Superior Energy-Storage Performance Enabled by Machine Learning Accelerated Composition Design for Lead-Free...

Area of Science:

• Materials Science
• Ceramics Engineering
• Computational Materials Science

Background:

• High-performance lead-free AgNbO3 (AN)-based antiferroelectrics (AFEs) are crucial for advanced energy-storage capacitors.
• Traditional trial-and-error methods for optimizing energy storage density (Wrec) and efficiency (η) are time-consuming and inefficient due to vast composition spaces.

Purpose of the Study:

• To develop a data-driven methodology for designing high-performance AN-based AFEs.
• To overcome the limitations of conventional screening methods using machine learning.
• To identify novel compositions with superior energy storage capabilities.

Main Methods:

• Implementation of a two-layer stacking machine learning (ML) framework, SS-PAN (stacking strategy for predicting AN-based ceramics).
• Cross-validation to assess model performance, achieving a high R² score of 0.82.
• SHapley Additive exPlanations (SHAP) analysis to identify key predictive features (tolerance factor, B-site electron affinity).
• Experimental validation using STEM and DFT calculations to decode local structure.

Main Results:

• The SS-PAN framework outperformed individual ML models in predicting AN-based AFEs.
• A predicted composition, Li0.01Ag0.99Nb0.5Ta0.5O3, demonstrated a quasi-linear P-E loop.
• Achieved ultrahigh Wrec of 16.6 J cm⁻³ and η of 92.6% at 108 kV mm⁻¹ in MLCC, with a figure of merit of 224.3 J cm⁻³.
• SHAP analysis highlighted the importance of tolerance factor and B-site electron affinity for prediction accuracy.

Conclusions:

• The data-driven SS-PAN framework offers a potent approach for discovering emergent relaxor AFEs.
• Li/Ta co-doping induces highly polar short-range antiferroelectric nanodomains, enhancing dielectric properties.
• This work paves the way for advanced dielectric capacitor applications through ML-guided materials design.