Machine Learning Prediction of Ionic Conductivity and Binding Energy in Doped Li₂ZrCl₆ Halides for Solid-State...

Area of Science:

• Materials Science
• Computational Chemistry
• Electrochemistry

Background:

• All-solid-state batteries require advanced solid electrolytes for improved safety and performance.
• Ternary halide compounds show promise for fast ion conduction, but their discovery is challenging.
• Efficient methods are needed to accelerate the identification of novel solid electrolyte materials.

Purpose of the Study:

• To accelerate the discovery of fast-ion-conducting halide solid electrolytes using machine learning and DFT.
• To predict ionic conductivity and binding energy of halide compounds.
• To explore substitutions and dopants for enhancing electrolyte properties.

Main Methods:

• Constructed a halide compound database from literature and merged it with an existing inorganic solid-state electrolyte database.
• Employed supervised machine learning models integrated with density functional theory (DFT) calculations.
• Utilized interpretable elemental features like Element Fraction and Element Property descriptors for model training.

Main Results:

• Achieved excellent predictive performance for ionic conductivity (MAE = 0.6012, R² = 0.8062) and binding energy (MAE = 0.0314, R² = 0.9957).
• Explored divalent and trivalent substitutions in Li₂₊₂ₓZr₁₋ₓMe²⁺ₓCl₆ and Li₂₊ₓZr₁₋ₓMe³⁺ₓCl₆ systems.
• Proposed Ti³⁺ and Sn²⁺ as dopants to enhance ionic conductivity and structural stability in halide electrolytes.

Conclusions:

• Machine learning, combined with DFT, is a powerful tool for efficiently discovering new solid-state battery electrolyte compositions.
• Easily obtainable elemental descriptors are crucial for developing effective machine-learning approaches in materials science.
• This study demonstrates a pathway to accelerate the development of next-generation solid electrolytes for all-solid-state batteries.