Google has introduced Graph Networks for Materials Exploration (GNoME), an AI tool that discovered 2.2 million new crystals, offering a wealth of potential materials for future technologies. In a Nature paper, Google outlined GNoME’s capabilities and the promising applications of these crystals, including superconductors for powering supercomputers, next-gen batteries for electric vehicles, and more.
GNoME’s predictions have been shared with the research community, and Google is contributing 380,000 predicted stable materials to the Materials Project for further exploration.
GNoME utilizes deep learning and graph neural network (GNN) models to predict the stability of new materials at an unprecedented scale. Google highlights that GNoME increases the efficiency and speed of material discovery significantly. External researchers have experimentally validated 736 of GNoME’s new structures, demonstrating the model’s accuracy.
Around 20,000 crystals from the ICSD database, validated through experiments, are computationally stable. Utilizing computational methods from databases like the Materials Project increased stable crystal count to 48,000. GNoME significantly expands the catalog of known stable materials to 421,000.
One notable achievement is the discovery of 52,000 new layered compounds similar to graphene, with potential applications in revolutionizing electronics through the development of superconductors. Additionally, the model identified 528 potential lithium-ion conductors, 25 times more than in previous studies, with implications for enhancing rechargeable battery performance.
Google emphasizes GNoME’s role in accelerating materials discovery, reducing costs, and supporting the development of greener technologies. The company’s collaboration with Berkeley Lab resulted in the successful synthesis of over 41 new materials using GNoME’s insights on stability and materials from the Materials Project.
The AI-driven approach to materials discovery showcased by GNoME demonstrates the potential of machine learning tools to guide experimentation and reshape the field’s future.