US supercomputer-backed AI aims to speed hunt for battery breakthroughs

Researchers at the University of Michigan, led by Venkat Vishwanathan, are leveraging U.S. Department of Energy (DOE) supercomputers at Argonne National Laboratory to accelerate the discovery of new battery materials. Traditionally, battery material development relied heavily on intuition and incremental improvements based on a limited set of materials discovered mainly between 1975 and 1985. The team is now using foundational AI models trained on billions of molecules via powerful supercomputers like Polaris and Aurora to predict key properties such as conductivity, melting point, and flammability. This approach enables rapid, data-driven identification of promising electrolytes and electrode materials, which are critical for developing next-generation batteries that are more powerful, longer-lasting, and safer. These foundational AI models differ from traditional AI by possessing a broad understanding of molecular structures, allowing them to efficiently tackle specific tasks in battery design. The researchers employed a text-based molecular representation system called SMILES, enhanced by a new tool named SMIRK, to improve prediction