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 existing deep 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, for any element from across the periodic table treated as a point defect or impurity, in (a) ABX₃ halide perovskites with a selected set of A, B and X species, and (b) zincblende group IV, III-V and II-VI binary and ternary semiconductors. ML models utilize unique encoding of the defect atom’s elemental properties, coordination environment, and cheaper unit cell defect DFT properties, along with rigorous training using random forests, Gaussian processes, and neural networks. We also apply multi-fidelity learning to increase the accuracy of predictions at a high level of theory, namely the HSE06 hybrid functional, employing large amounts of lower accuracy/expense theory (the GGA-PBE functional) data and modest amounts of HSE06 data, to make defect level predictions with near-experimental accuracy and lower uncertainty. Accelerated screening of optoelectronically-active functional impurities is thus performed, and useful lists of dopants that may induce n-type or p-type conductivity shifts in the semiconductor are generated. The extensive DFT datasets and best ML models are made available as online tools for easy prediction and screening across massive semiconductor-defect chemical spaces.<br/> <br/><b>References</b><br/>1. A. Mannodi-Kanakkithodi et al., "Comprehensive Computational Study of Partial Lead Substitution in Methylammonium Lead Bromide", Chemistry of Materials 31 (10), 3599–3612 (2019).<br/>2. A. Mannodi-Kanakkithodi et al., "Machine learned impurity level prediction in semiconductors: the example of Cd-based chalcogenides", npj Computational Materials 6, 39 (2020).<br/>3. A. Mannodi-Kanakkithodi et al., "Defect Energetics in Pseudo-Cubic Mixed Halide Lead Perovskites from First-Principles", Journal of Physical Chemistry C. 124, 31, 16729–16738 (2020).<br/>4. F. G. Sen et al., "Computational Design of Passivants for CdTe Grain Boundaries", Solar Energy Materials and Solar Cells 232, 111279 (2021).<br/>5. A. Mannodi-Kanakkithodi et al., "Universal Machine Learning Framework for Impurity Level Prediction in Group IV, III-V and II-VI Semiconductors", <i>under review</i>. PREPRINT: https://doi.org/10.21203/rs.3.rs-723035/v1 (2021).<br/>6. A. Mannodi-Kanakkithodi et al., "Accelerated Screening of Functional Atomic Impurities in Halide Perovskites using High-Throughput Computations and Machine Learning", <i>under review</i>.