Doubling possibilities: GNoME’s AI propels material discovery into new frontiers

This news has been read 750 times!

NEW YORK, Dec 3: Inorganic crystals, crucial for advanced technologies, often require time-consuming experiments for their discovery. Researchers, faced with limited outcomes from trial-and-error methods, have traditionally sought new crystal structures through experimentation. Computational methods, led by the Materials Project and others, have identified 28,000 novel materials in the past decade. However, the potential of AI to predict experimentally viable materials has been a challenge.

Lawrence Berkeley National Laboratory and Google DeepMind researchers present an AI tool, Graph Networks for Materials Exploration (GNoME), capable of predicting material stability. GNoME’s deep learning predicts 2.2 million crystals, doubling the number of feasible materials and offering promises for next-gen batteries and superconductors. Using a graph neural network (GNN) approach, GNoME leverages atomic connections to discover novel crystalline materials efficiently.

Active learning and Density Functional Theory enhance its predictive power, increasing the stability prediction discovery rate from 50% to 80%. The autonomous lab, guided by GNoME, synthesized over 41 novel materials, showcasing the potential of AI-driven materials synthesis. GNoME’s forecasts are released to the scientific community through the Materials Project, fostering further exploration and leveraging machine learning as experimental guidelines.

This news has been read 750 times!