Process optimisation for NASICON-type solid electrolyte synthesis using a combination of experiments and bayesian optimisation

Na superionic conductor (NASICON)-type LiZr 2 (PO 4 ) 3 (LZP) is an oxide-based solid electrolyte candidate for use in all-solid-state Li-ion batteries. However, as the ionic conductivity is insufficient, doping with aliovalent cations has been carried out to improve the Li-ion conductivity by controlling the composition and crystal structure. Li-ion conductivity is also affected by the microstructural properties of a sintered body, such as density, morphology, and elemental distribution, and thus, controlling process parameters, such as heating conditions during the solid-state reaction, improves conductivity. Using an exhaustive experimental approach, Ca and Si co-doped Li-rich NASICON-type LZP was synthesised via solid-state reactions under various two-step heating conditions to yield the highest Li-ion conductivity by optimising the conditions. The highest total Li-ion conductivity of 3.3 × 10 −5 S cm −1 was obtained when the sample was first heated at 1050 °C and then heated at 1250 °C. The crystal structures, relative densities, micromorphologies, and Li-ion conductivities of the materials were characterised, and their relationships were investigated. These relationships were complex, and intuitively determining the optimal conditions was challenging with only a few experiments. Instead, as a proof-of-concept study, the collected data were used to demonstrate that Bayesian optimisation (BO) efficiently improved the experimental determination of the optimal heating conditions. The BO-guided experimental investigation determined the optimal conditions more rapidly compared to conventional trial-and-error approaches employed in the materials industry. The efficiency factor was approximately double that of the exhaustive search. The optimal sintering conditions for LiZr 2 (PO 4 ) 3 , a NASICON-type solid electrolyte with high conductivity, were explored. It was also found that these optimum sintering conditions could be efficiently discovered by using Bayesian optimisation.