The AI from Google DeepMind has identified 2.2 million new crystal structures

Google DeepMind and researchers at Lawrence Berkeley National Laboratory have achieved a significant advancement in materials science by creating an AI system named GNoME that has identified 2.2 million novel inorganic crystal structures, with 700 of these having been experimentally validated. This research, published in Nature, highlights a 10-fold increase in known stable inorganic crystals facilitated by GNoME’s deep learning techniques and two distinct methods for discovering stable materials. Utilizing autonomous robotic systems, a subset of GNoME’s predictions reached a remarkable 71% synthesis success rate over 17 days of continuous experiments. This wealth of data, now publicly accessible via the Materials Project database, offers the potential to revolutionize the discovery of new materials—potentially accelerating innovation and reducing costs in producing technologies from batteries to computer chips. The studies demonstrate the power of AI and deep learning in scientific discovery, potentially transforming the process and efficiency of material discovery in various industries.