Optimization of Lead‐Free Perovskite Solar Cell Architecture Using Machine Learning and Numerical Simulations

Abstract The proposed perovskite device structure considers several factors to realize their significance on device performance. Initially, the PCE variation between the two absorber halides is investigated, yielding a maximum PCE of 24.31% for CH3NH3SnBr3 and 27.37% for CH3NH3SnI3. Additionally, the SCAPS‐1D simulation assesses the contribution of distinct HTMs and ETMs. By further optimizing these layers along with diverse intrinsic parameters, the device's PCE increased from 27.37% to 40.17%. To improve predictive capabilities, a dataset of 29565 is generated utilizing the SCAPS‐1D simulator for CH3NH3SnI3‐based solar cells. Data preprocessing in Python applied leakage‐safe Pearson correlation filtering: within each highly collinear group (|r| ≥ 0.90), one representative predictor is retained and the remainder are excluded to reduce multicollinearity and improve interpretability. Six machine learning models are tested, and Random Forest is validated to be the most credible performer with an R2 of 96% and an RMSE of 0.210. The optimized configuration — FTO/WS 2 (ETL)/CH 3 NH 3 SnI 3 (absorber)/V 2 O 5 (HTL)/Pt (back contact) — achieves a record simulated efficiency of 40.17%, surpassing prior reports. This performance is attributed to WS 2 ’s favorable band alignment, CH 3 NH 3 SnI 3 ’s strong absorption, and V 2 O 5 ’s stability. The combined SCAPS–ML framework not only accelerates optimization but also provides actionable design rules for environmentally sustainable, lead‐free PSCs.