Bayesian optimisation with transfer learning for NASICON-type solid electrolytes for all-solid-state Li-metal...

Area of Science:

• Materials Science
• Electrochemistry
• Solid-State Chemistry

Background:

• NASICON-type LiZr₂(PO₄)₃ (LZP) is a promising solid oxide electrolyte for all-solid-state batteries due to its high ionic conductivity and electrochemical stability.
• Improving the Li-ion conductivity of LZP is crucial for enhancing battery performance.
• Doping LZP with CaO and SiO₂ (Li_1+x+yCa_xZr_2-ySi_yP_3-yO₁₂, LCZSP) offers a route to enhanced conductivity.

Purpose of the Study:

• To enhance the Li-ion conductivity of Li-rich NASICON-type LZP by doping with CaO and SiO₂.
• To efficiently identify the optimal composition of LCZSP with the highest Li-ion conductivity.
• To demonstrate the efficacy of Bayesian optimization with transfer learning for materials discovery.

Main Methods:

• Synthesis and characterization of 49 LCZSP compositions, evaluating crystal structure, relative density, and Li-ion conductivity.
• Application of Bayesian optimization (BO) to efficiently search for optimal compositions.
• Implementation of BO with transfer learning, utilizing prior knowledge from Li_1+x+yY_xCa_yZr_2-x-yP₃O₁₂ (LYCZP) compositions.

Main Results:

• Simultaneous doping with Ca and Si at Zr and P sites improved Li-ion conductivity.
• Bayesian optimization with transfer learning successfully identified optimal LCZSP compositions.
• The transfer learning approach accelerated the discovery of optimal compositions by a factor of two compared to conventional BO.

Conclusions:

• Bayesian optimization with transfer learning is a highly effective and efficient method for discovering optimal compositions of functional materials.
• This approach significantly accelerates the materials discovery process for solid oxide electrolytes.
• The methodology is broadly applicable to the optimization of various functional materials and ionic conductors.