SC values of their individual subcells ranging from the largest 88.7% (ref. 29) to smallest 9.3% (ref. 32) (note, within working batches there are nevertheless some variation). Comparatively, the best III–V triple-junction solar cell made via metalorganic vapor phase epitaxy (another highly controllable vacuum deposition technique) has an EQE-J SC spread of only 4.5%. 43 The dotted lines in Fig. 2d show our optically modelled EQE, which closely aligns with the EQE of the experimental device, when using the same absorber thicknesses. This further conrms the accuracy of the thermal coevaporation technique. Indeed, the EQE spectral behaviour and integrated EQE-J SC of the actual experimental device matches well with that of the optical modelling, especially in the top (9.6 vs. 9.5 mA cm −2 ) and bottom (9.0 vs. 9.1 mA cm −2 ) subcells, respectively. However, for the middle subcell, there is a slight overestimation of the integrated EQE-J SC in the optical model (10.4 mA cm −2 ) compared to the experimental device (9.3 mA cm −2 ). The fact that the modelled EQE is shied upwards slightly is likely because in real devices there are effects such as charge generation/dissociation, non-radiative recombination, and extraction losses that are not accounted for in the optical model. Additionally, while the EQE-J SC values of the bottom subcell are similar for both the experimental and modelled, the experimental EQE curves are smoother than the optically modelled EQE curves, where the latter exhibit more interference fringe patterns. This is likely due to the idealized smooth layers in the optical model, where interference is more pronounced. In practice, layer thickness variations and surface roughness causes increased scattering of light at more oblique angles or a more Lambertian scattering process in the experimental device vs. the optically modelled scenario. Hence, we emphasise that in thermally co-evaporated layers (top and middle), the features in the EQE curves in Fig. 2d match well, which is actually an argument to say that the co-evaporated layers are of better optical quality, with less thickness roughness and more accurate deposition than the solution-processed layer (bottom). In addition, we show in ESI Fig. 13a and b†the optically modelled device stack and EQE (when the top and middle absorbers are 180 and 580 nm thick, respectively) highlighting that the maximum J SC for current matching is 9.7 mA cm −2 .We note that, in order to develop the nal monolithic PPP triplejunction solar cell structure, we performed a number of iterative steps with the development of suitable charge transport and interconnection layers (ESI Fig. 15–18 and Table 3†). The J–Vcharacteristics of the devices are presented in Fig. 2e. The average PV parameters across the batch were V OC =2.34 V, J SC =8.45 mA cm −2 ,FF=69.84%, and PCE =13.82% (reverse). The champion device achieved V OC =2.38 V, J SC =9.27 mA cm −2 ,FF=71.52%, and PCE =15.77% in reserve scan (forward: V OC =2.30 V, J SC =9.27 mA cm −2 ,FF=66.58%, and PCE = 14.22%). Notably, no passivation layers were applied to any of the three subcells. Previous work has demonstrated that passivation could signicantly improve V OC and overall efficiency by reducing surface recombination, 13,89–92 and efforts along these lines in future work will further increase V OC .We 46 |EES Sol.,2025,1,41–55 © 2025 The Author(s). Published by the Royal Society of Chemistry EES Solar Paper Open Access Article. Published on 21 January 2025. Downloaded on 8/12/2025 11:56:44 AM. This article is licensed under a Creative Commons Attribution 3.0 Unported Licence. View Article Online also estimate the series and shunt resistance of our champion device which were approximately 28.9 and 1.9 ×10 3 Ucm 2 , respectively. Additionally, the ALD-SnO x /GO recombination junctions at the top–middle and middle–bottom interfaces were not fully optimized in this instance, possibly contributing to shunt pathways. The champion device achieved a J SC of 9.27 mA cm −2 , one of the highest values reported amongst monolithic PPP triplejunction solar cells in literature (ESI Fig. 3b†). Moreover, as previously mentioned, it exhibits the smallest variation in the integrated EQE-J SC values across the individual top, middle, and bottom subcells (ESI Fig. 2b†), meaning that the currentmismatch is minimized. This is attributed to the use of the more transparent ALD-SnO x /GO interlayers 80 than the standard ALD-SnO x /Au and, crucially, the precise thickness control enabled by the thermal co-evaporation process. Furthermore, our EQE prole has coincidentally led to the bottom subcell with the lowest current at 9.0 mA cm −2 , and, as Boccard and Ballif 93 pointed out, such bottom-cell-limitation appears to be the most ideal in terms of maintaining high FF values in most realistic situations. As a result, the high Fill Factor (FF) we achieve for our champion device was 71.52%, and the highest FF we achieved in this batch was 75.53%, which is reasonably competitive with existing triple-junction PPP solar cells (highest at FF =81% (ref. 30)) (see ESI Fig. 3c†). This could also be due to the more conformal thermal co-evaporation process for the top and middle perovskite absorber layers. The J–Vcurve and parameters for the champion PPP triple-junction solar cell are presented in Fig. 2f, along with an inset showing the results of a 100 hours encapsulated max power stability test conducted in air with the temperature of the device held at 25 °C using a thermoelectric temperature-controlled stage (see ESI Fig. 19†). The device PCE shows an initial PCE gain (typical for p–i–n architectures 94 ) up to around 60 hours and then the gradual slow decrease. At 100 hours the device was still operating above its initial starting PCE. We believe this excellent stability result is due to a combination of our solution-free thermal co-evaporation process as well as our ALD-SnO x protective layers, especially the nal layer just before the Cu contact. Thus, we have demonstrated that thermal co-evaporation can be highly versatile and thickness-controllable for MJ congurations. Further optimisation opportunities for monolithic PPP triplejunction solar cells Finally, we use optical modelling to identify further areas for performance gain (Fig. 3a) based on our three perovskite absorber layers. We rst added a 100 nm MgF 2 antireection coating (ARC) at the front side of the glass substrate to improve light coupling into the active layers of our device. Secondly, we doubled the Pb–Sn bottom subcell thickness to 1600 nm to improve the long-wavelength absorption. Fig. 3b shows the results of this simulation in terms of a thickness optimisation heatmap for the top and middle perovskite absorbers of the stack shown in Fig. 3a. Fig. 3c shows the optimum EQE (corresponding to the maximum point in Fig. 3b where the top and bottom absorber thicknesses are 190 and 660 nm respectively) with current matching now slightly improved to 10.0 mA cm −2 . With the MgF 2 ARC and thicker Pb–Sn bottom subcell, the total current available in the new simulation increased from 29.1 mA cm −2 (sum of EQE-J SC values in ESI Fig. 14b†) to 30.2 mA cm −2 (sum of EQE-J SC values in Fig. 3c). This in turn has shied the optimum thicknesses of the top and middle absorber pairs from 180 and 580 nm (Fig. 2b) to 190 to 660 nm (Fig. 3c), respectively. We further improved this device stack by optimizing the thicknesses of the interlayers. As shown in Fig. 3d, we established realistic thickness parameters for each interlayer, including MgF 2 , ITO, C60, ALD-SnO x , and our three perovskite absorber layers. The aim was to minimize parasitic absorption and improve light coupling between the interlayers to optimize optical transparency to the perovskite absorbers. Furthermore, we sweep through suitable bandgap ranges for the top, middle, and bottom perovskites to nd the optimal bandgap combination. The optimal thicknesses and bandgaps of each perovskite layer as well as the thicknesses for each interlayer modelled are highlighted in bold on the right side of Fig. 3d. The top, middle, and bottom perovskite layers had optimum bandgaps of 1.87, 1.53 and 1.20 eV and thicknesses of 510.6, 1064.7, and 1488.8 nm, respectively (see ESI Fig. 20†for the simulated complex refractive index data). As a result of this optimization, Fig. 3e shows the optimal EQE based on these tuned layer thicknesses. Here we observe an improved J SC matching of 11.4 mA cm −2 (top subcell limited), representing a signicant improvement over our earlier stacks (ESI Fig. 14†with 9.7 mA cm −2 and Fig. 3c with 10.0 mA cm −2 ). This improvement is rstly due to the increased total absorption range with a lower bottom subcell bandgap of 1.20 eV and secondly, largely due to the signicant reduction in the parasitic absorption of the interlayers, especially in the C60 and ITO. It should be noted here that this result comes from an optical model that does not account for changes in material properties with thickness. For example, in this model, we assume that reducing the ITO thickness from 160 nm to 65.3 nm will maintain comparable sheet resistance. In reality, this could be difficult given the relationship between resistivity, transmittance, doping, and lm thickness in transparent conducting oxides 95,96 what may make this challenging. The enhanced light management and improved light absorption and coupling within the device itself also play a crucial role in the performance gains. In Fig. 3e we can see the various absorption components and losses. The glass absorbs more highly in the shorter wavelengths typically below 350 nm and then again towards the red and infrared regions. The ITO absorption losses also mainly occur at the shorter wavelength region, whereas the interlayer losses have a more even distribution in terms of the wavelength range. The perovskite layers contribute to the bulk of the absorption resulting in a total current density of 34.7 mA cm −2 . Finally, the inevitable reection losses in this case are minimized through optimizing for the thicknesses of the various interlayers, especially the front ARC. Note, we only show one ARC layer here, although more than one ARC 97 can be used, however this comes with diminishing returns. © 2025 The Author(s). Published by the Royal Society of Chemistry EES Sol.,2025,1,41–55 | 47 Paper EES Solar Open Access Article. Published on 21 January 2025. Downloaded on 8/12/2025 11:56:44 AM. This article is licensed under a Creative Commons Attribution 3.0 Unported Licence. View Article Online By allowing bandgaps and thicknesses of the perovskite absorbers to vary more widely together with the careful tuning of the thickness of the other layers, the practical efficiency limit for this new stack is thus much higher. We run a simple simulation based on the modelling to assess this practical efficiency limit. A simulated J–Vcurve including those of the subcells are shown in Fig. 3e. Here we see that the pseudo-V OC that can be achieved by the top, middle, and bottom subcells in the stack are 1.55, 1.19, and 0.89 V, respectively. Together with aJ SC matching of 11.4 mA cm −2 based on our perovskite absorbers with bandgaps (1.87, 1.53, and 1.20 eV) and estimated FF of 0.91 (see methods) we can therefore estimate an optimistic, but close to achievable, device efficiency of ∼37.6%. The improvements that have to be made versus existing monolithic PPP triple-junction solar cells include: (1) improving the V OC of the high-bandgap perovskite materials without suffering from the effects of halide segregation, 32,36,38,40,98 (2) increasing the bottom subcell Pb–Sn perovskite absorber thickness without loss in current extraction (where a thermally co-evaporated Pb–Sn perovskite could be made as thick as required without precursor concentration limitations 99 ), (3) nd intermediate layers that are thinner and more transparent, yet equally functional, and nally (4) better optoelectronic simulations to assist in fabrication of real devices. Conclusion In this work, we rst demonstrated the successful fabrication of thermally co-evaporated FAPbI 3 perovskite lms and solar cells using 2-sources, FAI and PbI 2 . This technique is especially useful compared to conventional solution processing, as it allows conformal deposition that is solution-free with excellent thickness control. We show that the ratio of the evaporation rates between the FAI and PbI 2 has a large effect on the properties of the nal FAPbI 3 perovskite. Specically, we found that a PbI 2 rich composition, which in our case means a lower FAI (0.9 Å s −1 ) to PbI 2 (0.6 Å s −1 ) evaporation rate, resulted in the most efficient working devices overall with higher V OC and FF. However, a slightly more stoichiometric FAI to PbI 2 ratio, where the FAI and PbI 2 rates were 1.2 Å s −1 to PbI 2 0.6 Å s −1 , respectively, yielded better J SC at the expense of lower V OC and FF. We then successfully demonstrated a working 2-terminal monolithic PPP triple-junction solar cell with V OC of 2.38 V, J SC of 9.27 mA cm −2 , FF of 71.52%, PCE of 15.77% using thermal coFig. 3 Further optimisation parameters for the demonstrated monolithic PPP triple-junction solar cell. (a) Schematic diagram of the optimal device stack from optical modelling with the addition of a front MgF 2 anti-reflection coating and increased bottom subcell absorber thickness, (b) optically modelled J SC heatmap of the top and middle subcell perovskite absorber thickness variation based on the stack shown in (a). Note that the optical modelling range for the top and middle perovskite absorber layers are 0–400 nm and 400–800 nm, respectively, (c) optically modelled EQE from the maximum point shown in (b). Showing each of the three subcells including the integrated EQE-J SC and total EQE-J SC , (d) further optimized device stack with a thickness sweep of all active layers with the ranges shown inside the stack and the optimum values on the right in bold, (e) optically modelled EQE from the stack shown in (d). Showing each of the three subcells including the integrated EQE-J SC and breakdown of the absorption and losses, and (f) optoelectronically simulated J–Vcurves of the monolithic PPP triple-junction solar cell including each of the subcell contributions, where the J SC is based on the values modelled in (e). The simulated PV parameters of the full triple-junction device are also shown. 48 |EES Sol.,2025,1,41–55 © 2025 The Author(s). Published by the Royal Society of Chemistry EES Solar Paper Open Access Article. Published on 21 January 2025. Downloaded on 8/12/2025 11:56:44 AM. This article is licensed under a Creative Commons Attribution 3.0 Unported Licence. View Article Online evaporation for the deposition of the 4-source Cs 0.3 FA 0.7 - Pb(I 0.56 Br 0.44 ) 3 top cell and our optimized 2-source FAPbI 3 middle cell absorbers. The integrated EQE-J SC for the top, middle, and bottom subcells were 9.6, 9.3, and 9.0 mA cm −2 respectively, which is one of the most effectively currentmatched for monolithic triple-junction solar cells in literature. These results are owing to our effective optical simulations and the excellent versatility of the thermal co-evaporation process, whereby the bandgap and thickness of the perovskite absorber can be controlled effectively. Furthermore, because the thermal co-evaporation technique is solution-free and in vacuum, subsequent layer deposition of the perovskite absorber avoids damaging, or at least, reduces the damage to underlying layers. Our work opens further doors to the vacuum-deposition thermal co-evaporation technique for perovskite research and commercialisation. Methods Materials All materials were used as received without further purication. FAI (99.99%) was purchased from Greatcell Solar Materials (Australia). MeO-2PACz (>98.0%), 2PACz (98.0%), PbI 2 (99.99% trace metal basis) and PbBr 2 (99.99% trace metal basis) were purchased from Tokyo Chemical Industry Co. (Japan). Cesium bromide (99.999% trace metals basis), tin iodide (beads, 99.99%), tin ouride (99%), cesium iodide (99.999% trace metals basis), N,N-dimethylformamide (anhydrous, 99.8%), dimethyl sulfoxide (anhydrous >99.9%), PTAA, toluene (anhydrous, 99.8%), ethanol (anhydrous, 99.5%) were purchased from Sigma-Aldrich (Global). BCP (>99.5% sublimed) and PEDOT : PSS was purchased from Ossila. C60 (99.99%) was purchased from Creaphys (subsidiary of MBraun) GmBH (Germany). Single-layer graphene oxide dispersion in water was purchased from Graphene Supermarket (USA). Patterned ITO (10–15 ohm per sq.) 25.4 ×25.4 ×1.1 mm substrates were purchased from Kintec (Hong Kong). TDMASn precursor was purchased from Strem Chemicals (USA). Device fabrication Substrate preparation. ITO substrates (Kintec) were cleaned in a sonication bath in Hellmanex solution, DI water, acetone and isopropanol (15 min each). The cleaned substrates were transferred to a UV-ozone chamber (UVC1014, NanoBioAnalytics) for another 15 minutes post-treatment. Wide-bandgap top perovskite and solar cell fabrication. 120 ml of MeO-2PACz (0.4 mg ml −1 ) in anhydrous ethanol was dropped on the cleaned ITO substrate and spincoated at 4000 r.p.m. for 30 seconds in a glovebox with an integrated spincoater followed by post-annealing at 100 °C for 10 min. The MeO-2PACz substrates were transferred to a PEROevap (CreaPhys/Mbraun) chamber inside a N 2 -lled glovebox for perovskite evaporation. The chamber was pumped down to below 2.0 ×10 −6 mbar. During the evaporation, the substrates stage was kept at 18 °C temperature, while the chamber walls were kept at −15 °C temperature. The precursor deposition rates for the four sources were 0.9 Å s −1 for FAI, 0.6 Å s −1 for PbI 2 , 0.1 Å s −1 for PbBr 2 , and 0.1 Å s −1 for CsBr. The PbI 2 powders were placed in an alumina crucible whereas the FAI powder, being a more sensitive organic molecule, was placed in a crucible of a special “ultra-low temperature”source xture designed for controlled heating. Aer the perovskite evaporation, the samples were post annealed at 150 °C or 170 °C for 20 minutes to form the Cs 0.3 - FA 0.7 Pb(I 0.56 Br 0.44 ) 3 . To complete the single-junction perovskite solar cells, the samples were transferred back into the perovskite evaporator for C60 (25 nm), BCP (8 nm) and Cu (120 nm) deposition. Mid-bandgap middle cell perovskite and single-junction solar cell fabrication. 120 ml of MeO-2PACz (0.4 mg ml −1 )in anhydrous ethanol was dropped on the cleaned ITO substrate and spincoated at 4000 r.p.m. for 30 seconds in a glovebox with integrated spin-coater, followed by post-annealing at 100 °C for 10 min. The MeO-2PACz substrates were transferred to a PEROevap (CreaPhys/Mbraun) chamber inside a N 2 -lled glovebox for perovskite evaporation. The chamber was pumped down to below 2.0 ×10 −6 mbar. During the evaporation, the substrates stage was kept at 18 °C temperature, while the chamber wall was at −15 °C temperature. The precursor deposition rates for the two sources were 0.9 Å s −1 for FAI and 0.6 for PbI 2 . The PbI 2 powders were placed in an alumina crucible whereas the FAI powder, being a more sensitive organic molecule, was placed in a crucible of a special “ultra-low temperature”source xture designed for controlled heating. Aer the perovskite evaporation, the samples were post annealed at 150 °C for 20 minutes to form the FAPbI 3 perovskite. To complete the single-junction perovskite solar cells, the samples were transferred back into the perovskite evaporator for C60 (25 nm), BCP (8 nm) and Cu (120 nm) deposition. Low-bandgap bottom perovskite and solar cell single-junction solar cell fabrication. 120 ml of 2PACz (0.3 mg ml −1 )in anhydrous ethanol was dropped on the cleaned ITO substrate and spincoated at 3000 r.p.m. for 30 seconds in a glovebox with integrated spin-coater, followed by post-annealing at 100 °C for 10 min. Anhydrous ethanol was then dropped on the lm and again spincoated at 3000 r.p.m. for 30 seconds to remove the excess 2PACz, followed by annealing at 100 °C for 2 minutes. To make the perovskite solution, the precursors were mixed in 1 × 4 ml vial in the following order, SnF 2 (0.1 M), SnI 2 (1 M), CsI (0.5 M), PbI 2 (1 M) and FAI (1.5 M) to achieve 2 M concentration in DMF : DMSO (4 : 1) for the composition of Cs 0.25 FA 0.75 Pb 0.5 - Sn 0.5 I 3 . The Pb–Sn perovskite solution was prepared in a N 2 - lled glovebox (H 2 O and O 2 below 1 ppm) and stirred for at least 3 hours before use. 120 ml perovskite solution was spread on the 2PACz/ITO substrate and spun at 5000 r.p.m. for 40 s for with N 2 gas quenching at 25 s into the process (some devices at 4000 rpm and 30 s N 2 gas quenching total). The samples were moved to a hotplate for post-annealing at 120 °C for up to 10 min. Monolithic triple-junction PPP solar cell fabrication. This method was based on the triple-junction shown in the main text. The conguration of the PPP triple-junction was Glass/ITO/ MeO-2PACz/1.83 eV Cs 0.3 FA 0.7 Pb(I 0.56 Br 0.44 ) 3 perovskite/C60/ SnO x /GO/MeO-2PACz/1.53 eV FAPbI 3 perovskite/C60/SnO x /GO/ 2PACz/1.25 eV Cs 0.25 FA 0.75 Pb 0.5 Sn 0.5 I 3 perovskite/C60/SnO x /Cu. © 2025 The Author(s). Published by the Royal Society of Chemistry EES Sol.,2025,1,41–55 | 49 Paper EES Solar Open Access Article. Published on 21 January 2025. Downloaded on 8/12/2025 11:56:44 AM. This article is licensed under a Creative Commons Attribution 3.0 Unported Licence. View Article Online