Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage.
Researchers at Google DeepMind and Lawrence Berkeley National Laboratory have made a remarkable scientific breakthrough by developing a new AI system called GNoME. This system has discovered over 2 million new materials that could be used for technology development, such as batteries, solar panels, and computer chips.
Published in the scientific journal Nature, the research details how DeepMind’s team scaled up deep learning techniques, allowing GNoME to explore potential material structures with unprecedented efficiency. Remarkably, in just 17 days, the AI system identified 2.2 million potentially stable new inorganic crystal structures, more than 700 of which have already been experimentally confirmed. This is nearly a tenfold increase in the number of known stable inorganic crystals.
GNoME employs two methods to find stable materials: one method generates similar crystal structures, while the other uses a more random approach. Both methods’ results are tested, enhancing the GNoME database for future learning.
The second paper describes how GNoME’s predictions were tested with autonomous robotic systems at Berkeley Lab. Over 17 days of continuous experiments, the system successfully synthesized 41 out of 58 predicted compounds, achieving a high success rate of 71%.
The newly discovered materials data has been made publicly available through the Materials Project database. Researchers can use this database to screen structures and identify materials with desired properties for real-world applications. For instance, the research has revealed 52,000 potential new 2D layered materials akin to graphene, 25 times more solid lithium-ion conductors than previous studies, and 15 new lithium-manganese oxide compounds that could replace lithium-cobalt oxide in batteries.
Google found that 736 of the materials predicted by GNoME were also independently discovered by other scientists around the world. Some examples include a unique optical material (Li4MgGe2S7) and a potential superconductor (Mo5GeB2).
The key to GNoME’s success is its use of sophisticated graph neural networks. These networks can predict the stability of proposed crystal structures in mere seconds, allowing GNoME to filter vast numbers of computer-generated candidates down to the most promising ones. This capability addresses previous challenges in estimating the energies and stability of new materials.
The researchers highlighted that the high success rate of GNoME showcases the effectiveness of AI-driven platforms for autonomous materials discovery. This success motivates further integration of computational techniques, historical knowledge, and robotics.
The implications of these studies are profound for the future of materials science. AI-driven approaches like this can significantly speed up the creation of new materials for specific applications, leading to faster innovation and cost reductions in product development. The use of AI and deep learning suggests a future where labor-intensive lab experiments are minimized or even eliminated, allowing scientists to focus more on designing and analyzing unique compounds.
Overall, these developments mark a new era in materials science, driving innovation across fields, from energy storage to advanced medical equipment. The fusion of artificial intelligence, deep learning, and scientific research continues to push the boundaries of what is possible.