First-Principles Calculations and Machine Learning Accelerated Superionic Conductivity Predictions for Novel Anti-Fluorite Solid Electrolytes

The development of solid-state electrolytes (SSEs) with high ionic conductivity is pivotal for the advancement of next-generation energy storage technologies, particularly solid-state batteries. In a previous study, we leveraged machine learning (ML) techniques to predict the ionic conductivities of nearly 8,000 known lithium-containing compounds, identifying approximately 30 promising candidates for lithium superionic conduction. Notably, these candidates encompass not only well-established crystal structure types for SSEs (e.g., LISICON, garnet, NASICON, perovskite) but also several anti-fluorite structures, a class of materials not traditionally associated with high ionic conductivity.<br/> <br/>The primary objective of the present research is to rigorously assess the ionic conductivity of these candidate materials through first-principles calculations. While ab initio molecular dynamics (AIMD) simulations have become the gold standard for calculating ionic conductivity, their computational intensity and time demands pose significant challenges when evaluating extensive candidate lists generated by data-driven material discovery approaches.<br/> <br/>To overcome this bottleneck, we have adopted machine learning force fields (MLFFs) to accelerate the prediction of ionic conductivity. MLFFs offer the potential to achieve reliable result comparable to AIMD while substantially reducing computational overhead. By utilizing the on-the-fly AIMD training method within the Vienna Ab initio Simulation Package (VASP), we have efficiently generated MLFFs tailored for predicting the ionic conductivity of our candidate materials.<br/> <br/>The results of our investigation indicate that the novel anti-fluorite structures identified in our candidate shortlist exhibit superionic conductivity behavior around 600-700 K. This contrasts with conventional anti-fluorite compounds such as Li₂S and Li₂Se, which exhibit superionic conductivity at temperatures above 1000 K. This observation, derived from our MLFF-based theoretical calculations, not only validates the efficacy of our data-driven material prediction methodology but also opens up new avenues for the design and development of SSEs. By demonstrating the potential of anti-fluorite structures for achieving high ionic conductivity at or near room temperature, our research contributes to the ongoing quest for materials that can enable safe, high-performance solid-state batteries.<br/> <br/>Furthermore, the successful integration of MLFFs into our computational workflow highlights the synergistic potential of combining machine learning and first-principles calculations for accelerated material discovery. This approach holds promise for expediting the identification and optimization of materials with tailored properties across a wide range of technological applications. Future work will focus on refining our workflow, exploring the broader chemical space of those compounds we validated, and experimentally validating the predicted ionic conductivities of our most promising candidates. This work was supported by JST Grant Number JPMJPF2016.