Machine learning-assisted optimization of CsPbI₃-based all-inorganic perovskite solar cells: A combined SCAPS-1D and XGBoost approach

• Innovative Methodology : Combines SCAPS-1D device simulations with Machine Learning (XGBoost/SHAP) for accelerated, interpretable optimization. • High Impact : 26% relative PCE improvement via defect/thickness engineering. • Scalability : Dataset of 56,390 configurations provides a benchmark for future studies. • Practical Insights : Identifies critical thresholds (e.g., defect densities, layer thicknesses) for experimental validation. The commercialization of perovskite solar cells (PSCs) is hindered by the instability of organic components and the resource-intensive nature of experimental optimization. Machine learning (ML) is revolutionizing the discovery and optimization of photovoltaic devices by reducing reliance on conventional trial-and-error approaches. This study aims to optimize the performance of CsPbI₃-based all-inorganic PSCs using a combined SCAPS-1D and machine learning (ML) approach. We generated 56,390 unique device configurations via SCAPS-1D simulations, varying layer thicknesses and defect densities. Five ML models were trained, with XGBoost achieving the highest accuracy (R² = 0.999). Feature importance was analyzed using SHAP. Optimization increased the PCE from 15.15% to 19.16%, with the perovskite layer thickness (2 µm) and defect density (<10¹⁵ cm⁻³) identified as critical parameters. This study highlights the potential of ML-driven optimization in perovskite solar cells, offering a systematic and data-driven approach to enhancing device efficiency and accelerating the development of next-generation photovoltaics.