2:45 PM - DS03.02.07
Tuning Optoelectronic Properties of Semiconductors with First Principles Modeling and Machine Learning
Arun Kumar Mannodi Kanakkithodi1,Maria Chan2,Xiaofeng Xiang3,Laura Jacoby3,Robert Biegaj3,Rishi Kumar4,David Fenning4
Purdue University1,Argonne National Laboratory2,University of Washington3,University of California, San Diego4
Show Abstract
Semiconductors with desirable electronic band structure and optical absorption are sought for solar cells, electronic devices, infrared sensors and quantum computing. Compositional manipulation via alloying at cation or anion sites, or via incorporation of point defects and impurities, can help tune the properties of semiconductors in known chemical spaces. In this work, we develop AI-based frameworks for the on-demand prediction and multi-objective optimization of the phase stability, band gap, optical absorption spectra, photovoltaic efficiency, dielectric constant, defect formation energies, and impurity energy levels in two broad classes of semiconductors, namely (a) halide perovskites with the general formula ABX3 (where A is a large organic or inorganic monovalent cation, B is a divalent cation and X is a halogen anion), and (b) group IV, III-V and II-VI semiconductors in binary, ternary and quaternary forms. These frameworks are powered by high-throughput density functional theory (DFT) computations, unique encoding of the atom-composition-structure information, and rigorous training of advanced neural network-based predictive and optimization models.
Bayesian optimization-based active learning approaches are applied to systematically improve prediction accuracies and comprehensively traverse the compositional chemical space. Multi-fidelity learning helps to bridge the gap between (high quantities of) low accuracy calculations and (lower quantities of) high-fidelity data, constituted of either accurate, expensive computations or experimental measurements collected from the published literature. High-accuracy predictions based on modest datasets are thus accomplished for (a) band gaps and formation energies at the HSE06 level of theory utilizing PBE-level data, (b) defect and impurity energy levels with experimental accuracy utilizing PBE-level data, (c) large supercell properties utilizing smaller cell calculations, and (d) properties of all alloy compositions utilizing data from end-point and selected mixed compositions. The best predictive models are combined with two different multi-objective optimization techniques, namely, state-of-the-art genetic algorithms and variational autoencoders, to determine optimal semiconductor atom-composition-structure combinations with desired stability, optoelectronic, and defect properties. Finally, AI-based recommendations are synergistically coupled with targeted synthesis and characterization, leading to successful validation and discovery of novel compositions for improved performance in solar cells.
References
1. A. Mannodi-Kanakkithodi, J. S. Park, N. Jeon, D. H. Cao, D. J. Gosztola, A. B. F. Martinson, M. K. Y. Chan, "Comprehensive Computational Study of Partial Lead Substitution in Methylammonium Lead Bromide", Chemistry of Materials 31 (10), 3599–3612 (2019).
2. A. Mannodi-Kanakkithodi, M. Toriyama, F. G. Sen, M. Davis, R. F. Klie, M. K. Y. Chan, "Machine learned impurity level prediction in semiconductors: the example of Cd-based chalcogenides", npj Computational Materials 6, 39 (2020).
3. A. Mannodi-Kanakkithodi, M. K. Y. Chan, "Computational Data-Driven Materials Discovery", Trends in Chemistry 3, 2, 79–82, (2021).
4. X. Xiang, L. Jacoby, A. Mannodi-Kanakkithodi, R. Biegaj, M. K. Y. Chan, "Universal Machine Learning Framework for Impurity Level Prediction in Group IV, III-V and II-VI Semiconductor", in preparation.
5. A. Mannodi-Kanakkithodi, R. E. Kumar, D. Fenning, M. K. Y. Chan, "Data-Driven Design of Novel Halide Perovskite Alloys", in preparation.