Harnessing machine learning to revolutionize perovskite solar cells: Optimizing RbTmCl3-based photovoltaic performance

Abstract

This study investigates the performance optimization of RbTmCl₃-based perovskite solar cells (PSCs) by combining machine learning (ML) models with the SCAPS-1D simulator. A comprehensive dataset was created by analyzing the effects of key parameters, including variations in the electron transport layer (ETL), hole transport layer (HTL), absorber thickness, as well as the presence of defects, doping, and impurities in the CsTmCl₃ absorber. Additional factors were examined, such as interface defects at the WS₂/ RbTmCl₃ and RbTmCl₃/CBTS junctions, temperature fluctuations, and resistance effects (series and shunt). Among eight ML models, the artificial neural network (ANN) outperformed the others, achieving a power conversion efficiency (PCE) with a determination coefficient (R²) of 0.86 and maintaining error fluctuations within a narrow range (±0.4 V). Key performance-determining parameters were identified using Pearson correlation and Shapley Additive Explanations (SHAP). SCAPS-1D simulations, guided by ANN-optimized inputs, yielded an optimal device architecture consisting of a 100 nm WS₂ ETL, an 800 nm RbTmCl₃ absorber, and a 500 nm CBTS HTL, resulting in a PCE of 28.21 %. The results highlight the efficacy of integrating ML with traditional simulation tools to accelerate PSC design and optimization. The findings also offer a solid basis for experimental validation and represent a significant step toward scalable, high-efficiency photovoltaic technologies.