Artificial intelligence can accelerate the process of finding and testing new materials, and now researchers have used that ability to develop a battery that is less dependent on the costly mineral lithium.
Lithium-ion batteries power many devices that we use every day as well as electric vehicles. They would also be a necessary part of a green electric grid, as batteries are required to store renewable energy from wind turbines and solar panels. But lithium is expensive and mining it damages the environment. Finding a replacement for this crucial metal could be costly and time-consuming, requiring researchers to develop and test millions of candidates over the course of years. Using AI, Nathan Baker at Microsoft and his colleagues accomplished the task in months. They designed and built a battery that uses up to 70 per cent less lithium than some competing designs.
The researchers focused on a type of battery that only contains solid parts, and they looked for new materials for the battery component that electric charges move through, called the electrolyte. They started with 23.6 million candidate materials designed by tweaking the structure of established electrolytes and swapping out some lithium atoms for other elements. An AI algorithm then eliminated the materials that it calculated would be unstable, as well as those in which the chemical reactions that make batteries work would be weak. The researchers also considered how each material would behave while the battery was actively working. After only a few days, their list contained just a few hundred candidates, some of which had never been studied before.
“But we’re not material scientists,” says Baker. “So I called up some experts who’ve worked on large battery projects with the Department of Energy… and said, ‘What do you think? Are we crazy?’”
Vijay Murugesan at the Pacific Northwest National Laboratory in Washington state was one of the scientists who picked up the phone. He and his colleagues suggested additional screening criteria for the AI. After more elimination rounds, Murugesan’s team ultimately picked one of the AI’s suggestions to synthesise in the lab. It stood out because half of what Murugesan would have expected to be lithium atoms were replaced with sodium. He says that this is a very novel recipe for an electrolyte and that having the two elements together opens questions about the basic physics of how the material works inside a battery.
His team built a working battery with this material, albeit with a lower conductivity than similar prototypes that use more lithium. Baker and Murugesan both say that lots of work is left to optimise the new battery. However, the process of making it – from the first time Murugesan spoke to the Microsoft team to the battery being functional enough to turn on a light bulb – took about nine months.
“The methods here are bleeding edge, in terms of machine learning tools, but what really elevates this is that things got made and tested,” says Rafael Gómez-Bombarelli at the Massachusetts Institute of Technology, who was not involved with the project. “It’s very easy to do predictions; it’s hard to convince someone to invest on actual experiments.” He says that the team used AI to accelerate and augment calculations that physicists have been doing for decades. But this approach may still run into obstacles in the future. The data needed to train the AI for this type of work is often sparse, and materials other than battery components may require a more complex way of combining elements, he says.