New battery material that uses less lithium found in AI-powered search
Microsoft announced Tuesday that a team of scientists used artificial intelligence and high-performance computing to plow through 32.6 million possible battery materials ― many not found in nature ― in 80 hours, a task the team estimates previously would have taken 20 years. The results kick off an ambitious effort to create a new generation of batteries less dependent on toxic and environmentally damaging lithium.
The company shared some of the best candidates with the government's Pacific Northwest National Laboratory in Richland, Wash., which investigated the most promising ones and built a prototype battery using a brand-new material.
While the dime-size prototype is not yet ready for a prime-time role powering the watches and car keys of today, it functions using less lithium than commercially available options and has the ability to recharge power. Moreover, the feat demonstrates the potential of new technologies to revolutionize the underappreciated but fast-evolving field of materials science.
"It's really not an overstatement to say that almost every major problem that we face as a society could be, at least in part, helped by having better materials," said Christopher M. Wolverton, a professor at Northwestern University's materials science and engineering department. Wolverton, who was not involved in Microsoft's battery project but has worked on similar projects of his own, called the company's claim to have screened 32.6 million materials in 80 hours "staggering."
"Thirty-two million tells me right away that they weren't screening [just] known materials," he added. "They were screening hypothesized new materials that they were hoping to discover."
Like other outside experts interviewed, he was able to review news releases from Microsoft and Pacific Northwest National Laboratory but was not able to see the scientific paper describing the work before its release. The paper, which has yet to be peer-reviewed, was posted Tuesday on the science preprint site arXiv.
- - -
High-powered mixing and matching
Although estimates vary, experts generally consider there to be about 200,000 known materials in the world, and materials scientists continue to seek new ones that may hold the key to solving some of the planet's serious challenges.
To combat climate change, scientists must find the best materials for capturing and storing atmospheric carbon dioxide. To reduce the vast accumulations of plastic littering beaches, clogging landfills and threatening human health, scientists must find safe ways to break it down into fuel and raw materials. To respond quickly to new pathogens capable of causing future pandemics, scientists must design effective drugs that use proteins to attack viruses and bacteria or to stimulate the body's defenses.
All depend on discovering the right materials, a process that has been slowed by hypothesis-driven trial-and-error approaches.
High-performance computing and artificial intelligence are now allowing scientists to rapidly use the elements in the periodic table like a painter might use a palette of colors, mixing them and arriving at new configurations.
"This is a new way to do science," said Nathan Baker, an author of the paper and senior director of partnerships for chemistry and materials at Microsoft. The project used the company's Azure Quantum Elements platform, which was unveiled in June with the goal of using advanced computing power to speed up the discovery process.
Starting with all of the known materials, Baker said, "we can basically pick through the periodic table and drop new atoms into [various] locations through a process of substitution."
Azure Quantum Elements quickly weeded out materials that are poor conductors, too unstable, too reactive or too costly to be used for batteries. The high-performance computing used by Microsoft essentially harnessed the power of about 5,000 typical laptops. Baker stressed that once they had early results, Microsoft consulted materials experts at the government laboratory in the belief that, "Okay, a computer gave us the answer. . . . Let's go prove it."
Officials with Microsoft and the national laboratory said that they will continue the battery project as part of a three-year collaboration aimed at speeding up the pace of innovation and discovery.
"This is not just a stunt announcement, a headline-grabbing sort of announcement. This is the nature of things to come," said Chirag Dekate, a vice president and analyst at the Stamford, Conn.-based research and consulting firm Gartner who was not on the research team.
- - -
The hunt for a better battery
The need for better battery technology has emerged as a major challenge facing scientists.
"A lot of the current battery technologies are highly inefficient," Dekate said. "If you look at one of the reasons why the promise of a green future has eluded us thus far despite advances in solar and wind [power], if you look at the underpinning reason, it is actually the challenges associated with battery technologies."
Lithium-ion batteries now power most electric cars as well as most of the electric scooters and electric bicycles that have become ubiquitous in modern cities around the globe. In 2022, the global lithium ion battery market size was valued at $46.2 billion, and the industry is projected to reach $189.4 billion by 2032.
But the current crop of batteries are unable to charge quickly or hold their charge for long periods. Moreover, some have proved to be volatile.
In 2023, the New York fire department reported that 18 people died in fires linked to electric-vehicle batteries.
Microsoft searched for materials to build what's called a solid-state electrolyte battery; these have a greater energy density than liquid ion batteries and do not present a fire or leakage risk. The material used for the prototype contained some lithium ― but up to 70 percent less than the amount found in existing batteries.
Microsoft's work on the battery project "began in earnest about nine months ago," Baker said.
The team trained its artificial-intelligence system by showing it examples of different materials, helping it learn their crystal structures and energetic properties. "As we go over different compositions, it starts to learn how the structure and composition relate to energetics," Baker said.
The Microsoft team then used AI to function like a funnel. A large amount of possibilities were fed into the top of the funnel and then passed through various filters that narrowed the list of candidates. The best ones emerged from the bottom of the funnel.
Given that a battery material must be stable, the artificial intelligence began by filtering the number of candidates to just under 590,000 that would stay in the form required to perform in a battery. From there, the AI essentially asked of each candidate: Is it going to react to oxygen? Will it do something strange when electrical current is applied across it? How do you get lithium atoms to move through the material, a necessary process for a functioning battery?
Eventually, Microsoft contacted the Pacific Northwest National Laboratory, which has deep experience in battery research.
"They asked, what are the properties that you would typically need from a material to make a better battery?" said Vijay Murugesan, a paper co-author and the lab's materials science group lead in the Physical Sciences Division. He called developing a solid material that will allow only lithium ions to move from one side of the battery to the other a huge challenge.
When he examined a list of the 120 to 130 best candidates, Murugesan said his reactions varied from "Why? What? Really?" for some of the most surprising candidates to "I see that is an obvious next step from what we are doing."
Scientists said they were encouraged by the fact that some of the materials recommended by artificial intelligence had already been flagged by experts as promising.
The team at the laboratory analyzed the top candidates and made a handful of the hypothetical materials dreamed up by computers. The new material chosen for the prototype contains some lithium along with sodium, chloride and yttrium. The lab is conducting further work on the material used for the prototype and is preparing to investigate at least two more candidate materials.
Responding by email to preliminary information presented by Microsoft, Aron Walsh, a professor and chair in materials design at Imperial College London, said, "AI is providing a new generation of approximate but practical tools, which enable us to tackle problems that seemed impossible before, such as the rapid exploration of large chemical spaces reported here."
Related Content
Christie is out but may not be done trying to stop Trump from winning
Secrets, lies and TMI: A history of White House health disclosures
The biggest wins and controversies from Bill Belichick's career
