Prediction of Efficiency for KSnI3 Perovskite Solar Cells Using Supervised Machine Learning Algorithms

Machine learning possesses enormous capability for accelerating materials research. A dataset of 40,845 data points, each containing 52 features for KSnI₃-based perovskite solar cells (PSCs), was curated in the present study for the first time. This dataset was generated by varying the concentration of defects at the layers and interfaces, thickness, doping density, work function of back contacts, series resistance, temperature, and shunt resistance for various combinations of inorganic and organic charge transport layers (CTLs) for a KSnI₃-based PSC. Various supervised machine learning regression algorithms were applied to the curated dataset to predict the power conversion efficiency (PCE) of the PSC, and the random forest regression (RFR) algorithm was found to provide the lowest error out of all the trained models. The RFR was then utilized to predict the PCE of the PSC based on KSnI₃, using SrTiO₃ and NiO as CTLs, with varying concentrations of defects and dopants and thickness of the layers. The predicted values were found to be in good agreement with the true values. The machine learning model and the dataset provided in the present study will not only aid in the selection of optimal CTLs but also help in the optimization of the PSC structure.