Machine-learning-assisted discovery of perovskite materials with high dielectric breakdown strength

Abstract

In this paper, we have built a stepwise model based on the XGBoost machine learning algorithm to screen perovskite materials with high dielectric breakdown strength by comparing six machine learning algorithms. Here, the phonon cutoff frequency of perovskite materials can be predicted as an instrumental variable from features that can be commonly and easily found in materials databases. This prediction model shows outstanding performance and greatly reduces the amount of computation. Then we have screened borate perovskite materials with high dielectric breakdown strength such as LaBO₃, AlBO₃ and LuBO₃ from a database of 760 perovskite oxides. The results can help search for possible perovskite materials with high dielectric breakdown strength for application in dielectric capacitors.