Experimental data-driven efficient exploration of the composition and process conditions of Li-rich NASICON-type solid electrolytes

Abstract

LiZr₂(PO₄)₃ has garnered widespread interest as a solid electrolyte for all-solid-state batteries. However, its Li ionic conductivity remains insufficient for practical use. Although attempts have been made to improve the Li ionic conductivity by doping with cations and controlling the synthesis conditions, the exploration space is vast, and optimisation remains challenging. In this study, the amount of dopants and heating conditions for Li_1+x+2yCa_yZr_2-ySi_xP_3-xO₁₂ co-doped with Ca²⁺ and Si⁴⁺ were optimised via experimental synthesis, evaluation, and Bayesian optimisation (BO) cycles. The BO technique suggests the next experimental samples in each cycle and reduces the number of experimental cycles by almost 80 % compared with an exhaustive search. In addition, the experimental results were subjected to machine-learning regression analysis to analyse the factors affecting the Li-ion conductivity.