Li diffusion in oxygen–chlorine mixed anion borosilicate glasses using a machine-learning simulation

Abstract

Lithium-ion conducting borate glasses are suitable for solid-state batteries as an interfacial material between a crystalline electrolyte and an electrode, thanks to their superior formability. Chlorine has been known to improve the electron conductivity of borate glasses as a secondary anion. To examine the impact of chlorine on lithium dynamics, molecular dynamics (MD) simulations were performed with a machine-learning interatomic potential (MLIP). The accuracy of the MLIP in modeling chlorine-doped lithium borate (LBCl) and borosilicate (LBSCl) glasses was verified by comparing with available experimental data on density, neutron diffraction S(q), and glass transition temperatures (Tg). While the MLIP-MD simulations underestimated the density when an isobaric–isothermal (NPT) ensemble was used, the glass models relaxed using the NPT ensemble after a melt-quench simulation employing a canonical (NVT) ensemble possessed reasonable density. The LBCl and LBSCl glass models exhibited increased lithium ion diffusion, and the ions were found to travel longer distances with an increase in the chlorine content. According to the structural analyses, it was observed that chlorine ions primarily interacted with lithium ions rather than the network formers. Consequently, lithium ions that interacted with a higher amount of chlorine showed a moderate increase in mobility. In summary, the MLIP demonstrated reasonable accuracy in modeling chlorine-containing borate glasses and enabled the investigation of the effect of chlorine on electron conductivity. In contrast, the first sharp diffraction peaks in S(q) deviated from the experimental diffractions, suggesting that additional efforts are required to accurately model the middle-range structure of the glasses.