AI-Guided Understanding and Design of Solid-Solid Interfaces in Batteries

Solid state electrolytes have potential to widely replace liquid electrolytes in rechargeable Li- and Na-ion batteries due to improved safety and energy density. In addition to ionic conductivity and electrochemical stability window, stability with the anode and the cathode is of utmost importance for a potential solid electrolyte. Therefore, it is necessary to predict and determine phase decomposition or unknown phase formation at the electrode electrolyte interfaces. Using AI-guided generative modeling, as implemented in FANTASTX, we sample structures from the relevant chemical system and compare their stability and match to experimental data, where available. We also compare the structures generated by FANTASTX using sampling methods such as genetic algorithm and basin-hopping in the original and transformed structure spaces where the latter is obtained using neural-network based models such as variational auto encoders (VAEs). Finally, we discuss ML-based approaches which allows us to detect impurity phases and bonding from core-level spectroscopy.