Building AI Foundation Models to Accelerate the Discovery of New Battery Materials - CleanTechnica

Researchers at the University of Michigan, leveraging the powerful supercomputers Aurora and Polaris at the Argonne Leadership Computing Facility (ALCF), are developing AI foundation models to accelerate the discovery of new battery materials. Traditionally, battery material discovery relied heavily on intuition and incremental improvements to a limited set of materials identified mainly between 1975 and 1985. The new AI-driven approach uses large, specialized foundation models trained on massive datasets of molecular structures to predict key properties such as conductivity, melting point, boiling point, and flammability. This enables a more efficient exploration of the vast chemical space—estimated to contain up to 10^60 possible molecular compounds—by focusing on promising candidates for battery electrolytes and electrodes. The team’s foundation model, trained on billions of molecules using text-based molecular representations (SMILES) and enhanced by a novel tool called SMIRK, allows for more precise and consistent learning of molecular structures. This approach helps overcome the limitations of traditional trial-and-error methods by providing