Charting a New Material World – Sponsor Content

In 1905, a young American scientist named William Coolidge ready to work on an ambitious task as part of his new job at General Electric: creating a better light bulb. The carbon fiber used in Thomas Edison’s world-changing invention was not very energy efficient. Scientists experimented with tungsten, which had the advantage of having a higher melting point, but the metal wire was too brittle and broke easily.

A few years later, after numerous trials, errors and accidents – by some accounts the breakthrough came when a tungsten rod fell into liquid mercury flowing from a heating system – Coolidge created “malleable tungsten”, a more bendable form of metal that eventually became wiring, which can still be found in modern light bulbs.

Ekin Dogus Cubuk, a research associate at Google DeepMind, points to malleable tungsten as an example of how new materials can revolutionize existing technologies — often after careful, labor-intensive efforts full of unexpected twists. More than 100 years after this breakthrough innovation, his team has created an artificial intelligence tool that could change the way scientists discover new materials. As with malleable tungsten, each of these compounds could improve current materials and lead to next-generation electric vehicle batteries or better solar panels. “Many material discoveries require a lot of trial and error, luck and coincidence,” Cubuk says. “With artificial intelligence, we hope to reduce dependence on luck. We have the potential to accelerate every step of the process, from discovery to development, with more accurate and scalable forecasts.”

GNOME catapults material discovery

The artificial intelligence tool created by Cubuk’s team is called GNOME (Graph Networks for Materials Exploration) and uses deep learning to determine the structure of 2.2 million inorganic crystals. According to simulations, the GNOME project identified 380,000 materials that are stable at low temperatures. “We hope that having this information will help the community catapult further breakthroughs in materials discovery and design,” explains Cubuk.

Since these crystals became publicly available in 2023, scientists have been experimenting with AI-identified compounds in labs around the world — including three recently synthesized in Japan by scientists from Hokkaido University.

“There is a wealth of knowledge about artificial intelligence, but there is currently a gap between translating these insights into materials synthesis in the lab. This early research shows how we can fill the gap and explore new materials more efficiently,” says Akira Miura, an associate professor at Hokkaido University who led the study.

Meanwhile, according to Cubuk, the GNOME project has independently identified hundreds of crystals that scientists have already discovered on their own as stable, including “exciting” superconductor candidates. Moreover, hundreds of these crystals have been independently synthesized in laboratories around the world. This is a strong signal that the new material candidates in the GNOME project arrangements have real feasibility.

Initially, GNOME was trained using crystal structure data available from the Materials Project, an open source resource that has played a key role in the discovery of new materials. Identifying the structure of a new material is just the beginning of what could be a long and intensive development process. According to founder and UC Berkeley professor Kristin Persson, the original idea of ​​the Materials Project was to shorten invention time by focusing experiments on compounds with the greatest potential for success. This in itself was a game changer from when Persson was a student in the mid-1990s and had to comb through large volumes to check the chemical structure and properties of the material. In the lab, researchers combined intuition with previous results to inform their experiments. “There was no recipe for it outside of the researcher’s brain,” Persson says. “They would use their experience to create new materials, and that would take a long time.”

Not long ago, it was relatively rare to be able to calculate a new material on a computer that was thermodynamically stable, explains Chris Wolverton, a professor of materials science and engineering at Northwestern University. First, the sheer amount of computing power required was an obstacle. “Google is particularly rich when it comes to time and computing power,” he says. “They were able to calculate the properties of several million compounds in a very short time.”

The availability of this huge data resource opens up new opportunities for materials researchers, says Persson. “They give this work back to the community and say, ‘Do with it what you want,’” he says. “This is something new in our offer when it comes to the tools we have. If you’re a materials company and you’re not thinking about machine learning, you’re behind, and Google has shed a great light on that.”

This doesn’t mean that piles of new materials will suddenly start appearing on their lab benches. The tool identified crystals that the calculations predict will be stable at absolute zero, around -273°C – the ideal state where particle movement stops. The key question remains: Can these crystals be created in a stable form that will not change or decompose at room temperature? Myriad factors, including temperature and pressure, could influence whether the theoretical compound could exist in the form that would become, say, the next EV battery. Scientists will need to develop appropriate experimental methods to create materials useful for innovation. “This is the next huge challenge in this field: predicting synthesis recipes,” Wolverton says.

Materials scientists are exploring next steps

Indeed, the enormous scale of AI discoveries has spurred important conversations about how best to identify and synthesize new materials. A researcher can spend an entire career experimenting on just a few compounds, which is an extremely slow and labor-intensive process. The key is to strike a balance between building AI-powered capabilities and advancing established science, explains University of Liverpool professor Andy Cooper, whose research group focuses on ways to accelerate the discovery of functional materials: “I think that’s a challenge for the future. Where is the right place on this spectrum?”

New materials improve the foundations of modern life

At Google DeepMind, Cubuk sees a future where new materials can reshape today’s most exciting technologies – from semiconductors to new power sources for supercomputers. As AI tools evolve, a key challenge is determining which potential materials best suit the technologies and products, Cubuk explains. For example, knowing what criteria make a good lithium-ion conductor helps researchers better identify the most promising candidates. The GNOME team has already identified 528 potential battery materials — more than 25 times the number identified in a 2017 study led by Austin Sendek, CEO of Aionics, a technology company that designs high-performance batteries using artificial intelligence.

Artificial intelligence contributions can also open researchers’ minds to unforeseen possibilities, says Jakoah Brgoch, a chemistry professor at the University of Houston whose work examines functional inorganic materials. Perhaps the compounds suggested by AI are not feasible at current laboratory temperatures and pressures, for example; but what happens if you change these variables? “Learning is often hard work,” says Brgoch. “The hope and the hype is that we won’t have to go through this ordeal. Generative AI can predict the next big thing. As we exhaust what is known, we must develop new ways to push boundaries, and this is one such path. Of course, experimentally verifying these predictions is easier said than done.

According to current industry knowledge, on average, discovering new materials requires at least 10 years of work and between $10 million and $100 million. Although the cost of building the factory is the same, a computer simulation that previously took weeks can now be done in seconds thanks to artificial intelligence, Sendek says.

“AI-based materials discovery methods will unlock unconventional solutions that would otherwise elude human intuition, while also leading to fewer dead ends that cannot be commercialized,” says Sendek, pointing to the recent discovery of Aionics’ artificial intelligence system which involves adding a luxury perfume molecule to the battery electrolyte can improve the overall performance of the battery. Aionics is now actively testing this molecule in real batteries.

“Making significant progress in materials innovation requires us to discover better materials faster and then bring them to market in a fraction of the time.”