Machine Learning Prediction of Organic–Inorganic Halide Perovskite Solar Cell Performance from Optical Properties

Abstract

Numerical toolsets have the potential to enable accelerated development of hybrid organic–inorganic halide perovskite (HOIP) solar cells. In the present study, we develop a machine learning (ML) approach that accurately predicts the current–voltage behavior of 3D/2D-structured (FAMA)Pb(IBr)₃/OABr HOIP solar cells under AM1.5 illumination. Using measured responses from 368 devices, we train a neural network (NN) with three optical inputs of constituent HOIP films (transmission, spectrally resolved photoluminescence, and time-resolved photoluminescence) to predict the solar cell’s open-circuit voltage (Voc), short-circuit current (Jsc), and fill factors (FF). The model achieves 91%, 94%, and 89% accuracy for 95% of Voc, Jsc, and FF predictions, with coefficient of determination (R²) values of 0.47, 0.77, and 0.58, respectively. By linking ML predictions to extracted physical parameters from the measured HOIP films optical properties, we identify key factors influencing the prediction results. Furthermore, we develop separate ML-classification algorithms that identify degraded solar cells with >90% accuracies using the same optical data. This work demonstrates an efficient, nondestructive approach for HOIP solar cell assessment that can assist in accelerating the next generation of perovskite solar cell developments.