The search for superionic solid-state electrolytes using a physics-informed generative model.

Area of Science:

• Materials Science
• Computational Chemistry
• Solid-State Chemistry

Background:

• Discovering superionic solid-state electrolytes for cation batteries is hindered by limited materials in databases.
• Generative AI shows promise for exploring new materials but struggles with stability criteria.

Purpose of the Study:

• To develop a physics-informed generative framework for efficient discovery of stable superionic solid-state electrolytes.
• To overcome limitations in current AI approaches for generating chemically valid and structurally stable material candidates.

Main Methods:

• Introduced a physics-informed hierarchical generative framework leveraging symmetry-aware crystallographic principles.
• Integrated empirical physical constraints and reinforcement learning with a hierarchical state representation.
• Proposed the symmetry-aware hierarchical architecture for flow-based traversal with density (SHAFT-density) model.

Main Results:

• Discovered new binary and ternary metastable phases with potential as solid-state electrolytes.
• Identified highly conductive LiBr, LiCl, Li2IBr, and Li3CBr2 materials.
• Demonstrated efficient exploration of the material search space, prioritizing stability and conductivity.

Conclusions:

• The SHAFT-density model successfully identifies stable, diverse, and potentially superionic compounds.
• The discovered materials offer promising candidates for next-generation solid-state electrolytes.
• This approach advances the development of advanced battery technologies through AI-driven materials discovery.