摘要
高通量计算与材料数据库推动了数据驱动的机器学习方法的发展.机器学习已经成为材料计算研究的重要方法,在分析材料数据、加速材料计算、预测材料性质、推进新材料发现、筛选和设计等方面展现出极大的潜力.众多与材料计算相交叉的机器学习方法、模型以及框架不断涌现.本文综述了近年来光电半导体材料计算设计领域内机器学习方法的最新进展与应用.介绍了机器学习的流程与类型,基于不同材料表示方法的浅层模型、集成模型和深度神经网络,以及相关材料数据库和相关工具.我们还讨论了这些模型在预测材料稳定性与光电性质、材料逆向设计、构建材料构效关系等方面的应用.最后,本文对目前机器学习方法存在的机遇与挑战,即数据数量与质量、材料的表示、材料逆向设计做了进一步总结与讨论.
Abstract
The development of high-throughput compu-tation and materials databases has laid the foundation for the emergence of data-driven machine learning methods in recent years.Machine learning has become a crucial methodology propelling researches in computational materials.It has de-monstrated tremendous potential in analyzing materials data,expediting materials calculations,predicting material prop-erties,advancing the discovery,screening,and design of new materials.Consequently,an increasing number of methodo-logies,models,and frameworks of machine learning have emerged.This review provides a comprehensive overview of the latest advancements and applications of machine learning in computational design of optoelectronic semiconductors.We introduce the workflow and strategies of machine learning shallow models,ensemble models,and deep neural networks based on various material representation methods.The asso-ciated material databases and toolkits are also discussed.Furthermore,we delve into the applications of these models in predicting material stability,optoelectronic properties,ma-terials inverse design,and establishing relationships between material structures and properties.Finally,we summarize and discuss the key challenges existing in current machine learn-ing,with a specific focus on issues related to the size of available data,data quality,material representation,and ma-terials inverse design.
NETL
NSTL
万方数据