Deep learning-enabled modeling of FA1-xCsxSnI3 solar cells: Impact of cesium composition and temperature on device efficiency

Abstract

This work presents a comprehensive framework that integrates SCAPS-1D numerical simulations with deep learning (DL) techniques to investigate and optimize the performance of lead-free FA_1-xCs_xSnI₃ perovskite solar cells (PSCs). The study focuses on the effects of varying cesium (Cs) mole fraction and operating temperature on key photovoltaic parameters, including power conversion efficiency (PCE), open-circuit voltage (Voc), short-circuit current density (Jsc), and fill factor (FF). A dataset comprising over 500 simulated device configurations was generated to capture the influence of multiple structural and environmental factors. A long short-term memory (LSTM)-based DL model was employed to classify device performance and identify the most critical parameters through feature importance analysis. The results revealed that the electron transport layer (ETL) had the strongest influence on overall efficiency, followed by HTL thickness, perovskite bandgap, and ETL thickness. Optimization showed that incorporating Cs at a mole fraction of 0.15 improved PCE from 15.3 % to 18.7 %, with corresponding enhancements in Voc (0.83 V–0.89 V), Jsc (22.1–23.8 mA/cm²), and FF (72.3 %–79.4 %). These findings highlight the synergistic role of compositional tuning and interfacial engineering in boosting PSC performance. The proposed DL-SCAPS framework offers a powerful tool for guiding the design of efficient, stable, and eco-friendly perovskite solar cells, which could be applied in flexible photovoltaics, building-integrated solar panels, and portable power generation systems.