Data-Driven Identification of an Interpretable Model for Diffusivity by Generating the Largest Amorphous Materials Database from Ab Initio Molecular Dynamic Calculations

All-solid-state batteries (SSB) are promising to address a portion of the increasing need for energy storage. Amorphous materials demonstrate unique advantages in tunability and processability in electrodes, electrolytes, and their interface coating materials. Investigating the relationship between ion conduction in amorphous materials can provide guidance in selecting materials. However, a comprehensive database is missing to enable such analysis. In this work, we generated the largest diffusivity database of amorphous materials to fill this gap using the AIMD workflow. The database covered a total of 4886 distinct compositions, and 3383 of them include Li. The amorphous structures are generated using the melt-quench procedure. The diffusivities of a subset of amorphous materials at four temperatures from 2500 K to 1000 K with 500 K intervals were calculated. We developed an analytical machine learning model from this diffusivity database using the sure independence screening and sparsifying operator (SISSO) method. The model can predict the temperature-dependent diffusion coefficient from compositional and structural features with high accuracy (R²=0.95). The model enables a rapid evaluation of diffusion in amorphous solids of different compositions, which can be used to screen suitable candidates for different parts in SSB.