Compositionally-informed machine learning for solid-state electrolyte design: a structure-free approach.

Area of Science:

• Materials Science
• Computational Chemistry
• Energy Storage

Background:

• Next-generation energy storage requires advanced solid-state electrolytes (SSEs).
• Machine learning (ML) is a promising tool for designing new materials.
• Current ML methods often need atomic positions, limiting early-stage design when structural data is unavailable.

Purpose of the Study:

• To develop a structure-free ML framework for predicting SSE ionic conductivity.
• To address challenges in composition-based ML, such as capturing ionic size effects and compositional complexity.
• To accelerate the discovery of high-performance SSEs for energy storage applications.

Main Methods:

• Introduced a compositionally-informed ML (CI-ML) framework utilizing ionic radius mismatch (IonicRad_Mis) as a key descriptor.
• Employed a compositionally-stratified modeling strategy with an XGBoost model.
• Applied SHAP analysis to identify descriptor importance and trends.

Main Results:

• The CI-ML model achieved robust accuracy on global and stratified datasets (e.g., R² of 0.707 on test data).
• Ionic radius mismatch was identified as the most significant descriptor, negatively correlating with ionic conductivity.
• Designed and synthesized a Br-substituted Li₃InCl₆ SSE, demonstrating a conductivity increase from 0.88 to 1.30 mS cm⁻¹.

Conclusions:

• The structure-free CI-ML approach effectively overcomes the limitations of structure-dependent ML models.
• This framework enables accelerated early-stage discovery of SSEs and other functional materials.
• Ionic radius mismatch is a critical factor for designing high-performance solid-state electrolytes.