Exploring the Possibility of Machine Learning for Predicting Ionic Conductivity of Solid-State Electrolytes.

Area of Science:

• Materials Science
• Electrochemistry
• Computational Chemistry

Background:

• Solid-state electrolytes (SSEs) offer enhanced safety and performance for all-solid-state lithium-ion batteries (ASSBs) compared to conventional liquid electrolytes.
• However, challenges like low ionic conductivity and interface instability hinder SSE development.
• Discovering new SSEs through traditional methods is time-consuming and resource-intensive.

Purpose of the Study:

• To develop a machine learning (ML) based architecture for predicting the ionic conductivity of SSEs.
• To identify key material characteristics influencing ionic conductivity.
• To accelerate the discovery of novel SSEs for ASSBs.

Main Methods:

• Utilized machine learning to predict ionic conductivity based on features like activation energy, temperature, lattice parameters, and unit cell volume.
• Employed ensemble-based predictor models, including stacked models, to enhance prediction accuracy and mitigate overfitting.
• Split the dataset into 70:30 ratios for training and testing eight predictor models.

Main Results:

• Developed an ML architecture capable of predicting SSE ionic conductivity with high accuracy.
• Identified distinct patterns in SSE characteristics using a correlation map.
• Achieved low prediction errors, with the Random Forest Regressor model showing maximum mean squared error of 0.001 and mean absolute error of 0.003.

Conclusions:

• Machine learning provides an effective and reliable tool for screening and discovering new SSE materials.
• The developed ML model can significantly accelerate the identification of SSEs with improved ionic conductivity for ASSBs.
• Ensemble and stacked models offer robust predictions, addressing challenges in SSE material development.