Materials Graph Networks Assisting the Discovery of New Solid-State Electrolyte Materials

Abstract

Finding novel solid-state electrolyte materials with specific desired properties is one of the main challenges of first principles-based modelling due to its high computational cost. Recently, machine learning (ML) has been used exclusively in the materials discovery field due to its remarkable capabilities to process large amounts of data and extract useful information insights. The implementation of ML into atomic-scale materials modeling can accelerated materials sampling with first principles accuracy, this should provide a short workflow and highly reduce the computational cost. In this work, we facilitate the ML model that can effectively screen for targeted properties from the entire chemical database possibility, including those materials that have never been examined. Currently, we are developing a materials graph networks framework for representing periodic crystal systems with the capability to learn atomistic chemical insights, especially for the discovery of new solid-state electrolyte materials. In this talk, we will discuss the basic theory and our recent work towards the development of ML model that can extensively discover and explore materials. We will highlight the computational-guided evolution approaches and screening high-performance Li conductor materials. Finally, we will end on discussing the future of ML-assisted materials discovery to the future of all-solid-state lithium batteries.