Machine learning-enabled chemical space exploration of all-inorganic perovskites for photovoltaics

Abstract

The vast compositional and configurational spaces of multi-element metal halide perovskites (MHPs) result in significant challenges when designing MHPs with promising stability and optoelectronic properties. In this paper, we propose a framework for the design of B-site-alloyed ABX3 MHPs by combining density functional theory (DFT) and machine learning (ML). We performed generalized gradient approximation with Perdew–Burke–Ernzerhof functional for solids (PBEsol) on 3,159 B-site-alloyed perovskite structures using a compositional step of 1/4. Crystal graph convolution neural networks (CGCNNs) were trained on the 3159 DFT datasets to predict the decomposition energy, bandgap, and types of bandgaps. The trained CGCNN models were used to explore the compositional and configurational spaces of 41,400 B-site-alloyed ABX3 MHPs with a compositional step of 1/16, by accessing all possible configurations for each composition. The electronic band structures of the selected compounds were calculated using the hybrid functional (PBE0). Then, we calculated the optical absorption spectra and spectroscopic limited maximum efficiency of the selected compounds. Based on the DFT/ML-combined screening, 10 promising compounds with optimal bandgaps were selected, and from among these 10 compounds, CsGe0.3125Sn0.6875I3 and CsGe0.0625Pb0.3125Sn0.625Br3 were suggested as photon absorbers for single-junction and tandem solar cells, respectively. The design framework presented herein is a good starting point for the design of mixed MHPs for optoelectronic applications.