Novel fast Li-ion conductors for solid-state electrolytes from a fine-tuned foundational model

We present a high-throughput computational screening for fast lithium-ion conductors to be used as solid-state electrolytes. Starting with more than 30,000 experimentally reported crystal structures sourced from the COD, ICSD and MPDS repositories, we perform highly automated calculations using AiiDA, at the level of density functional theory to identify electronic insulators. On these ~1000 materials, we use molecular dynamics simulations to estimate Li-ion diffusivities using a universal machine learning interatomic potential called PET-MAD [1], extensively fine-tuned on first-principles data of Li-containing materials. This work revisits and expands a previous screening that used the pinball model, to run fast and efficient MD simulations, leading to the discovery of 7 new fast Li-ion conductors [2]. To our knowledge, this is the first screening of this magnitude using a foundational model. For the most promising, previously unreported fast conductors we perform MD simulations at several temperatures to estimate activation barriers. We further present the entire screening protocol, along with the finetuning methodology.

[1] A. Mazitov, M. Ceriotti et al; arXiv:2503.14118v2

[2] T. S. Thakur, L. Ercole and N. Marzari; submitted to Energy and Environmental Science (2025)