Hybrid Simulation and Machine Learning Framework for Optimizing Cs₂InAgBr₆-Based Lead-Free Perovskite Solar Cells

Abstract The development of lead-free perovskite solar cells (PSCs) is a crucial step toward environmentally sustainable photovoltaics. In this study, we present a hybrid modeling approach that integrates SCAPS-1D simulations with machine learning (ML) to optimize the photovoltaic performance of Cs₂InAgBr₆ double perovskite solar cells. Using SCAPS-1D, we explore the effects of transport layer selection, absorber thickness, doping, and defect density—identifying WS₂ as the electron transport layer (ETL) and CuI as the hole transport layer (HTL) in a device that achieves a simulated power conversion efficiency (PCE) of 19.26%. We then generate a dataset of 1,690 device configurations and train ten supervised ML models to predict PCE. The artificial neural network (ANN) model achieves high predictive accuracy (R² = 0.9983, RMSE = 0.2025), while SHAP and LIME analyses reveal fill factor and short-circuit current density as the most influential parameters. This work demonstrates a scalable and accurate pathway for optimizing lead-free photovoltaic materials by bridging device physics with data-driven modeling. To our knowledge, this is the first study that combines SCAPS-1D simulations and explainable AI (SHAP, LIME) to optimize and interpret the performance of Cs₂InAgBr₆-based perovskite solar cells. This integrative method enables both accurate prediction and physical interpretability, accelerating the design of sustainable lead-free photovoltaics.