Quantifying Perovskite Solar Cell Degradation via Machine Learning from Spatially Resolved Multimodal Luminescence Time Series
Giulio Barletta 1,†, Simon Ternes 2,†, Saif Ali 2, Zohair Abbas 2, Chiara Ostendi 2, Marialucia D’Addio 2, Erica Magliano 2, Pietro Asinari 1,3, Eliodoro Chiavazzo 1,, Aldo Di Carlo 2,4, · 2026
Read Quantifying Perovskite Solar Cell Degradation via Machine Learning from Spatially Resolved Multimodal Luminescence Time Series by Giulio Barletta 1,†, Simon Ternes 2,†, Saif Ali 2, Zohair Abbas 2, Chiara Ostendi 2, Marialucia D’Addio 2, Erica Magliano 2, Pietro Asinari 1,3, Eliodoro Chiavazzo 1,, Aldo Di Carlo 2,4, (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-12857.txt.
Introduction
Giulio Barletta and co-authors track perovskite solar cell decay with machine learning on luminescence images. These cells are efficient, yet their working stability blocks wide use. Electrical sweeps give good metrics but are slow and miss local damage. Their LumPerNet reads electroluminescence and photoluminescence images taken as a device ages. It estimates how much efficiency the cell retains. Global brightness change held most of the signal, while spatial detail added robustness.
Catalog record
| CREATOR | Giulio Barletta 1,†, Simon Ternes 2,†, Saif Ali 2, Zohair Abbas 2, Chiara Ostendi 2, Marialucia D’Addio 2, Erica Magliano 2, Pietro Asinari 1,3, Eliodoro Chiavazzo 1,, Aldo Di Carlo 2,4, |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 87896 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.12857
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Quantifying Perovskite Solar Cell Degradation via Machine Learning from Spatially Resolved Multimodal Luminescence Time Series Giulio Barletta 1,†, Simon Ternes 2,†, Saif Ali 2, Zohair Abbas 2, Chiara Ostendi 2, Marialucia D’Addio 2, Erica Magliano 2, Pietro Asinari 1,3, Eliodoro Chiavazzo 1,, Aldo Di Carlo 2,4, Abstract Perovskite solar cells (PSCs) have rapidly gained in power conversion efficiency (PCE) within the last 15 years, providing a promising alternative or add-on to established silicon semiconductors for future’s large-scale photovoltaic (PV) deployment. Yet, besides scalability of fabrication routines, operational stability remains a key bottleneck for commercial roll-out of perovskite based PV. In this context, reliable and fast assessment of the state of health and degradation mechanisms that is compatible with field deployment are of utmost importance. Electrical diagnostics such as current–voltage (J–V) sweeps under illumination provide accurate performance metrics but are time-consuming and do not resolve spatially localized degradation, motivating non-invasive imaging-based alternatives. Here, we present a deep-learning framework to estimate PSC efficiency retention R PCE = PCE t / PCE 0 R_{\mathrm{PCE}}=\mathrm{PCE}_{t}/\mathrm{PCE}_{0} directly from multimodal luminescence imaging acquired during device aging, and we demonstrate that leveraging spatially resolved luminescence patterns may reveal key to improving prediction performance and robustness compared to intensity-only baselines. In our approach, each learning sample consists of electroluminescence (EL), open-circuit photoluminescence (PL oc ), and short-circuit photoluminescence (PL sc ) acquired at an aged state together with device-specific reference images at t = 0 t=0 , enabling the model to learn degradation-relevant spatial changes relative to the initial condition. These data were acquired over a time span of 5 5\text{\,} – 70 h 70\text{\,}\mathrm{h} in an automated, house-built measurement platform. We introduce LumPerNet, a compact convolutional neural network that regresses R PCE R_{\mathrm{PCE}} from stacked multimodal image tensors, and benchmark it against an intensity-only baseline multilayer perceptron regressor. Using a leakage-aware evaluation protocol with device-level held-out testing and four-fold cross-validation, and restricting analysis to the operational window R PCE ∈ [ 0.8 , 1.2 ] R_{\mathrm{PCE}}\in[0.8,1.2] , LumPerNet achieves a substantially higher and more robust performance (MAE: − 23.4 % -23.4\% , RMSE: − 25.6 % -25.6\% , R 2 : + 0.417 R^{2}:+0.417 ) with respect to the intensity-only baseline, confirming that spatially resolved EL/PL patterns carry predictive information beyond global luminescence decay. To quantify the role of multimodality, we further perform an exhaustive modality ablation study over all single- and paired-modality inputs. We find that multimodal training yields the best generalization, while carefully chosen bimodal configurations can retain most of the full-model performance, whereas other pairings can be unstable across device splits, highlighting that complementary physical contrast, rather than channel count, governs robustness. Representative device trajectories further show temporally coherent R PCE R_{\mathrm{PCE}} estimates and uncertainty quantification via model ensembling. Overall, this work establishes a reproducible pipeline linking automated luminescence imaging to electrical R PCE R_{\mathrm{PCE}} labels, and positions spatially resolved imaging as a practical route for accelerated stability testing and non-invasive degradation monitoring in perovskite photovoltaics. 2 2 footnotetext: Authors contributed equally 1 Introduction Perovskite solar cells (PSCs) have emerged as one of the most promising emerging photovoltaic (PV) technologies green_solar_2025 , kojima_organometal_2009 , green_emergence_2014 , jung_perovskite_2015 , kim_high-efficiency_2020 , correa-baena_promises_2017 , achieving a remarkable leap in power conversion efficiency (PCE), from around 3.8% in 2009 to over 27% for single junction PSCs xiong_homogenized_2025 , noauthor_longi_2024 at the time of writing. Furthermore, the flexible bandgap of perovskite absorber thin films has enabled their deployment in the tandem device architecture, yielding 34.85% for perovskite–Si tandem cells noauthor_3485_2025 , and 30.1% for perovskite–perovskite tandem cells in 2025 lin_all-perovskite_2025 . Despite these performance milestones, their medium- to long-term operational stability remains a critical challenge, with reports demonstrating stable operation on the order of 10^{3} h under standardized stress protocols, although full T_{80} values are not always reached or consistently reported khenkin_consensus_2020 , wang_review_2019 , mokabane_review_2025 , yaghoobi_nia_co-crystal_2026 . Unlike silicon PV, which have demonstrated moderate degradation over three decades defining the industry standard fazal_progress_2023 , cui_technoeconomic_2022 , chen_technology_2023 , PSCs are highly sensitive to environmental stressors such as moisture, oxygen, UV light, and thermal cycling. On the short to medium timescale (up to 10^{2} h), the performance of perovskite based PV was shown to fluctuate according to mostly reversible transient effects such as light soaking and path dependence bian_unraveling_2025 , lin_light_2023 . However, on the medium to long timescale ( >10^{2} h), irreversible degradation of the perovskite crystal structure to lead halide (rich) phases is routinely observed ahmad_instability_2017 , kalasariya_how_2026 , dubose_hole_2022 , guo_unveiling_2025 . This instability is mediated by intrinsic material vulnerabilities (including phase transitions, ion migration, and interfacial degradation) as well as extrinsic factors like poor encapsulation and crystallographic defects zhu_long-term_2023 , mokabane_review_2025 , ahmedchowdhury_stability_2023 , ahmad_instability_2017 , wang_review_2019 , kim_material_2016 , silverman_durability_2025 . While stability has historically been one of the main barriers for perovskite PV, recent years have witnessed significant progress toward addressing these limitations: for instance, early outdoor field trials have demonstrated stable operation despite the technology’s relative youth jiang_towards_2023 , jost_perovskite_2020 , meheretu_recent_2024 . Furthermore, several studies document thousands of hours of continuous operation under realistic conditions and under standardized stress conditions similar to those applied to silicon jacobsson_open-access_2022 , nur-e-alam_current_2025 , azmi_damp_2022 , mavlonov_thermal_2024 , lintangpradipto_single-crystal_2023 , such as elevated-temperature storage or damp heat (ISOS D2 and D3 elevated stress tests khenkin_consensus_2020 , reese_consensus_2011 ). However, translating these stability advances into rapid iteration and scalable quality control still requires monitoring tools that are faster than repeated full electrical characterization and that can localize where degradation initiates. This motivates non-invasive luminescence imaging as a practical readout for accelerated testing, where data-driven models can translate spatially resolved patterns into device-level performance retention even when the target metric is spatially averaged, as explored in this work. The operational stability of metal-halide perovskites is governed by coupled chemical and ionic processes that are strongly influenced by the environment and by electrostatic boundary conditions within the device stack khenkin_consensus_2020 , wang_review_2019 . Operational stability drifts can be broadly grouped into reversible (metastable) and irreversible contributions, which may coexist and act on different time scales khenkin_consensus_2020 , vkhenkin_reconsidering_2018 . A central irreversible failure route is the partial or complete conversion of the perovskite absorber into PbI 2 (and volatile organic species in MA/FA-based compositions), which can be triggered by moisture-assisted hydrolysis and hydrate formation, photo-oxidation in the presence of oxygen, and elevated temperature, each of which perturbs the ionic lattice and facilitates defect formation and ion motion wang_review_2019 , aristidou_role_2015 . At the device level, both irreversible and reversible performance changes are closely linked to mobile ionic defects (notably halide-related vacancies and interstitials) that redistribute under illumination and bias, leading to internal field screening, interfacial ion accumulation, and time-dependent changes in recombination, hysteresis, and apparent performance khenkin_consensus_2020 , thiesbrummel_ion-induced_2024 . In mixed-halide compositions, ionic redistribution can also manifest as light-induced halide segregation, with partial reversibility upon returning to dark conditions kalasariya_how_2026 . In addition to degradation, repeated or continuous bias and/or illumination (“light soaking”) can induce temporary performance improvements, often attributed to evolving interfacial electrostatics and contact properties mediated by ionic redistribution rather than permanent bulk passivation kress_persistent_2022 , moghadamzadeh_spontaneous_2020 , shao_elimination_2016 . Consistently, partial or even near-complete recovery of device performance after dark storage has