Advances in AI and high-performance computing are changing the way scientists look for new battery materials.

Artificial intelligence (AI) and large-scale cloud computing is speeding up the search for new battery materials. An AI-enhanced collaboration between Microsoft and the Pacific Northwest National Laboratory (PNNL) has already produced one promising new material, which the two are sharing publicly today.

They discovered a new kind of solid-state electrolyte, the kind of material that could lead to a battery that’s less likely to burst into flames than today’s lithium-ion batteries. It also uses less lithium, which is getting harder to come by as demand soars for rechargeable EV batteries.

There’s still a long road ahead to see how viable this material is as an alternative to traditional lithium-ion batteries. What scientists are most excited about is the potential for generative AI to speed up their work. This discovery is just the first of many materials they’ll test in search of a better battery.

“The big point to make is the speed by which we got to a new idea, a new material. If we can see that kind of acceleration, my bet would be on that this is the way of the future to find these kinds of materials,” says Karl Mueller, a physical chemist and program development office director at PNNL.

Microsoft reached out to PNNL researchers last year to offer its Azure Quantum Elements (AQE), a platform that brings together high-performance computing and AI — and eventually, quantum computing, according to Microsoft. The company launched it last year as a tool tailored for discoveries in chemistry and materials science.

The researchers queried AQE for battery materials that use less lithium, and it quickly suggested 32 million different candidates. From there, the AI system had to discern which of those materials would be stable enough to use — which wound up being around 500,000. They used more filters to deduce how well each material might conduct energy, simulate how atoms and molecules move within each material, and suss out how practical each candidate would be when it comes to cost and availability.

Eventually just 23 candidates were left, of which five were already known materials. All the whittling down took just 80 hours — a feat so speedy it would have been virtually impossible without AI and AQE.

“Thirty-two million is something that we would never ever be able to do ... Imagine a human sitting and going through 32 million materials and choosing one or two out of it. It’s just not going to happen,” says Vijay Murugesan, a staff scientist and materials sciences group lead at PNNL.