Machine Learning-Driven Optimization of Transport Layer Parameters in CsSn 0.5 Ge 0.5 I 3 Perovskite Solar Cells

Perovskite solar cells (PSCs) offer high efficiency and low cost compared to conventional photovoltaics, but their lead content raises significant environmental and health concerns, limiting commercial potential. This work explores optimising the performance of lead-free CsSn₀.₅Ge₀.₅I₃-based PSCs by adjusting the hole transport layer (HTL) and electron transport layer (ETL) properties. To this end, machine learning models, i.e., RF, ANN, CatBoost, and XGBoost, were trained to estimate the photovoltaic performance of the solar cells. This concept potentially minimizes the time and resources required during the experiment compared to traditional methods. We used the SCAPS-1D simulation software to simulate the PV performance of the perovskite configuration, PCBM/CsSn₀.₅Ge₀.₅I₃/Spiro-MeOTAD, across different ETL and HTL mobilities, ranging from 0.001 cm²/Vs to 0.2 cm²/Vs and 0.0001 cm²/Vs to 0.0021 cm²/Vs, respectively, and carrier concentrations ranging from 10¹⁵ cm⁻³ to 10¹⁹ cm⁻³ to generate a dataset. Based on these variations, a dataset of 3025 performance points, including JSC, FF, VOC, and PCE, was generated and subsequently used to train and validate the ML models. The models' predictions were compared with the SCAPS-1D results, and both Catboost and XGboost better represent the simulated data, as reflected by lower RMSE values of 0.01 and 0.02 and higher R² values of 0.9999. Sensitivity analysis using SHAP plots indicated that ETL and HTL mobility, along with carrier concentrations, significantly influenced PCE, FF, JSC, and VOC, with mobility showing a consistent and dominant effect across all parameters. Machine learning accelerates Cs-based PSC optimization, providing a faster alternative to time-consuming device simulations.