Machine Learning Defect Properties of Semiconductors

Defects and impurities in semiconductors can reduce photovoltaic absorption via nonradiative recombination of charge carriers or enhance absorption via intermediate bands. Defect levels in the band gap may also be used as qubits for quantum computing. Quick and accurate predictions of defect properties are thus desired in technologically important semiconductors, but are complicated by difficulties in sample preparation and assigning measured levels to specific defects, as well as by the expense of large-supercell first principles computations that involve charge corrections and advanced functionals. In this work, we address this issue by combining high-throughput density functional theory (DFT) with machine learning (ML) to develop predictive models for defect formation energy and charge transition levels in (a) ABX₃ halide perovskites, and (b) zincblende group IV, III-V and II-VI semiconductors. ML models utilize unique encoding of the defect atom’s elemental properties, coordination environment, and cheaper DFT properties, along with rigorous training using random forests, Gaussian processes, and neural networks. The extensive DFT datasets and best ML models are made available as online tools for easy prediction and screening across large semiconductor-defect chemical spaces.