Machine Learning Accelerated Study of Defect Energy Levels in Perovskites

Manipulatingthe defect tolerance is one of the effectivewaysto maintain the high power conversion efficiency and keep the stabilityof perovskite semiconductor materials. So, rapid screening for defectsand trap states in the perovskite semiconductor candidates is urgentlyneeded. Theoretical investigations of defects based on density functionaltheory (DFT) are still limited by their extremely high consumptionof computational resources and time. We implement an accelerated materialdiscovery approach using artificial intelligence and DFT, which canpredict the defect transition levels in the candidate perovskite semiconductormaterials. To verify the accuracy of our models, Cs3Sb2Br9 and Cs2SnBr6, which areout of the dataset that we used in machine learning (ML) model construction,are taken as examples. The extrapolation of ML prediction models andthe results given by DFT calculations are compared for defect transitionenergy levels. The two methods are consistent with each other withvery small errors. Our strategy avoids complex and time-consumingcomputational work based on DFT and provides quick and efficient screeningof physical properties with low cost.