Accelerate materials development using Machine learning: from Photovoltaic to other functional materials

Abstract

We have seen an urgent need to accelerate materials development during the COVID-19 pandemic. Materials development for vaccination could transform society. Meanwhile, to mitigate the adverse effects of global warming, cheaper, cleaner, and robust materials are needed to meet the IPCC 2030 targets. Leveraging upon the recent developments in high-performance computing (HPC), high-throughput experiment (HTE) and artificial intelligence (AI), this work aims to develop a methodological framework to accelerate materials development by combining theory, materials synthesis, and device optimization. The key enabler of this framework are machine learning methods that bridge different aspects of computation and experiment. Different from machine learning methods developed for computer vision or natural language processing, algorithms for materials need to handle sparse data that is scattered in various formats, leverage on the established physics knowledge, and efficiently converge to the best experimental conditions. This thesis addresses these challenges and develops various physics-informed algorithms.