Machine Learning-Assisted Bayesian Optimization for the Discovery of Effective Additives for Dendrite Suppression in Lithium Metal Batteries

Abstract

In the pursuit of enhancing the performance and safety of lithium (Li)-metal batteries, the discovery of effective electrolyte additives to suppress Li dendrites has emerged as a paramount objective. In this study, we employ an inverse design strategy to identify potential additives for dendrite mitigation. Two key mechanisms, namely, the formation of robust solid electrolyte interphase layers and the leveling mechanism, serve as the foundation for our investigation. Our inverse design strategy is guided by molecular properties such as the lowest unoccupied molecular orbital energy and interaction energy upon Li surface adsorption. An active learning process utilizing Bayesian optimization (BO) was utilized to identify potential molecules with ideal properties. Through this screening process, we uncover a collection of 62 molecules with the potential to act as SEI-forming additives, along with 106 molecules for leveling additives, both surpassing the performance of established additives reported in the literature. This work highlights the potential of BO methods in computationally based inverse design of materials for many applications, and the discovered additives could potentially boost the commercialization of Li–metal batteries.