Discover Qatar-Media.tv

GNOME: DeepMind Ushers in a New Era in Crystalline Material Research through Deep Learning

GNOME: DeepMind Ushers in a New Era in Crystalline Material Research through Deep Learning: Crystal synthesis plays a fundamental role in various fields, from electronics and photonics to catalysis and medicine, forming the foundation of modern technologies. A significant breakthrough in this domain has been shared by Google DeepMind in the Nature journal, unveiling the revolutionary AI tool GNoME. Based on deep learning, it has led to the discovery of 2.2 million new crystals, with 380,000 deemed stable and promising for future technological applications. This advancement is equivalent to nearly 800 years of cumulative knowledge, marking a significant milestone in crystalline material research.

With GNoME (Graph Networks for Materials Exploration), the Google DeepMind team showcases the potential of using AI to discover and develop new materials on a large scale.

GNoME: A Revolution in Material Discovery In the past, discovering new crystalline structures was a laborious and expensive process, involving the modification of existing crystals or experimenting with new combinations of elements. This trial-and-error process could take months to produce results, often with limitations. Over the last decade, computer-based approaches, including those led by the Materials Project and other groups, have led to the discovery of 28,000 new materials.

However, these approaches have faced a fundamental limitation: the challenge of accurately predicting which materials might be experimentally viable, posing a hurdle in the research and development of new materials.

The stable crystals uncovered by GNoME have the potential to transform various technological sectors. Possible applications include superconductors, power supplies for supercomputers, and next-generation batteries improving the efficiency of electric vehicles. The discovery of 52,000 graphene-like compounds offers promising prospects for electronics, while highlighting 528 potential lithium-ion conductors could revolutionize rechargeable batteries.

Operation of the GNoME Neural Network Model The input data for GNNs (Graph Neural Networks) comes in the form of a graph that can be likened to connections between atoms, making GNoME well-suited for the discovery of new crystalline materials.

Initially trained with data on crystalline structures and their stability, freely accessible within the Materials Project, it uses two distinct pipelines: one based on structures similar to known crystals and the other on chemical formulas, allowing for comprehensive exploration. An active learning process significantly enhanced GNoME's performance, increasing the material stability discovery rate from 50% to 80%.

The model's efficiency elevated the discovery rate from less than 10% in previous approaches to over 80%. Credit DeepMind GNoME employs two pipelines to discover (stable) materials with low energy consumption. The structural pipeline creates candidates with structures similar to known crystals, while the compositional pipeline takes a more random approach based on chemical formulas. The results from both pipelines are evaluated using established density functional theory calculations, and these results are added to the GNoME database, informing the next cycle of active learning.

These results have been validated by independent experiments conducted by researchers worldwide: GNoME facilitated the discovery of 736 new materials in external laboratories, demonstrating the reliability of its predictions. In partnership with the team, researchers from the Lawrence Berkeley National Laboratory demonstrated the ability to autonomously synthesize new materials using AI predictions.

For researchers:

"The rapid development of new technologies based on these crystals will depend on the ability to manufacture them. In a paper led by our collaborators from the Berkeley Lab, researchers showed that a robotic laboratory could quickly manufacture new materials through automated synthesis techniques. Using materials from the Materials Project and GNoME's stability knowledge, the autonomous laboratory created new recipes for crystalline structures and successfully synthesized over 41 new materials, opening new possibilities for AI-driven material synthesis."

They have made GNoME's data and predictions available to the global research community, fostering collaborative research and paving the way for new discoveries.