Webmaster's Home (ChinaZ.com) News on November 30: Google DeepMind's research team discovered 2.2 million theoretically stable but experimentally unrealized crystal structures through the artificial intelligence tool GNoME. This result was published in " Published in the journal Nature. The discovery of more than 45 times the number of crystal structures discovered in the history of science offers potential advancements in areas such as renewable energy and advanced computing.
The researchers plan to make the 381,000 most promising structures available to other scientists for fabrication and testing in areas ranging from solar cells to superconductors. This move highlights the potential for leveraging artificial intelligence to streamline years of experimental work and potentially lead to improved products and processes .
Ekin Dogus Cubuk, one of the paper’s co-authors, said: “For me, materials science is basically the intersection of abstract thinking and the physical universe. It’s hard to imagine a technology that wouldn’t benefit from better materials. Improvement."
The DeepMind team aims to discover new crystals to add to the previously calculated 48,000 identified species. Known substances range from copper and iron, known for thousands of years, to more recent discoveries. The
DeepMind team identified new materials by using machine learning to first generate candidate structures and then evaluate their likely stability. DeepMind estimates that, based on the 28,000 stable materials discovered in the past decade, this amount is equivalent to the knowledge gained experimentally from nearly 800 years ago..
The paper in the journal Nature states: "From microchips to batteries and photovoltaic materials, the discovery of inorganic crystals has been limited by expensive trial-and-error methods. Our work represents an order of magnitude expansion of the number of known stable materials." Two potential applications for the new compound include the creation of multifunctional layered materials and the development of biomimetic computing, which uses chips to mimic the workings of the human brain,
Cubuk said.
Researchers at the University of California, Berkeley, and Lawrence Berkeley National Laboratory have used these findings as part of experimental work to create new materials, described in a separate paper published in the journal Nature.
The team deployed computing, historical data, and machine learning to guide an automated laboratory called the A-lab to create 41 new compounds from a list of 58 targets—with a success rate of more than 70 percent.
Gerbrand Ceder, a professor at the University of California and co-author of the paper, said this high success rate is surprising and may even increase further. He added that the key to improvements in is to use in conjunction with AI technology and existing resources, such as large datasets of past synthetic reactions.
He said: "While A-lab's robotics is cool, the real innovation is combining various sources of knowledge and data with A-lab to intelligently drive synthesis."
Professor Bilge Yildiz of MIT The techniques outlined in the two Nature papers will enable the identification of new materials "at the speed needed to address the world's grand challenges," said the researcher, who was not involved in either study.
"This vast database of inorganic crystals should be filled with 'gems' waiting to be discovered to advance solutions to clean energy and environmental challenges," said Yildiz, who works in MIT's Departments of Materials Science and Engineering and Nuclear Science and Engineering.
