Prediction of band gap and optimum electrical parameters of a thin homojunction perovskite solar cell based on FA1−xCsxSnyPb1−yI3 through a combination of SCAPS-1D and machine learning based modelling

Abstract

This paper presents a machine learning (ML) model developed to predict the band gap and optimum electrical parameters of a thin homojunction perovskite solar cell (PSC) based on FA1−xCsxSnyPb1−yI3. For different cesium (x) and tin (y) compositions, band gaps are tabulated, and electrical parameters such as open circuit voltage, short circuit current, and power conversion efficiency are generated in standard test condition (STC) using the SCAPS-1D platform. The data obtained are used to build two sub-models wherein first Model 1 uses the composition to predict the band gap and second Model 2 uses the predicted band gap to predict the electrical parameters using a polynomial regression algorithm. The sub-models are then connected in series. Train and test scores, correlation (r), and root mean square error (RMSE) are calculated to investigate the accuracy of the models while the cross-validation score checks the stability and the overfitting of the models. The final, best performance model’s metrics–R2, r, and RMSE, are 0.99, 0.99, and 0.09 respectively for Model 1 and 0.97, 0.99, and 0.28 for Model 2. These are achieved for a fourth order polynomial. In addition, we realize that the homojunction PSC based on FACsSnPbI3 offers a high efficiency of 17.17% for a band gap of around 1.44 eV. We also reveal that the presence of tin induces lattice contraction, affects the stability, and changes the configuration of the crystal structure from cubic to octahedral tilting when tin composition increases.