Efficient Experimental Search for Discovering a Fast Li-Ion Conductor from a Perovskite-Type LixLa(1–x)/3NbO3 (LLNO) Solid-State Electrolyte Using Bayesian Optimization

LixLa(1–x)/3NbO3 (LLNO) is an A-site-deficient perovskite material that has a larger unit cell volume, a lower La3+ concentration, and a higher intrinsic vacancy concentration than (LixLa(2–x)/3TiO3), which is known to be one of the fastest Li-ion conductive oxides. These advantages make LLNO a potential oxide-based solid electrolyte candidate for all-solid-state Li-ion batteries. The A-site and B-site elements in this perovskite-type material can be substituted by ions with various charges and radii in a wide range of ways to form complicated solid solutions; hence, this type of material can be adapted to a variety of application requirements. Doping with monovalent or divalent metal compounds is a promising method for improving the ionic conductance of this perovskite-type material. In this study, the (LiyLa(1–y)/3)1–xSr0.5xNbO3 (0 ≤ 0.5 x ≤ 0.15, 0 ≤ y ≤ 0.3) composition formed by co-doping with Li2CO3 and SrCO3 was optimized using an exhaustive experimental approach. Sixty-four samples with different compositions were structurally analyzed, and their electrochemical performance was experimentally characterized, which revealed that the co-doped samples have higher ionic conductivities and superior sintered morphologies compared to those prepared by single doping. Because Li+ and Sr2+ doping improves the ionic conductivity for different reasons, and many factors, such as higher carrier concentrations, enhancements through sintering, and changes in the microstructure, play important roles, it is difficult or inefficient to determine the best composition using only traditional trial-and-error or intuitive searching. Instead, as a proof-of-concept study, we show that the Bayesian optimization (BO) method efficiently searches for the best composition and that material retrieval during experimental exploration can benefit from BO because it significantly reduces the high workload associated with the trial-and-error approach employed by the materials industry.