Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
Nature Communications volume 17, Article number: 4512 (2026) Cite this article
Despite rapid advances in perovskite solar cells, solvent selection remains a central determinant of safety, process robustness, and end-of-life outcomes. These constraints are multi-dimensional and involve competing trade-offs, making them challenging to resolve through experimental optimization alone. This Perspective integrates green solvent engineering with artificial intelligence (AI) and life cycle assessment (LCA) to provide a unified sustainability framework. We discuss solvent-precursor coordination and processing-window robustness as governing factors. We also highlight how AI can accelerate solvent discovery and reduce key life cycle inventory gaps, while LCA quantifies trade-offs and mitigates burden shifting. This combined lens clarifies sustainability-relevant priorities for the field.
Metal–halide perovskite solar cells (PSCs) have progressed rapidly over the past decade, achieving remarkable power-conversion efficiencies (PCEs) and demonstrating strong potential for scalable manufacturing¹. Their compatibility with low-temperature solution processing and physically versatile device forms has positioned PSC as a leading next-generation photovoltaic technology. However, environmental impact, occupational safety, and long-term material sustainability are still major challenges for practical deployment^2,3. These concerns mainly arise from the solvent-intensive process and the need of viable end-of-life strategies.
Solution-processed PSCs commonly use polar aprotic solvents, such as N,N-dimethylformamide (DMF), dimethyl sulfoxide (DMSO), and N-methyl-2-pyrrolidone (NMP) for preparing the perovskite precursor solution^4,5,6. These solvents are essential for dissolving lead halide salts and forming high quality film, but they are highly toxic raising significant environmental concerns. Achieving high efficiencies alone is no longer sufficient for PSC translation. Sustainability constraints, particularly solvent hazard, process energy demand, and end-of-life circularity, must be incorporated as core design criteria rather than treated as an afterthought. Solution processes, such as spin coating, blade coating, and inkjet printing are also distinguished by their solvent requirements and energy consumption, which directly affect the associated environmental burdens. To ensure the sustainability of PSCs, it is also important to further advance environmentally viable recycling routes. Most reported strategies rely on solution-based separation, where consume substantial amounts of organic solvents, indicating that solvent choice and management are critical for minimizing environmental impacts⁷.
As sustainable processing becomes more important, environmental assessment has become a key basis for selecting materials and processing methods. Life cycle assessment (LCA) provides a structured approach for quantifying the environmental impacts of PSC technologies⁸. Recently, LCA studies have identified the principal contributors to the environmental burden of PSCs^9,10, underscoring the necessity of improved materials and process design. Integrating LCA into early-stage PSC research helps identify major sources of environmental impact and prioritize lower-impact alternatives, especially as solvent substitution, process optimization, and recycling strategies continue to develop. Yet LCA is often limited by incomplete life cycle inventory (LCI) data for emerging formulations and lab-scale process conditions, making prospective and uncertainty-aware assessment essential for meaningful design guidance.
In addition to these analytical frameworks, data-driven approaches are providing new opportunities for predictive sustainability design. Artificial intelligence (AI) is emerging as a useful tool for both green solvent design and prospective environmental evaluation^11,12. Machine-learning models can rapidly screen multi-component solvent mixtures and predict their solubility and coordination behavior. They can also help identify solvent candidates that combine low hazard with favorable crystallization behavior. AI has also been proposed for integrating uncertainty analysis into LCA frameworks^13,14.
Together, these tools help shift PSC research away from trial-and-error experimentation and toward predictive, sustainability-oriented design (Fig. 1). AI accelerates the development of green solvent strategies by enabling data-driven screening and optimization of solvent systems and processing conditions. Experimental validation then confirms technical feasibility and provides feedback that refines subsequent AI-guided selection. In parallel, LCA quantifies the environmental implications of validated solvent strategies, thereby identifying hotspots and informing the selection of lower-impact processing options. In addition, AI can mitigate a major limitation of LCA, insufficient life cycle inventory (LCI) data, by predicting missing inventory parameters and enabling uncertainty-aware assessment. Collectively, these interactions provide a practical basis for aligning process feasibility with environmental performance in PSC development. In practical research, this integrated framework begins with AI-assisted prescreening of solvent candidates and solvent mixtures using descriptors related to precursor coordination, processability, and hazard. The shortlisted formulations are validated through experiments on precursor-solution stability, intermediate-phase evolution, film quality, and device performance. LCA is further employed to evaluate the validated systems and identify environmental hotspots and trade-offs associated with solvent use, energy demand, and end-of-life management. AI can also assist this stage by predicting missing LCI parameters for emerging solvent systems and process conditions, thereby improving prospective and uncertainty-aware environmental evaluation. The resulting experimental and environmental data are fed back into the model, allowing iterative refinement of solvent selection and process optimization.
Conceptual scheme of an integrated framework linking green solvent strategy, AI technology, and LCA for sustainable PSC development.
Here, we review recent progress in green solvent process, LCA, and AI related for sustainable PSC development. We first summarize advances in solvent systems for perovskite precursor inks and recycling routes, focusing on toxicity, processability, and compatibility. Then, we discuss AI research that screen green solvent candidates, guide mixed-solvent formulation, and support the design of recycling process. Finally, we analyze LCA studies that quantify the environmental impacts of PSCs from materials production to end-of-life management. Based on these aspects, this review suggests practical directions for advancing PSC technologies toward environmentally responsible scale-up and long-term viability.
Solution processing remains one of the most promising approaches for the fabrication of PSCs owing to its simplicity, cost-effectiveness, and compatibility with large-area deposition. Nevertheless, the continued reliance on hazardous solvents and precursor formulations raises concerns regarding long-term environmental sustainability, as shown in Fig. 2a. Accordingly, recent studies have placed increasing emphasis on developing green solvent engineering strategies across the entire device fabrication sequence, including perovskite precursor preparation, anti-solvent treatment, charge-transport-layer (CTL) deposition, and solvent recovery within recycling processes³. As green solvent engineering for PSCs advances, solvent selection should be evaluated not only by laboratory film quality but also by its viability under manufacturing-scale constraints and circular economy requirements. In practice, system-level sustainability outcomes depend on solvent consumption and recovery, compatibility with scalable coating methods, energy demand associated with drying and annealing, and the feasibility of solvent reuse across both fabrication and end-of-life separation steps^15,16,17,18. Cross-technology learning from established photovoltaic manufacturing and recycling infrastructure further highlights that process integration and end-of-life logistics can ultimately determine whether a proposed “green” solvent strategy is implementable at scale^19,20. Accordingly, the following sections discuss green solvent candidates with solvent management considerations and scale-relevant constraints, rather than treating solvent substitution as an isolated materials replacement problem¹⁷.
a Illustration of the toxic solvents used in PSC fabrication. b Reported PCE of PSC devices and hazardous scores of solvents used for perovskite layer. c Anti-solvents and PCE of PSC devices using them. d Scheme of PSC device fabricated with all green solvent system. a and b adapted from ref. ³. Copyright 2024, Wiley-VCH GmbH. c adapted from ref. ⁴¹. Copyright 2021, Springer Nature. d adapted from ref. ³⁶., Copyright 2021, Royal Society of Chemistry.
For the perovskite layer, the solvent system must effectively dissolve lead halide precursors while governing nucleation, intermediate formation, and film crystallization. Widely used solvents, such as DMF and NMP are now subject to regulatory pressure due to reproductive toxicity and occupational exposure concerns. As a response, alternative solvents, including DMSO, γ-valerolactone (GVL), 2-propanol (IPA), and ethanol, have been proposed as greener candidates (Fig. 2b)^21,22,23,24. Solvent selection guides, such as the CHEM21 have already been applied in structured decision workflows for perovskite processing, where EHS-based screening is combined with physicochemical criteria (e.g., solubility, coordination descriptors) and device and process feasibility to identify safer solvent systems for scalable fabrication^25,26,27. Recent reports show that GVL-based formulations not only yield uniform, high-crystallinity films but also significantly reduce environmental burdens, particularly for climate-change and human-toxicity indicators when quantified through LCA^23,28,29. Beyond GVL, alcohol and water based precursor routes have been investigated to reduce solvent hazards. Low-polarity alcohol, such as ethanol and IPA offers safer processing conditions and advantageous volatility. However, they inherently show poor solubility for lead halide salts, which makes them unsuitable as stand-alone solvents for perovskite precursors. Ethanol has been used for perovskite precursor preparation, but only when additional coordinating additives, such as N,N-dimethylacetamide and alkylammonium chlorides, are introduced to enable complexation between PbI₂ and propylammonium chloride²². IPA shows a similar limitation. Reported IPA-based processes do not use IPA alone. They rely on mixed-solvent systems in which water or another polar co-solvent system assists the dissolution of lead salts^21,24,30,31. In these formulations, IPA mainly adjusts solvent polarity, improves drying behavior, and supports more uniform crystallization. Pb halide salts, such as PbI₂ are insoluble in pristine water. Therefore, water-based perovskite processing commonly avoids direct Pb halide dissolution and instead employs soluble Pb salts, such as Pb(NO₃)₂ as the lead precursor. Zhai et al. demonstrated high efficiency water-processed PSCs using a Pb(NO₃)₂/H₂O precursor enabled light-modulation control of conversion/film formation³². More recently, Zhang et al. reported that sodium dodecyl sulfonate surfactant modulation can accelerate the Pb(NO₃)₂ to perovskite transformation and improve film quality in aqueous processed planar devices³³. Although alcohol- or water-assisted systems offer higher environmental compatibility, they still suffer from slow crystallization, incomplete conversion, and the need for elevated annealing temperatures³³. These limitations are consistent with the distinct precursor coordination environment in protic media, which can reduce crystallization control and limit device performance relative to conventional DMF/DMSO routes³⁴.
Ionic liquids (ILs), a class of room-temperature molten salts, provide a non-volatile and chemically stable medium suitable for perovskite precursor formulation^31,35,36,37. Protic ILs containing methylammonium (MA) cations exhibit strong coordination toward Pb²⁺ species, enabling effective dissolution of lead halide salts without relying on conventional hazardous polar aprotic solvents. However, the inherently high viscosity of many ILs can compromise wettability and hinder uniform film formation, often necessitating the use of co-solvents. When appropriately diluted with greener solvents, such as water, ethanol, acetonitrile (ACN), or IPA, IL-based precursor systems achieve improved fluidity and facilitate controlled intermediate formation, ultimately supporting perovskite crystallization under more environmentally compatible processing conditions³¹. This dilution strategy, however, introduces a practical trade-off between improved processability and the need to manage residual ionic species and drying dynamics to maintain film uniformity and stability. For example, protic IL-enabled inks have been demonstrated for water/alcohol-based precursor formulations compatible with scalable coating, highlighting viscosity/processability as a key design constraint even when solvent is reduced³⁸. Accordingly, IL-based processing should be assessed by coupled criteria including viscosity reduction strategy, residual species control, and the feasibility of solvent recovery/reuse, in addition to film quality and operational stability³⁹.
The choice of anti-solvent critically influences perovskite nucleation, intermediate formation, and final film morphology (Fig. 2c)^40,41. Green anti-solvents, such as methoxybenzene, ethyl acetate (EA), acetate derivatives (methyl acetate, propyl acetate, butyl acetate), and various ethers have been explored as alternatives to toxic chlorobenzene (CB) and toluene^42,43,44,45. These solvents can promote larger grain sizes, smoother films, and improved device performance, with PCEs exceeding 20% in several reports. A representative example is EA-based antisolvent processing, which has been used to form uniform perovskite films under ambient-air processing conditions, illustrating the practical motivation for greener substitution^44,45. However, challenges remain, including roughness induced by fast-evaporating EA, sensitivity of bisolvent systems to ambient conditions, and safety concerns for ethers like diisopropyl ether due to peroxide formation^25,46. Recent assessments of green anti-solvents further emphasize the trade-off between film quality and process robustness, where changes in volatility and miscibility can narrow the processing window and influence long-term environmental stability under manufacturing-relevant conditions⁴⁶.
The CTL plays a key role in PSC performance, yet conventional hole transport layer (HTL) processing typically relies on toxic nonpolar solvents, such as CB or toluene. Recent efforts have focused on replacing these solvents with greener alternatives. EA has been used both as an anti-solvent and as a spiro-OMeTAD solvent, yielding improved PCEs up to 19.43%⁴². Fig. 2d shows a fully green blade-coated process using water, methylammonium acetate (MAAc), and EA which delivers a PCE of 20.21%³⁶. Anisole, a less toxic solvent, also enabled uniform HTL deposition with performance comparable to CB⁴⁷. Additional green solvents, such as tetraethyl orthocarbonate further enhanced device efficiency in inverted architectures, demonstrating the growing potential of environmentally benign CTL processing routes⁴⁸.
Overall, green-solvent routes are intrinsically constrained by precursor solubility/coordination and narrow processing windows, whereas others are primarily engineering-limited and thus realistically scalable when coupled with robust coating compatibility and closed-loop solvent management (recovery/reuse) under manufacturing relevant conditions. At the process level, long-term feasibility also depends on precursor/ink stability, since solution speciation can evolve during storage and handling via coordination changes and solvent degradation, thereby affecting crystallization behavior and coating reproducibility^23,49,50.
Solvent-based separation is widely used in PSC recycling, making solvent selection a critical determinant of the overall environmental impact. However, many established solvent-based recycling processes rely on toxic organic solvents, which motivates both solvent substitution and tighter solvent management^51,52,53. Recent studies show that greener options, such as water or low-toxicity organic solvents, can substantially reduce chemical burdens during recovery processes^54,55. In mesoscopic carbon PSCs, GVL enabled selective removal of the perovskite absorber while preserving the printed mesoporous scaffold for reuse, with the remanufactured devices recovering up to 89% of the initial PCE⁵⁶. Notably, aqueous-based recycling has been demonstrated as a practical route to reduce reliance on hazardous organic solvents while enabling effective materials recovery from PSCs⁵⁵. Beyond solvent choice alone, recycling of solvents themselves is also essential for improving sustainability. Several reports demonstrate that solvents used in the recycling step can be purified and reused for subsequent PSC fabrication without significant performance loss^15,57. In addition, several studies report solvent-assisted routes that facilitate reuse of recovered components or precursors rather than single-pass recovery, supporting more circular process designs^55,57. Implementing such closed-loop solvent systems reduces waste generation and meaningfully lowers the life cycle environmental footprint of perovskite technologies. At manufacturing-relevant scales, however, the environmental outcome is often governed by process-level factors, such as solvent intensity per module area, recovery yield during distillation or purification, and the associated energy demand for separation, drying, and thermal treatments^17,54,55,58. In addition, realistic devices and modules impose system-level constraints arising from multi-layer stacks, encapsulation, and coated-glass components. These features introduce additional processing steps and can shift burdens toward auxiliary materials and energy unless carefully managed^59,60. Accordingly, PSC recycling should be evaluated not only by recovery efficiency, but also by the combined trade-offs among solvent management, energy requirements, and end-of-life process integration that ultimately determine practical circular implementation^17,54,60.
Traditional sustainability research in materials science is often limited by post-experimental evaluation, which can result in a lack of forward-looking perspectives. In studies on device recycling and green solvent design, researchers often select experimental materials based on experience or existing knowledge, using trial-and-error methods^23,61,62. In perovskite research, the selection of candidate solvents has typically focused on single solvents or simple mixtures, with limited research on multi-solvent mixtures. Multi-solvent mixtures have been shown to enhance solubility and regulate nucleation effectively. However, trial-and-error approaches are inadequate for studying complex, multi-component solvent systems. Trial-and-error is not effective for identifying global optimal solutions^63,64,65,66. This limitation happens to designing solvents and adsorbents for wet recycling of PSCs. Meanwhile, the systems of complex, multicomponent solvents and adsorbents demonstrate significant potential^67,68. Fortunately, AI shows great promise in identifying optimal solutions.
Presently, there exists a scarcity of published works concerning AI-assisted green solvent design for PSCs, suggesting substantial research potential in this domain. Furthermore, the two cases just introduced focus on experimental engineering applications and lack a physicochemical perspective on the discovery of novel green solvents. In contrast, studies on organic solar cells (OSCs) in this regard have been reported on multiple occasions. The dissolution of conjugated organic molecules in OSCs relies on intermolecular π–π interactions and polarity matching, whereas PSCs depend on Lewis acid-base interactions and coordination complexes between ionic inorganic salts and organic cations (PbI₂, FAI, MAI, etc.)^69,70. Despite differing design requirements, we can discern common AI application workflows and similar Hansen solubility matching logic; however, for PSCs these should be viewed as a transferable framework that still requires PSC-specific validation and descriptor adaptation due to coordination-driven solution chemistry.
Designing green solvents for OSC active layers requires considering the replacement of harmful halogenated solvents while ensuring the solubility of acceptors and donors. Mahmood and Wang used multiple AI models to predict Hansen solubility⁷¹. Random forests produced accurate predictions for this metric (r = 0.96). More than 3000 molecular descriptors were calculated and screened to capture structural, topological, and polarity-related information beyond simple frontier orbital energies. Based on these descriptors, multiple machine-learning models were developed for PCE classification/regression and HOMO/LUMO prediction, followed by a two-step virtual screening strategy in which candidate non-fullerene acceptors were first filtered by energy-level compatibility and then ranked by predicted PCE. They screened 87 green solvents from 252 candidates using high-throughput screening and Hansen solubility parameter (HSP) profiles as a filter. As Fig. 3b shows, these solvents effectively dissolve Poly(3-hexylthiophene) (P3HT) and five high-performance non-fullerene acceptors. Subsequent work expanded their research in descriptor collection and model application¹¹. They benchmarked more than 40 machine-learning algorithms and showed that models trained on molecular descriptors consistently outperformed those trained on molecular fingerprints. Ultimately, they recommended four green solvents for each of 30 small-molecule donors. This study also highlighted that chemically meaningful descriptors can effectively encode dispersion, dipolar, and hydrogen-bonding interactions that govern solution-processability. Their studies primarily demonstrate transferable AI workflows, comprising descriptor generating, surrogate modeling, high-throughput screening, experimental validation. For PSCs, the same workflow can be adopted, while the screening labels and descriptors must be reformulated to reflect coordination-driven solvation and crystallization kinetics rather than molecular solubility alone. Lee emphasizes the importance of explainable green solutions beyond the black-box mechanisms of AI⁷². He presents an interpretable AI framework for predicting the PCE of bulk heterojunction (BHJ) OSCs processed with non-halogenated green solvents. This framework uses descriptors with physical meanings, such as molecular weight and HSPs. The data come from 97 devices with a PCE range of 1.02% to 16.52%. The trained gradient boosting regression model has an R² value of 0.74 and an RMSE value of 2.09. Additionally, Lee used Shapley additive explanations (SHAP) to determine the influence of key descriptors on efficiency. High SHAP values were identified for donor molecular weight (MW_donor) and nonpolar dispersive interaction (δ_d), at 1.91 and 1.15, respectively. Physically, MW_donor influences the aggregation behavior of polymer donors in solution, thereby controlling surface morphology. Meanwhile, δ_d facilitates the evolution of the morphology. Further, a 2D contour map was plotted to show the PCE change with two significant descriptors. Finally, he predicted that the BHJ-based OSC with a 120-kDa MW_donor and a 17.3-MPa δ_d would bring an efficiency of 17.5%. The explainable-AI paradigm is transferable to PSC green solvent developments. Meanwhile, perovskite-relevant interpretable descriptors should emphasize solvent coordination strength, solvation environment, evaporation/rheology, and their impact on intermediate phases, defect formation, and film uniformity.
a Data preparation process that mixing DMSO, DMF with solvent candidates with a ratio of 0.6:0.32:0.02:0.02:0.04. The prepared precursors were deposited on the TiO₂ substrates for aqueous stability checking. b Schematic workflow of the determining the low-toxicity solvent process in the air condition by a two-step Bayesian method. c A gradient tree as a PCE prediction model, and SHAP as interpretable tool. d NLP pipeline encompassing data extraction, data identification, descriptor introduction, model training, and model implement. a adapted from ref. ⁷³. Copyright 2023, American Chemical Society. b adapted from ref. ⁷¹. Copyright 2025, Elsevier. c adapted from ref. ⁷⁴. Copyright 2021, The Royal Society of Chemistry. d adapted from ref. ¹². Copyright 2022, American Chemical Society.
Natural-language processing (NLP) offers a scalable pathway to green-solvent discovery for perovskite solar cells by converting the rapidly expanding, yet fragmented, solvent knowledge scattered across thousands of papers into machine-readable evidence. By systematically extracting solvent identities, processing contexts, and safety-related cues, NLP enables more transparent, traceable, and risk-aware solvent selection. When designing perovskite solvent molecules, it is essential to comprehensively consider processes, such as their reaction, exchange, and coordination with precursors. Multi-solvent mixtures have repeatedly demonstrated their ability to effectively stabilize the intermediate phase and form high-quality films. However, the diversity of solvents and the flexible adjustment of their proportions constitute an extremely complex virtual design space. Identifying optimal solvent compositions and ratios within the vast space presents a challenge. Huang et al.‘s research offers insights for designing multi-green solvents, though their focus is not on green⁷³. They investigated the aqueous stability of perovskite films prepared with pentameric solvents. In this work, DMF and DMSO are fixed candidates, and the solvent ratio is fixed as 0.6:0.32:0.02:0.02:0.04. By systematically replacing the other three solvents with 21 solvent molecules, they constructed a dataset comprising 59 samples (see Fig. 3a). In the created ‘stable/unstable’ classification task, the extra tree achieved a test receiver operating characteristic (ROC) score of 0.85, demonstrating the highest accuracy. This model was then used to predict the water stability of 6720 solvent additives at high throughput, yielding 1608 candidate solvents. Furthermore, statistical analysis emphasized the significance of hydroxyl groups, although the study did not examine solvent ratios. SHAP analysis showed that solvent geometric eccentricity, dipole moment, polarizability, and relative vapor pressure are closely related to stability. Subsequent DFT further demonstrated that this optimal multi-solvent system can stabilize the perovskite surface through O···Pb Lewis acid-base interactions, hydrogen bonding, and weak interactions between solvents, and can modulate the electronic structure without introducing deep-level defects. For green solvents, efforts should focus on replacing DMF and conducting in-depth exploration of ratio adjustments. Machine-learning can model intermediate constraints related to coordination chemistry and process manufacturability before guiding performance optimization. Ma et al. used AI to explore low-toxicity solvents for the fabrication of PSCs under air conditions⁷⁴. They designed a two-step Bayesian method that integrates precursor solubility prediction and device efficiency prediction, the process of which is shown in Fig. 3c. The solubility model corresponds to the baseline constraints of precursor-solvent coordination and solution stability, while the efficiency model corresponds to the comprehensive results of grain size, growth kinetics, and defect control after film formation. First, they used hypercube samples of solvent and additive amounts for solubility testing. The Bayesian optimization of this process narrowed down the range of solvent and additive amounts. Next, they introduced an annealing temperature and prepared PSC devices. The second Bayesian algorithm was adopted to find the optimal process combination: 580 μL of triethyl phosphate, 25% MACl, and annealing at 120 °C. Furthermore, SHAP analysis revealed that TEP and MACl were the most critical variables. A moderate amount of TEP determined the precursor concentration and crystal quality, while approximately 25% MACl facilitated the formation of the mesophase, delayed crystallization, and promoted the formation of large grains and uniform thin films. In the work of Giri et al., AI was employed to screen for uncertain information from perovskite solvent literatures, thereby identifying safe solvents¹². The workflow is illustrated in Fig. 3d. They categorized endocrine disrupting solvents as hazardous solvents. This study employed context-aware NLP to extract over 30,000 text segments concerning perovskite synthesis from thousands of research papers. Subsequently, 35 solvent molecules were further identified, and Simplified Molecular-Input Line-Entry System (SMILES) descriptors were assigned. The established convolutional neural network (CNN)+long short-term memory (LSTM) binary classification model achieves 90% accuracy. Together with uncertainty metrics, potential endocrine disruptors could be screened even under data-scarce conditions. In the future, more convincing machine learning frameworks should simultaneously integrate molecular structure descriptors, precursor coordination/dissolution capabilities, specific processing roles, and toxicological safety labels, thereby promote truly feasible green manufacturing.
Given the potential applications of perovskite in numerous optoelectronic fields, such as solar cells, photodetectors, lasers, and light-emitting diodes, the future recycling of this material will represent a substantial market. The directional recycling of each layer within PSC devices holds significant importance. Sprague et al. pioneered an AI-assisted proof-of-concept method for PSC recycling, establishing a unified protocol for applying sentiment analysis language processing techniques⁷⁵. A neural network trained using this protocol achieved 70% accuracy in predicting the optimal recycling strategy for valuable and harmful components within unknown devices. To prevent environmental factors, such as moisture, from affecting device performance, PSC modules must undergo encapsulation during industrialization. Due to its high technical maturity and excellent barrier properties, ethylene-vinyl acetate (EVA) is currently the most popular encapsulation material. However, effectively de-encapsulating EVA for recovery remains a significant concern. Lu et al. established regression and classification models for high-throughput screening of 70 organic reagents used in the wet de-encapsulating process of photovoltaic modules⁵⁹. Among these, a random forest regression model with an RMSE of 0.167 predicted 10 reagents possessing de-encapsulating capability. Furthermore, a support vector machine (SVM) classification model with a total reliability score (TRS) of 0.819 identified 23 agents possessing de-encapsulation capability. The findings indicate that upon disruption of C = O bonds within EVA’s cross-linked and branched structures, reagent molecules fully occupy corresponding sites, inducing finite swelling of EVA. Conversely, cleavage of C-C bonds in the main chain structure leads to infinite swelling of EVA.
LCA is a methodology for systematically evaluating the environmental impacts of a product, process, or activity across its entire life cycle. It considers every stage, resource extraction, production, transportation, use, maintenance, re-use, recycling, and final disposal, while quantifying energy and material inputs and outputs and assessing the associated environmental burdens to identify opportunities for improvement. In this way, LCA enables a comprehensive, cradle-to-grave understanding of the environmental performance of products or systems.
LCA proceeds through four phases. First, goal and scope definition clarifies the purpose of the study and sets the basic rules, including the choice of the functional unit and the specification of the system boundary. In LCA studies of perovskite photovoltaics, energy performance is often the priority, so kilowatt-hour (kWh) is typically chosen as the primary functional unit to facilitate comparisons across different material and process designs, as well as with other electricity-generating technologies, such as non-photovoltaic renewable energy systems. Depending on the study’s objectives, additional functional units like kilowatt-peak (kWp) or square meter (m²) may also be used^76,77. Similarly, the system boundary should be tailored to the goal by delineating the stages of the full life cycle. For example, a cradle-to-gate boundary is suitable for LCAs focusing on how the composition of the perovskite absorber affects impacts, whereas a gate-to-grave boundary can be applied to LCAs of end-of-life module recycling^60,78. However, such differences in functional units and system boundary definitions across studies can substantially limit the possibility of direct, one-to-one comparison of reported LCA results, even when they target similar perovskite photovoltaic technologies or device architectures. Even when the functional unit and system boundaries are harmonized, the comparability of LCA results remains limited. This is because key assumptions, such as assumed module lifetime, site-specific irradiation conditions, background databases, and impact assessment methods, often differ across studies. Therefore, the numerical results presented here should be interpreted as indicative ranges rather than strictly comparable absolute values. When comparing LCA studies, the reported results should be interpreted primarily as indicative trends, recurring hotspots, and major trade-offs, rather than as strict quantitative rankings. This is particularly important when functional units, system boundaries, lifetime assumptions, and inventory completeness differ across studies. Such a cautious interpretation is also consistent with recent circular-economy perspectives, which stress that LCA should guide eco-design while avoiding burden shifting across production, use, and end-of-life stages⁷⁹. Second, the life cycle inventory (LCI) compiles and organizes all relevant material and energy flows within the defined system boundary, normalized to the functional unit. Third, the life cycle impact assessment (LCIA) translates the inventory into impact indicators aligned with the study’s goals. For perovskite photovoltaics—where decarbonization and environmentally friendly energy technologies are central—global warming potential (GWP, kg CO₂-eq) is typically a primary environmental indicator. To advance eco-friendly manufacturing, recycling, and the use of green solvents, toxicity-related indicators, such as human, terrestrial, and marine toxicity are also frequently considered. In addition, energy-related metrics, such as cumulative energy demand (CED) are often selected⁸⁰. Because different studies may adopt distinct impact categories, characterization models, and indicator metrics, apparent differences in environmental performance should not be overinterpreted as true technological gaps but rather examined by considering these methodological choices. Finally, the interpretation phase ensures that the results are consistent with the defined goal and scope, identifies key contributors and uncertainties, and supports decision-making^81,82.
PSCs are typically fabricated by depositing multiple layers, using techniques that differ in energy demand and required raw materials. LCA have evaluated two representative deposition routes, vapor deposition (R1) and solution spin coating (R2), using previously reported device structures with PCEs of 15.4 % for R1 and 11.5 % for R2 (Fig. 4a)⁸³. The scope of LCA follows a cradle-to-gate and the 1 kWh of electricity produced by a PSCs with an assumed lifetime of one year as functional unit. It enables direct comparison across different device designs and technologies by normalizing impacts per kWh of electricity delivered. Both routes show the same ordering of dominant categories: freshwater ecotoxicity and human toxicity. Across the categories shown in Fig. 4b–d, R2 consistently exhibits lower impacts than R1, indicating a lower overall burden for the solution process. However, this should be interpreted in the context of methodological uncertainties inherent to LCA, including laboratory-scale inventories, system-boundary definitions, and assumed lifetimes, which can influence absolute values and the magnitude of the observed differences. Meanwhile, the principal contributors differ by route, with fluorine-doped tin oxide (FTO) dominating in R1 and the perovskite layer dominating R2. Although the input mass of perovskite required to fabricate the perovskite layer is smaller in R2 than in R1, the perovskite layer exerts a higher environmental impact in R2. This is because R2 requires electricity-intensive annealing, which accounts for about 95% of that layer’s burden and elevates impacts across categories. Although the use of Pb in PSCs raises critical concerns, its contribution to the human toxicity impact is relatively lower than even MAI. Accordingly, optimization should encompass solvents, electricity, and resource use in addition to Pb containment.
a Illustration of PSC device structures, preparing with vapor deposition (R1) and spin coating (R2) methods. Environmental impacts of PSCs fabricated via R1 and R2, expressed per functional unit of 1 kWh of produced electricity: b Freshwater ecotoxicity, c Human toxicity_cancer, and d Climate change. e Environmental impacts of different solution processes. f CED versus climate change impact for precursor iodides used to synthesize perovskite. g LCA results for perovskite A-site cation precursors, and h Climate change impact mapped across the ternary phase space for A-site perovskite precursor composition. a–d adapted from ref. ⁸³. Copyright 2015, Elsevier. e adapted from ref. ⁸⁴. Copyright 2024, EDP Sciences. f adapted from ref. ⁶¹. Copyright 2025, Wiley-VCH GmbH. g, h adapted from ref. ⁷⁸. Copyright 2020, American Chemical Society.
Recently, Rossi et al. compared four different solution processes, blade coating in glovebox, blade coating, spin coating, and spin coating + Press, for forming the perovskite layer using LCA to quantify environmental burdens (Fig. 4e)⁸⁴. Among these processes, blade coating shows the lowest impacts, which is associated with lower electricity demand and more efficient use of materials. The use of relatively benign solvents, for example IPA in place of CB, further reduced the indicators. Spin coating and spin coating + press produced similar overall results; the electricity advantage of spin coating was offset by greater reliance on hazardous solvents. These observations suggest that practical process selection should consider both energy consumption and solvent hazard simultaneously.
Taken together, the precursor and composition effects reinforce that processing choices must be evaluated alongside material choices and solvent management. These observations suggest that practical process selection should consider both energy consumption and solvent hazard simultaneously. This factor also influences perovskite composition because precursor production routes differ in energy and emissions. Depending on composition, different cationic iodides are used, and their environmental impacts vary accordingly (Fig. 4f). Notably, PbI₂ shows the lowest CED (85 MJ kg⁻¹) and the second-lowest climate-change impact (5.9 kg CO₂-eq kg⁻¹), values comparable to CuI. FAI, SbI₃, CsI, and BiI₃ follow with progressively higher indicators, whereas AgI exhibits the highest CED and climate-change values. The large burden for AgI arises mainly from silver mining and beneficiation in the life cycle inventory. More detailed LCAs of cationic iodide precursors have been reported⁷⁸. Contribution analysis for FAI, MAI, and CsI shows that solvent use and end-of-life treatment account for a major share of climate-change impact (Fig. 4g). MAI synthesis attributes ~48% of its climate-change indicator to spent-solvent incineration, and that of CsI is ~54.4%. Because precursor impacts differ, mapping the ternary composition space can indicate low-impact formulations when composition varies (Fig. 4h). For example, the Cs/FA region tends to show lower indicators, reflecting the relatively high burden associated with MAI. Nevertheless, composition also governs device properties, such as absorption edge and PCE, so precursor selection should balance environmental performance with photovoltaic function. Overall, these results reinforce that solvent choice, usage, and recovery are central to sustainable processing.
An analysis of eight polar aprotic solvents widely used in PSC fabrication has been reported (Fig. 5a, b)². The study integrated production, use, removal, and end-of-life scenarios in a life cycle framework, updating toxicity factors beyond simple carcinogenic labels. It highlighted regulatory concern for DMF-class solvents and quantified burdens associated with drying and post-processing. Among the candidates, DMSO shows the lowest combined human-health and environmental impacts, and solvent recovery is generally preferable to incineration. These results provide a quantitative basis for choosing solvents and for integrating solvent capture and recycling in scale-up. At the same time, process details, such as deposition route, solvent consumption, recovery yield, and device performance should be explicitly considered when interpreting environmental burdens.
a Schematic of LCA boundary for possible PSC production pathways. b Human health impacts associated with different solvents, expressed as disability-adjusted life years (DALYs) per quantity of emitted solvent. c Normalized environmental impacts (global warming potential (GWP), human toxicity (HT), Marine aquatic ecotoxicity (MAE)) of different perovskite precursor solvent systems: Ink 1 (DMF), Ink 2 (DMF and IPA), and Green solvent (DMSO and GBL), d Environmental impacts of synthesis procedure for MAPbI₃ using different solvents. e LCA contribution analysis of PSM prepared with different solvent systems. a, b adapted from ref. ². Copyright 2021, Springer Nature. c adapted from ref. ⁹. Copyright 2022, Elsevier. d adapted from ref. ¹⁰. Copyright 2025, The Royal Society of Chemistry. e adapted from ref. ²⁸. Copyright 2025, The Royal Society of Chemistry.
Subsequent LCA at device level examined inkjet printing as a scalable deposition route using a functional unit of 1 kWh (see the Fig. 5c)⁹. Using a cradle-to-gate boundary, the study reported markedly lower GWP and CED for inkjet-printed PSCs relative to spin-coated process. In addition, a green-solvent ink based on DMSO and γ-butyrolactone (GBL) showed lower impacts across categories than DMF (Ink 1) and DMF with IPA (Ink 2).
More recent analyses focused on biomass-derived GVL as alternative perovskite precursor for alleviating environmental concerns¹⁰. For MAPbI₃ and FAPbI₃, using GVL reduced overall environmental footprints versus GBL and DMF, with consistent improvements across midpoint and endpoint indicators, supporting solvent substitution with GVL and motivate parallel attention to precursor chemistry (Fig. 5d). A complementary system level study evaluated a GVL with EA as a green anti-solvent, combining device performance with techno economic analysis and LCA (Fig. 5e)²⁸. The assessment reported substantial reductions in manufacturing cost and climate-change impact compared with DMF/DMSO systems, while maintaining high device efficiency. It also identified break even conditions under different lifetimes and recycling assumptions, indicating that solvent substitution, electricity reduction, and recovery should be addressed together for scale up. Overall, the GVL and EA system emerges as a strong candidate for PSC commercialization and underscores the importance of anti-solvent selection. An anti-solvent LCA compared anisole with CB under controlled device parity and a cradle-to-grave boundary⁸⁵. Anisole shows lower carcinogenic human toxicity and freshwater ecotoxicity but a higher climate-change indicator because of its multistep synthesis. In practice, the required volume of anisole is much smaller than that of chlorobenzene, approximately one fifth. Considering actual usage, the overall burden during PSC fabrication can be substantially reduced. These observations indicate that LCA should be performed at the process level and that a clearly defined functional unit is essential for rational comparison.
Perovskite nanocrystals are also employed to form the perovskite active layer, so the environmental impact of their synthesis process should be considered. A recent study examined solvent substitution in CsPbX₃ (X = Cl, Br, I) nanocrystal synthesis by replacing 1-octadecene with limonene, a citrus-derived solvent, and coupled the laboratory results with LCA⁸⁶. The nanocrystals obtained in limonene showed structure and optical properties comparable to those produced in conventional media, indicating the feasibility of replacing more hazardous solvents. The LCA reported large reductions in global-warming potential, reaching about 83% for CsPbBr₃ and up to 95% when solvent recovery was implemented. These findings underscore that solvent choice is a major contributor to the overall burden and that closed-loop solvent recycling can further improve sustainability.
Beyond device fabrication, most recycling routes are solution based and consume solvents at each step. Many efforts have been made to recycle Pb from PSCs^{52,87,88,89,90,91}, and a comparative LCA has now been published that evaluates the principal processes⁵⁴. That analysis shows that organic solvents, such as DMF and EA markedly increase GWP and human-toxicity indicators, whereas water-based routes are substantially lower. Additionally, for recycling routes based on water, EA, and DMF, solvent reuse further reduces burdens; reusing the dissolution solvent ten times lowers GWP by approximately 89.55%. A recycling process using water, ethanol, and EA has also been evaluated by LCA to quantify the environmental impact of the recovery strategy¹⁷. Each solvent is purified by distillation and reintroduced into the process loop. Compared with landfilling, the recycling scenario shows lower energy requirements and a reduced environmental footprint, and the advantage persists over multiple cycles while maintaining 98.4% of the initial device efficiency. These observations indicate that PSC recycling should prioritize green solvents, minimize high-hazard organics, and integrate solvent recovery and reuse. Accordingly, LCA of recycling processes should be applied with care to ensure sustainable outcomes.
ILs are room-temperature molten salts composed entirely of ions, exhibiting negligible vapor pressure, low volatility, tunable coordination strength, and high thermal stability. In PSCs, these properties enable ambient-condition processing without an anti-solvent, since ILs can dissolve precursors, and regulate nucleation and grain growth³⁹. Their negligible vapor pressure reduces worker exposure to volatile organics, and toxic precursor solvents can be eliminated, making IL routes environmentally favorable⁹².
However, early LCA questioned whether ILs are genuinely green⁹³. Considering the full life cycle, ILs may exhibit greater environmental impacts than conventional organic solvents. Also, the environmental impacts depend on composition and production route⁹⁴, suggesting that life cycle performance should be evaluated carefully before a solvent is labeled green for use in sustainable technologies.
Fig. 6 compares solvent guidance developed by CHEM21, an EU public–private consortium, with LCA indicators. CHEM21 produced a solvent selection guide that classifies solvents by safety, health, and environmental criteria aligned with the Globally Harmonized System (GHS)²⁵. Fig. 6a, b contrasts scores for 12 representative PSC solvents: CHEM21 hazard rankings versus LCA endpoint results for environment and human health. Notably, significant gaps between the two frameworks are observed in most solvents. A solvent labeled hazardous by CHEM21 may still show comparatively favorable LCA outcomes, and the reverse can also occur. This arises from methodological scope. CHEM21 ratings are built primarily on intrinsic hazard and laboratory safety: flammability, acute and chronic toxicity, persistence, and regulatory status. These criteria are essential for worker protection and for immediate risk management, yet they do not quantify upstream energy use, greenhouse-gas emissions, or end-of-life burdens. By contrast, LCA aggregates cradle-to-gate or cradle-to-grave flows into endpoint categories, such as human health, ecosystems, and resources. Fig. 6c, d provide a detailed comparison of representative perovskite precursor solvent and anti-solvent. Consistent with the earlier discussion, the scoring differs between CHEM21 hazard guidance and LCA indicators. In particular, anisole is classified as favorable across the CHEM21 Safety, Health, and Environment categories, yet the LCA shows very high burdens in the corresponding endpoints for resource use, ecosystem damage, and human health. This contrast reflects LCA elements not included in CHEM21’ hazard ranking, namely multistep synthesis with high upstream energy and emissions, additional process electricity during use, and greater solvent consumption per step. Meanwhile, GVL shows the smallest LCA values in the set, including a negative value. The negative value in the resource category reflects allocation credits from its bio-based feedstock and energy integration in the inventory. Compared with DMSO, GVL receives a higher hazard score in CHEM21; however, in the LCA it exhibits much smaller impacts.
a CHEM21 Environment scores versus the LCIA Ecosystems endpoint; b CHEM21 Health scores versus the LCIA Human Health endpoint. Comprehensive comparisons for c perovskite light-absorber solvents and d anti-solvents include CHEM21 scores (Safety (S), Health (H), Environment (E)) and LCIA endpoints (Human Health (HH), Ecosystems (E), Resources (R)). CHEM21 scores were obtained from ref. ²⁵. and LCIA results were obtained with the ReCiPe 2016 Endpoint method, with LCIA values normalized to anisole as a reference to enable relative comparisons across solvents. Detailed values are reported in Supplementary Tables 2 and 3.
The application of AI in bottom-up LCAs as an alternative model for post-experimental evaluation is nothing new^95,96. The alternative models allow for the early evaluation of sustainability metrics and LCA scores. However, challenges include process scaling behaviors, electrification options, and uncertainties within chemical supply chains exist⁹⁷. Continuously updating datasets through text and data mining enables the sustainable design of PSCs by identifying environmental hotspots in a timely manner and providing real-time decision support. AI can help fill key knowledge gaps in prospective LCAs to provide decision support. AI will accelerate the shift from a “performance-driven” to a “sustainability-driven” PSC model.
Ramón et al. explored the feasibility of using AI models in LCA instead of complex first-principles models¹³. To improve the accuracy and efficiency of AI-integrated LCA, they conducted a bibliometric analysis of 387 publications. This analysis, which incorporated cluster and trend analyses, identified three modes of AI-LCA integration. Firstly, there is the rapid parameter-based prediction of LCI. Secondly, AI is employed as a surrogate model for a specific segment within the LCA process. Thirdly, comprehensive predictive models are constructed directly for large-scale LCA systems. The specific workflow is illustrated in Fig. 7a. It is widely recognized that cognitive uncertainty and random uncertainty within the modeling process constitute model uncertainty, quantitative uncertainty, and scenario uncertainty in LCA. The evaluation about AI research usually focuses on performance metrics, such as R² and mean square error (MSE), but neglects uncertainty assessment. Consequently, concerns have arisen about the lack of rigorous uncertainty analysis in LCA + AI combinations. Akrami et al. expanded upon research integrating NGBoost models with LCA. They developed a comparative analysis of GWP that considers uncertainty in AI predictions, traditional LCA, and combined uncertainty (see Fig. 7b)¹⁴. Their results show that incorporating a AI + LCA combining uncertainty analysis increases the range of potential GWPs, thus complicates decision making. As illustrated in Fig. 7c. It is suggested that modeling processes should integrate uncertainty analysis and sensitivity analysis to comprehensively evaluate variability in outcomes across different AI approaches. This approach can be applied to scenarios where data is scarce.
a Three AI application patterns were identified via cluster analysis and trend analysis. b Visualize the flow of information within AI and LCA models. This model incorporates uncertainties from ML and/or LCA components into a comprehensive modeling framework. c GWP distribution comparison in four different uncertainty analysis: control, NGBoost, LCA, and NGBoost+LCA. d–f Scatter plots of LCA indicators predicted by SMOreg: embodied energy for all structural attributes, net energy for all structural attributes, and net energy for active material attributes. a adapted from ref. ¹³. Copyright 2025, Elsevier. b and c adapted from ref. ¹⁴. Copyright 2022, Elsevier. d–f adapted from ref. ⁹⁸, copyright 2022, American Chemical Society.
Preliminary research on LCA + AI has been conducted in organic photovoltaics. David and Kettle investigated how to minimize environmental impact from materials and processes while maintaining device performance⁹⁸. Using device architecture information, performance metrics, and LCA indicators (embodied energy, E_Emb, and net energy, E_Net) from 1580 OSCs, they designed two AI models: sequential minimum optimization regression (SMOreg) and genetic algorithm clustering. To be specific, SMOreg evaluates the impact of different layer materials on LCA indicators, while genetic algorithm clustering identifies the optimal architecture for net efficiency output. Fig. 7d–f display scatter plots of LCA indicators predicted by SMOreg. For all structural attributes, the correlation coefficients for E_Emb and E_Net were 0.988 and 0.838, respectively. Focusing solely on active material attributes, E_Net achieved a correlation coefficient of 0.719, demonstrating reliable accuracy. Subsequent clustering identified the OSC architecture with the highest predicted E_Net as PET/Ag/PEDOT:PSS/P3HT/PCBM/PEDOT:PSS/Ag.
Green solvent approaches have gained more attention for reducing both environmental and occupational burdens in PSC fabrication. Several alternative solvent systems can yield high-quality perovskite films and competitive device performances while lowering toxicity. Nevertheless, important challenges remain in precursor solubility, nucleation and crystallization control, viscosity management, and the scala-up processing. Alcohol- and water-based routes still suffer from low lead-salt solubility and incomplete crystallization. IL-based systems require careful control of viscosity and thus the use of co-solvents. Many green anti-solvents are also highly sensitive to evaporation behavior and ambient processing conditions. Further work should improve solubility and coordination chemistry in benign media to support more robust green solvent systems. In addition, more stable crystallization pathways under realistic coating conditions, as well as robust solvent systems that are suitable for practical applications and compatible with solvent recovery are needed.
AI provides an additional tool for developing more sustainable PSC processes. Machine-learning models have been used to screen greener solvent candidates and predict film stability or crystallization behavior, reducing extensive repeated experiments. AI has also been applied to PSC recycling by using NLP-based protocols to predict suitable recovery strategies. However, most current studies rely on small datasets, simple solvent descriptors, and narrow design spaces are still at the proof-of-concept level. Thus, further studies should expand and diversify the available datasets and develop descriptors that more accurately represent coordination chemistry and phase evolution. Closer integration of AI, and experiments will be important to translate these tools into practical guidance for materials and process design.
LCA provides a quantitative framework to evaluate environmental impacts over the full life cycle of PSCs, including materials, solvents, and processing routes, encompassing materials, solvents, and processing routes. Previous work has identified solvent production and use, electricity-intensive annealing, and end-of-life treatment as major contributors to the overall impacts. However, differences in functional units, system boundaries, and inventory assumptions still make it difficult to compare values across LCA studies. More transparent and standardized reporting that reflects realistic process parameters would improve consistency between studies. AI can partly address data gaps in LCA by providing surrogate models for inventory data and environmental indicators, but uncertainty needs to be treated explicitly and kept consistent with experimental results.
For translation beyond laboratory demonstrations, sustainability gains must remain viable manufacturing-relevant constraints and practical end-of-life implementation. Solvent substitution should therefore be considered together with solvent intensity, energy demand, process integration, and realistic collection and handling conditions. Recycling pathways should likewise be assessed not only by recovery yield, but also by operational simplicity of solvent management.
Overall, advancing PSC sustainability will require coordinated progress in green solvent engineering, AI-guided design, and LCA-based evaluation. Green solvents reduce hazards at the material and process levels. AI accelerates the screening and optimization for robust solvent systems and process conditions. LCA then verifies whether proposed routes truly reduce environmental burdens. Using these three approaches together can shift PSC research from purely performance-driven optimization toward environmentally viable technologies and support responsible commercialization.
The data supporting this Perspective are available in the article, the Supplementary Information, and from the corresponding cited references.
Best Research-Cell Efficiency Chart.). NREL.
Vidal, R. et al. Assessing health and environmental impacts of solvents for producing perovskite solar cells. Nat. Sustain. 4, 277–285 (2021).
Article  Google Scholar 
Kim, H.J., Kim, Y.J., Han, G.S. & Jung, H.S. Green solvent strategies toward sustainable perovskite solar cell fabrication. Solar RRL 8, 2300910 (2024).
Park, J. et al. Controlled growth of perovskite layers with volatile alkylammonium chlorides. Nature 616, 724–730 (2023).
Article  ADS  CAS  PubMed  Google Scholar 
Zhao, K. et al. Peri-fused polyaromatic molecular contacts for perovskite solar cells. Nature 632, 301–306 (2024).
Guo, H. et al. Bifunctional modified biopolymer for highly efficient and stable perovskite solar cells and modules. Chem. Eng. J. 460, 141699 (2023).
Kim, H.J., Han, G.S. & Jung, H.S. Managing the lifecycle of perovskite solar cells: Addressing stability and environmental concerns from utilization to end-of-life. eScience 4, 100243 (2024).
Vidal, R. et al. Sustainable solvent selection for the manufacture of methylammonium lead triiodide (MAPbI3) perovskite solar cells. Nat. Sustain 4, 227–285 (2021).
Google Scholar 
Okoroafor, T. et al. Life cycle assessment of inkjet printed perovskite solar cells. J. Cleaner Prod. 373, 133665 (2022).
Campana, F., Lanari, D., De Angelis, F. & Vaccaro, L. Assessment of the environmental and human health impact in the synthesis and processing of metal halide perovskite active layers using GVL. Green. Chem. 27, 11903–11913 (2025).
Article  CAS  Google Scholar 
Mahmood, A., Sandali, Y. & Wang, J.-L. Easy and fast prediction of green solvents for small molecule donor-based organic solar cells through machine learning. Phys. Chem. Chem. Phys. 25, 10417–10426 (2023).
Article  CAS  PubMed  Google Scholar 
Giri, D., Mukherjee, A. & Rajan, K. Uncertainty informed screening for safer solvents used in the synthesis of perovskite based solar cells via machine learning. J. Chem. Inf. Model. 65, 7901–7918 (2022).
Martínez-Ramón, N., Calvo-Rodríguez, F., Iribarren, D. & Dufour, J. Frameworks for the application of machine learning in life cycle assessment for process modeling. Clean. Environ. Syst. 14, 100221 (2024).
Article  Google Scholar 
Akrami, M., Porter, M. D. & Colosi, L. M. Addressing uncertainty in machine learning-integrated life cycle assessment (ML + LCA). J. Environ. Manag. 389, 126225 (2025).
Article  Google Scholar 
Kim, H. J. et al. Environmentally viable solvent management in perovskite solar cell recycling process. ACS Energy Lett. 8, 4330–4337 (2023).
Article  CAS  Google Scholar 
Whitaker, J. B. et al. Scalable slot-die coating of high performance perovskite solar cells. Sustain. Energy Fuels 2, 2442–2449 (2018).
Article  CAS  Google Scholar 
Larini, V. et al. Circular management of perovskite solar cells using green solvents: from recycling and reuse of critical components to life cycle assessment. EES Solar 1, 378–390 (2025).
Bruening, K. et al. Scalable fabrication of perovskite solar cells to meet climate targets. Joule 2, 2464–2476 (2018).
Article  CAS  Google Scholar 
Iseri F, Iseri H, Iakovou E, Pistikopoulos EN. A circular economy systems engineering framework for waste management of photovoltaic panels. Indus. Eng. Chem. Res. 64, 14986–14997 (2025).
D’Adamo, I., Ferella, F., Gastaldi, M., Ippolito, N. M. & Rosa, P. Circular solar: evaluating the profitability of a photovoltaic panel recycling plant. Waste Manag. Res. 41, 1144–1154 (2023).
Article  PubMed  Google Scholar 
Hsieh, T.Y., Wei, T.C., Wu, K.L., Ikegami, M. & Miyasaka, T. Efficient perovskite solar cells fabricated using an aqueous lead nitrate precursor. Chem. Commun. 51, 13294–13297 (2015).
Article  CAS  Google Scholar 
Yun, H.S. et al. Ethanol-based green-solution processing of α-formamidinium lead triiodide perovskite layers. Nat. Energy 7, 828–834 (2022).
Article  ADS  CAS  Google Scholar 
Miao, Y. et al. Green solvent enabled scalable processing of perovskite solar cells with high efficiency. Nat. Sustain. 6, 1465–1473 (2023).
Zhai, P. et al. Performance-limiting formation kinetics in green water-processed perovskite solar cells. Energy Environ. Sci. 7, 3742 (2023).
Prat, D. et al. CHEM21 selection guide of classical-and less classical-solvents. Green. Chem. 18, 288–296 (2016).
Article  Google Scholar 
Doolin, A. J. et al. Sustainable solvent selection for the manufacture of methylammonium lead triiodide (MAPbI 3) perovskite solar cells. Green. Chem. 23, 2471–2486 (2021).
Article  CAS  Google Scholar 
Chen, L. et al. Solvent engineering for scalable and sustainable fabrication of lead-tin perovskite solar cells. Adv. Energy Mater. 15, 2405941 (2025).
Bae, J., Cha, J., Kim, M. & Han, J. Advancing perovskite solar cells with biomass-derived solvents: a pathway to sustainability. Green. Chem. 27, 10598–10611 (2025).
Article  CAS  Google Scholar 
Satale, V.V. et al. Green solvent enabled perovskite ink for ambient-air-processed efficient inkjet-printed perovskite solar cells. Adv. Funct. Mater. 35, 2503717 (2025).
Feng, Y. et al. Solution-processed perovskite solar cells using environmentally friendly solvent system. Thin Solid Films 636, 639–643 (2017).
Article  ADS  CAS  Google Scholar 
Oez, S. et al. Protic ionic liquid assisted solution processing of lead halide perovskites with water, alcohols and acetonitrile. Nano Energy 51, 632–638 (2018).
Article  CAS  Google Scholar 
Zhai, P. et al. Light modulation strategy for highest-efficiency water-processed perovskite solar cells. Matter 5, 4450–4466 (2022).
Article  CAS  Google Scholar 
Zheng, N. et al. Controlling Crystallization of Aqueous-Processed Planar Perovskite Films via Sodium Dodecyl Sulfonate Surfactant Modulation. Molecules 30, 2146 (2025).
Article  CAS  PubMed  Google Scholar 
Radicchi, E. et al. Combined computational and experimental investigation on the nature of hydrated iodoplumbate complexes: insights into the dual role of water in perovskite precursor solutions. J. Phys. Chem. B 124, 11481–11490 (2020).
Article  CAS  PubMed  Google Scholar 
Chen, H. et al. A solvent-and vacuum-free route to large-area perovskite films for efficient solar modules. Nature 550, 92–95 (2017).
Article  ADS  CAS  PubMed  Google Scholar 
Fang, J. et al. Microstructure and lattice strain control towards high-performance ambient green-printed perovskite solar cells. J. Mater. Chem. A 9, 13297–13305 (2021).
Article  CAS  Google Scholar 
Du, Y. et al. Ionic liquid treatment for highest-efficiency ambient printed stable all-inorganic CsPbI3 perovskite solar cells. Adv. Mater. 34, 2106750 (2022).
Article  CAS  Google Scholar 
Ünlü, F. et al. Toward green processing of perovskite solar cells: protic ionic liquids enable water-and alcohol-based MAPbI3 precursors inks for slot-die coating. Adv. Energy Mater. 15, 2403626 (2025).
Article  Google Scholar 
Chao, L., Niu, T., Xia, Y., Chen, Y. & Huang, W. Ionic liquid for perovskite solar cells: an emerging solvent engineering technology. Acc. Mater. Res. 2, 1059–1070 (2021).
Article  CAS  Google Scholar 
Nowee, S. M., Abbas, A. & Romagnoli, J. A. Antisolvent crystallization: model identification, experimental validation and dynamic simulation. Chem. Eng. Sci. 63, 5457–5467 (2008).
Article  Google Scholar 
Taylor, A. D. et al. A general approach to high-efficiency perovskite solar cells by any antisolvent. Nat. Commun. 12, 1878 (2021).
Article  ADS  CAS  PubMed  Google Scholar 
Bu, T. et al. Synergic interface optimization with green solvent engineering in mixed perovskite solar cells. Adv. Energy Mater. 7, 1700576 (2017).
Article  Google Scholar 
Lee, H. B., Jeon, M. K., Kumar, N., Tyagi, B. & Kang, J. W. Boosting the efficiency of SnO2-triple cation perovskite system beyond 20% using nonhalogenated antisolvent. Adv. Funct. Mater. 29, 1903213 (2019).
Article  Google Scholar 
Kim, S. et al. Morphology control of perovskite in green antisolvent system for MAPbI3-based solar cells with over 20% efficiency. Sol. Energy Mater. Sol. Cells 203, 110197 (2019).
Article  CAS  Google Scholar 
Yang, F. et al. Dependence of acetate-based antisolvents for high humidity fabrication of CH3NH3PbI3 perovskite devices in ambient atmosphere. ACS Appl. Mater. interfaces 10, 16482–16489 (2018).
Article  CAS  PubMed  Google Scholar 
Wang, L. et al. The mechanism of universal green antisolvents for intermediate phase controlled high-efficiency formamidinium-based perovskite solar cells. Mater. Horiz. 7, 934–942 (2020).
Article  CAS  Google Scholar 
Cao, X. et al. Green solvents processed all functional layers for efficient perovskite solar cells. RSC Sustain. 1, 1290–1297 (2023).
Article  CAS  Google Scholar 
Wang, M. et al. Systematic optimization of perovskite solar cells via green solvent systems. Chem. Eng. J. 387, 123966 (2020).
Article  CAS  Google Scholar 
Rizzo, A., Listorti, A. & Colella, S. Chemical insights into perovskite ink stability. Chem 8, 31–45 (2022).
Article  CAS  Google Scholar 
Wang, X. et al. Perovskite solution aging: what happened and how to inhibit? Chem 6, 1369–1378 (2020).
Article  CAS  Google Scholar 
Kim, B. J. et al. Selective dissolution of halide perovskites as a step towards recycling solar cells. Nat. Commun. 7, 11735 (2016).
Article  ADS  CAS  PubMed  Google Scholar 
Binek, A. et al. Recycling perovskite solar cells to avoid lead waste. ACS Appl Mater. Interfaces 8, 12881–12886 (2016).
Article  CAS  PubMed  Google Scholar 
Kadro, J. M. et al. Proof-of-concept for facile perovskite solar cell recycling. Energy Environ. Sci. 9, 3172–3179 (2016).
Article  CAS  Google Scholar 
Lee, J. M., Kim, H. J. & Jung, H. S. Life cycle assessment of lead recycling processes in perovskite solar cells. Chem. Commun. 61, 1850–1853 (2025).
Article  CAS  Google Scholar 
Xiao, X. et al. Aqueous-based recycling of perovskite photovoltaics. Nature 638, 670–675 (2025).
Article  ADS  CAS  PubMed  Google Scholar 
Valadez-Villalobos, K., Worsley, C., Garcia-Rodriguez, R., Watson, T. & Davies, M. Sustainable remanufacturing of mesoscopic carbon perovskite solar cells using green solvents. RSC Sustain. 4, 754–759 (2026).
Wang, K. et al. One-key-reset” recycling of whole perovskite solar cell. Matter 4, 2522–2541 (2021).
Article  CAS  Google Scholar 
Zheng, Z. et al. High-value organic solvent recovery and reuse in perovskite solar cell manufacturing. Sci. Adv. 11, eadt6008 (2025).
Article  ADS  CAS  PubMed  Google Scholar 
Lu, J., Wang, C., Zhu, J., Wu, Y. & Ruan, J. Recycling of end-of-life photovoltaic modules: high throughput screening reagents by machine learning and classification of decapsulation mechanisms. Chem. Eng. J. 521, 166477 (2025).
Rodriguez-Garcia, G. et al. Life cycle assessment of coated-glass recovery from perovskite solar cells. ACS Sustain. Chem. Eng. 9, 15239–15248 (2021).
Article  CAS  Google Scholar 
Vidal, R. et al. Assessing the environmental impact of pnictogen-based perovskite-inspired materials for indoor photovoltaics. Adv. Energy Mater. 15, 2403981 (2025).
Article  CAS  Google Scholar 
Gallant, B. M. et al. A green solvent enables precursor phase engineering of stable formamidinium lead triiodide perovskite solar cells. Nat. Commun. 15, 10110 (2024).
Article  ADS  CAS  PubMed  Google Scholar 
Wei, J. et al. Machine learning in materials science. InfoMat 1, 338–358 (2019).
Article  CAS  Google Scholar 
Chen, W. et al. Artificial intelligence high-throughput prediction building dataset to enhance the interpretability of hybrid halide perovskite bandgap. J. Energy Chem. 109, 649–661 (2025).
Jo, B., Chen, W. & Jung, H. S. Comprehensive review of advances in machine-learning-driven optimization and characterization of perovskite materials for photovoltaic devices. J. Energy Chem. 101, 298–323 (2025).
Article  CAS  Google Scholar 
Chen, W., Mularso, K.T., Jo, B. & Jung, H.S. Indoor light energy harvesting perovskite solar cells: from device physics to AI-driven strategies. Mater. Horizons 12, 3691–3711 (2025).
Rezaee, E. et al. Green solvent ethanol-based inks for industrially applicable deposition of high-quality perovskite films for optoelectronic device applications. Small Methods 8, 2300564 (2024).
Article  CAS  Google Scholar 
Huang, X. et al. Solvent racing crystallization: Low-solvation dispersion cosolvents for high-quality halide perovskites in photovoltaics. Joule 7, 1556–1573 (2023).
Article  CAS  Google Scholar 
Pang, S. et al. High-efficiency organic solar cells processed from a real green solvent. Mater. Horiz. 10, 473–482 (2023).
Article  CAS  PubMed  Google Scholar 
Meng, F. et al. Sustainable green solvents for lead halide perovskites. Adv. Sustain. Syst. 9, 2400695 (2025).
Article  CAS  Google Scholar 
Mahmood, A. & Wang, J.-L. A time and resource efficient machine learning assisted design of non-fullerene small molecule acceptors for P3HT-based organic solar cells and green solvent selection. J. Mater. Chem. A 9, 15684–15695 (2021).
Article  CAS  Google Scholar 
Lee, M.-H. Interpretable machine-learning for predicting power conversion efficiency of non-halogenated green solvent-processed organic solar cells based on Hansen solubility parameters and molecular weights of polymers. Sol. Energy 261, 7–13 (2023).
Article  ADS  CAS  Google Scholar 
Huang, Y., Li, S. & Zhang, L. Accelerated multisolvent prediction for aqueous stable halide perovskite materials. ACS Appl. Mater. Interfaces 15, 48771–48784 (2023).
Article  CAS  PubMed  Google Scholar 
Ma, L. et al. Low-toxicity solvent processing in ambient air for perovskite solar cells via two-step Bayesian machine learning. J. Energy Chem. 111, 737–743 (2025).
Sprague, C.I., De, La, Asunción-Nadal, V. & García-Fernández, A. Implementing AI in Advanced Recycling of Perovskite Solar Cells. In: 2024 9th International Conference on Smart and Sustainable Technologies (SpliTech)), (IEEE, 2024).
Frischknecht, R. et al Methodology guidelines on life cycle assessment of photovoltaic 2020.). National Renewable Energy Laboratory (NREL), Golden, CO (United States, 2020).
Alsema, E. Energy requirements of thin-film solar cell modules—a review. Renew. Sustain. Energy Rev. 2, 387–415 (1998).
Article  CAS  Google Scholar 
Khalifa, S. A., Spatari, S., Fafarman, A. T. & Baxter, J. B. Environmental sustainability of mixed cation perovskite materials in photovoltaics manufacturing. ACS Sustain. Chem. Eng. 8, 16537–16548 (2020).
Article  CAS  Google Scholar 
Charles, R. G., Doolin, A., García-Rodríguez, R., Villalobos, K. V. & Davies, M. L. Circular economy for perovskite solar cells–drivers, progress and challenges. Energy Environ. Sci. 16, 3711–3733 (2023).
Article  Google Scholar 
Weyand, S., Wittich, C. & Schebek, L. Environmental performance of emerging photovoltaic technologies: assessment of the status quo and future prospects based on a meta-analysis of life-cycle assessment studies. Energies 12, 4228 (2019).
Article  CAS  Google Scholar 
Iso ISO. 14040. Environmental management—Life Cycle Assessment—principles and Framework 578, 235–248 (2006).
Iso ISO. 14044: 2006. Environmental management—Life Cycle Assessment—requirements and Guidelines, 1–46 (2006).
Espinosa, N., Serrano-Luján, L., Urbina, A. & Krebs, F. C. Solution and vapour deposited lead perovskite solar cells: ecotoxicity from a life cycle assessment perspective. Sol. Energy Mater. Sol. Cells 137, 303–310 (2015).
Article  CAS  Google Scholar 
Rossi, F. et al. Comparative life cycle assessment of different fabrication processes for perovskite solar mini-modules. EPJ Photovolt. 15, 20 (2024).
Article  ADS  CAS  Google Scholar 
Kamali, A. K. et al. Validating the “greenness” of chemicals via life cycle assessment: the case of anisole as an anti-solvent in perovskite solar cells. RSC Sustain. 2, 3036–3046 (2024).
Article  CAS  Google Scholar 
Pratolongo, D. et al. Lead halide perovskites nanocrystals synthesized in a green, reusable solvent. Small 21, 2500535 (2025).
Park, S. Y. et al. Sustainable lead management in halide perovskite solar cells. Nat. Sustain. 3, 1044–1051 (2020).
Article  Google Scholar 
Feng, X. et al. Close-loop recycling of perovskite solar cells through dissolution-recrystallization of perovskite by butylamine. Cell Rep. Phys. Sci. 2, 100341 (2021).
Ren, M. et al. Lead stabilization and iodine recycling of lead halide perovskite solar cells. ACS Sustain. Chem. Eng. 9, 16519–16525 (2021).
Article  CAS  Google Scholar 
Hong, JS. et al. High-throughput Pb recycling for perovskite solar cells using biomimetic whitlockite. Energy Environ. Mater. 6, e12374 (2022).
Chen, B. et al. Recycling lead and transparent conductors from perovskite solar modules. Nat. Commun. 12, 5859 (2021).
Article  ADS  CAS  PubMed  Google Scholar 
Kim, H.S. et al. Sustainable green process for environmentally viable perovskite solar cells. ACS Energy Lett. 7, 1154–1177 (2022).
Article  CAS  Google Scholar 
Zhang, Y., Bakshi, B. R. & Demessie, E. S. Life cycle assessment of an ionic liquid versus molecular solvents and their applications. Environ. Sci. Technol. 42, 1724–1730 (2008).
Article  ADS  CAS  PubMed  Google Scholar 
Mehrkesh, A. & Karunanithi, A. T. Life-cycle perspectives on aquatic ecotoxicity of common ionic liquids. Environ. Sci. Technol. 50, 6814–6821 (2016).
Article  ADS  CAS  PubMed  Google Scholar 
Kleinekorte, J. et al. APPROPRIATE life cycle assessment: a process-specific, predictive impact assessment method for emerging chemical processes. ACS Sustain. Chem. Eng. 11, 9303–9319 (2023).
Article  CAS  Google Scholar 
Sun, Y., Wang, X., Ren, N., Liu, Y. & You, S. Improved machine learning models by data processing for predicting life-cycle environmental impacts of chemicals. Environ. Sci. Technol. 57, 3434–3444 (2022).
Article  ADS  PubMed  Google Scholar 
Blanco, C., Pauliks, N., Donati, F., Engberg, N. & Weber, J. Machine learning to support prospective life cycle assessment of emerging chemical technologies. Curr. Opin. Green. Sustain. Chem. 50, 100979 (2024).
Article  CAS  Google Scholar 
David, T. W. & Kettle, J. Design for a sustainability approach to organic solar cell design: the use of machine learning to quantify the trade-off between performance, stability, and environmental impact. J. Phys. Chem. C. 126, 4774–4784 (2022).
Article  CAS  Google Scholar 
Download references
This study was supported by the National Research Foundation of Korea (NRF) (RS-2025-00522430, RS-2025-02316700), and by the Korea Institute of Energy Technology Evaluation and Planning (KETEP) and the Ministry of Climate, Energy & Environment (MCEE) of the Republic of Korea (RS-2025-25450823). This research was supported by the SungKyunKwan University and the BK21 FOUR(Graduate School Innovation) funded by the Ministry of Education (MOE, Korea) and National Research Foundation of Korea (NRF).
These authors contributed equally: Hee Jung Kim, Wenning Chen, Jae Myeong Lee.
School of Advanced Materials Science and Engineering, Sungkyunkwan University, Suwon, South Korea
Hee Jung Kim, Jae Myeong Lee & Hyun Suk Jung
Department of Future Energy Engineering, Sungkyunkwan University, Suwon, South Korea
Wenning Chen
SKKU Institute of Energy Science and Technology (SIEST), Sungkyunkwan University, Suwon, South Korea
Hyun Suk Jung
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
H.S.J. and H.J.K. conceived the study. H.J.K., W.C., and J.M.L. performed the investigation and contributed to the writing, review, and discussion of the manuscript. H.S.J. supervised the study and contributed to the review and discussion of the manuscript. H.J.K., W.C., and J.M.L. contributed equally to this work. H.S.J. is the corresponding author.
Correspondence to Hyun Suk Jung.
The authors declare no competing interests.
Nature Communications thanks Matthew L. Davies, Eva Unger Unger, and Xun Xiao for their contribution to the peer review of this work.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
Reprints and permissions
Kim, H.J., Chen, W., Lee, J.M. et al. AI-driven green processing and life cycle assessment for sustainable perovskite solar cells. Nat Commun 17, 4512 (2026). https://doi.org/10.1038/s41467-026-73255-1
Download citation
Received: 28 November 2025
Accepted: 07 May 2026
Published: 20 May 2026
Version of record: 20 May 2026
DOI: https://doi.org/10.1038/s41467-026-73255-1
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative
Advertisement
Nature Communications (Nat Commun)
ISSN 2041-1723 (online)
© 2026 Springer Nature Limited
Sign up for the Nature Briefing: Anthropocene newsletter — what matters in anthropocene research, free to your inbox weekly.

source