Advanced Computational Techniques for Optimizing Manganese-Based Perovskite Solar Cells: From SCAPS-1D Simulations to Machine Learning Predictions

Conventional perovskite materials face several limitations in the design of solar cell stability and toxicity for the environment due to lead (Pb) content. To overcome the toxicity, Pb has been replaced by manganese (Mn) in this work. In this context, NH₂(CH₂)₂NH₃MnCl₄ as a lead-free perovskite absorber has been explored owing to its low exciton binding energy, high carrier mobility, and tunable bandgap. With the help of the SCAPS 1d simulator, the doping (Nd), bulk defect density (BDD), and thickness of the absorber layer have been varied. The optimized champion cell obtained a power conversion efficiency (PCE) of 17.70%, with a J_SC of 13.86 mA/cm², V_OC of 1.45 V, and fill factor of 84.60%. Further, different ML models, including Random Forest, XGBoost, Linear Regression, and Support Vector Regression, have been applied to predict the PCE from a dataset of 2400. The XGBoost model outperformed the other ML models, achieving an MSE of 0.05 and an R² of 99.73%, demonstrating superior accuracy in the prediction of the PCE of the cell. These results demonstrate the effectiveness of integrating novel materials with machine learning techniques to accurately predict solar cell performance, offering a more efficient and less computationally demanding alternative to traditional simulation methods.