Machine Learning-Assisted Crystal Structure Prediction of Solid-State Electrolytes Reveals Superior Ionic...

Area of Science:

• Materials Science
• Solid-State Chemistry
• Computational Materials Science

Background:

• Developing novel solid-state electrolytes (SSEs) with high ionic conductivity is crucial for advancing all-solid-state batteries (ASSBs).
• Existing research often prioritizes compositional changes over the impact of intrinsic crystal structures on ion transport.
• Understanding structure-property relationships is key to designing efficient SSEs.

Purpose of the Study:

• To introduce a theoretical crystal structure prediction (CSP) approach using machine learning for discovering novel SSEs.
• To investigate the influence of crystal structure on Li-ion transport properties in promising SSE candidates.
• To identify design principles for high-performance SSEs based on structural characteristics.

Main Methods:

• Utilized machine-learning moment tensor potential (MTP) for theoretical crystal structure prediction (CSP).
• Employed a phase-diagram-guided strategy to apply CSP to Li-ion conducting materials (Li₂SiS₃, Li₂GeS₃, Li₄SiGeS₆, Li₄SiSnS₆).
• Analyzed polyhedral connectivity, relative stability, Li-ion accessible volume (packing ratio), and sublattice distortion.

Main Results:

• The CSP approach successfully identified novel SSE structures and reproduced known experimental ones.
• Metastable edge-sharing SSE phases demonstrated superior Li-ion mobility compared to stable corner-sharing phases.
• Enhanced conductivity in metastable phases correlates with higher packing efficiency, larger Li-S₄ sublattice volume, and greater dynamic distortion.

Conclusions:

• Crystal structure plays a fundamental role in determining Li-ion transport in SSEs.
• The CSP method is a powerful tool for designing novel SSEs with tailored properties.
• This work provides insights into designing high-performance ASSBs by controlling SSE crystal structures.