High-Throughput Computational Screening and Machine Learning Modeling of Janus 2D III–VI van der Waals Heterostructures for Solar Energy Applications

Two-dimensional Janus III–VI monolayers and corresponding van der Waals (vdW) heterostructures present immense application potential in the solar energy conversion areas. In this work, we present material screening and machine learning modeling to accelerate the discovery of promising photocatalytic and photovoltaic candidates in Janus III–VI vdW heterostructures. A comprehensive database with a total of 19926 heterostructures has been proposed according to the high-throughput first-principles calculations. It is highlighted that we develop an accurate machine learning model using only atoms and bonds as descriptors based on the crystal graph convolutional neural network framework. Besides, 1035 Janus III–VI vdW heterostructures have been screened out according to the essential criteria of stability. Moreover, we find 66 and 71 potential candidates for photocatalysis and solar cells, respectively, from further application-driven screening. Interestingly, the screened type-II SeInAlS/SeGaAlTe heterostructure with a band gap of 1.18 eV is highlighted as an internal electric field-driven asymmetrical photocatalyst. On the other hand, the type-II Al2STe/Al2SSe heterostructure solar cell presents power conversion efficiencies higher than 21% both from the microscopic and mesoscopic point of views. We believe that our study will provide a feasible strategy for the design of III–VI monolayers for solar energy applications.