AI’s Quest: Finding the Perfect Solid Electrolyte for Li-ion Batteries

Microsoft and the US Pacific Northwest National Laboratory (PNNL) have joined forces in a multi-year project aimed at exploring material science using artificial intelligence (AI) and super-computing. As part of this collaboration, the team has made a breakthrough discovery of a promising solid electrolyte for lithium-sodium-ion batteries.

The research process involved several steps, starting with the use of AI algorithms to evaluate elements for suitability and suggest combinations. This initial step generated a list of 32 million starting compounds. Subsequent algorithms were then employed to eliminate unstable materials, filter out candidate molecules based on reactivity, and identify those with the potential to conduct energy. Through this process, the list was whittled down to about 500,000 mostly new stable materials, and eventually to 800.

Following the AI-driven selection, theoretical modeling was employed to further analyze the remaining materials. While this approach provided more accurate estimates, it required significantly more computing power and time compared to AI. The modeling involved density functional theory to calculate the energy of each material relative to all other states it could be in. Molecular dynamics simulations, combining AI and high-performance computing (HPC), were then used to analyze the movements of atoms and molecules within each material.

After these rigorous computational analyses, 150 chemicals remained as potential candidates. HPC was then utilized to narrow down the selection based on practical metrics such as cost and availability, resulting in a final list of 23 materials. From these, PNNL materials scientists selected six to be synthesized and tested in batteries. One of these materials has already shown promise as a conductor of both lithium and sodium ions.

Previously, it was believed that sodium and lithium ions could not be used together in a solid-state electrolyte system due to their similar charges but different sizes. However, the research team discovered that these ions actually support each other, enabling their movement within the solid-state electrolyte material.

While it remains uncertain whether any of these materials will ultimately be used in batteries, the speed at which a workable battery chemistry was found is highly compelling. This collaborative effort between Microsoft and PNNL demonstrates the potential of AI and super-computing in accelerating material science research and discovery.