Chemical space-property predictor model of perovskite materials by high-throughput synthesis and artificial neural networks

Abstract

Lead-halide perovskites exhibit highly tunable optical properties, making them suitable for applications in photovoltaics and optoelectronics. Although considerable effort has gone into the development of methods that accurately predict the optical properties of perovskite materials based on structure, the reverse—predicting composition from optical data—is far less explored. In this study, high-throughput approaches were employed to synthesize and spectroscopically analyze a wide array of perovskites composed of mono-halide, di-halide, and tri-halides with a general formula, MA_xCs_1−xPb(Cl_xBr_yI_1−x−y)₃. The spectroscopic data were used to train an artificial neural network (ANN)-based chemical space-property predictor model designed to work with multiple responses and multiple predictors. The model predicted the chemical composition of perovskites from terahertz (THz) Raman spectroscopic data with approximately 85% accuracy. When the dataset also incorporated UV-visible spectroscopic data, the accuracy increased to nearly 92%. This study opens the possibility of real-time monitoring and defect detection, degradation analysis, and streamlined material selection and optimization of perovskite materials in industrial production.