Bits With Brains
Curated AI News for Decision-Makers
What Every Senior Decision-Maker Needs to Understand About AI and its Impact
AI Breakthrough in Materials Science
12/10/23
Editorial team at Bits with Brains
Google DeepMind, the organization that gifted AlphaFold to the world, has now unveiled a powerful new AI tool called Graph Networks for Materials Exploration (GNoME) that has dramatically accelerated the discovery of new materials.
As materials are critical for technologies from batteries to solar cells, this innovation could unlock major advances across industries.
GNoME leverages graph neural networks, an AI technique effective at learning from complex structured data like molecules. After training on 50,000 known stable materials, GNoME was able to generate over 2 million completely new potential structures. It also predicts the likelihood that each material is stable and viable to produce experimentally.
Out of GNoME's 2.2 million materials, DeepMind published 381,000 of the most promising. This exponentially grows the known stable inorganic crystals 10x from 48,000 to over 400,000. The achievement is equivalent to nearly 800 years of normal experimental materials discovery!
In a demonstration of real-world viability, over 700 of GNoME's predicted materials have already been synthesized. DeepMind is collaborating with labs on further synthesis tests and application screening for batteries, electronics, solar and more.
With the ability to shortcut years of painstaking research, advanced materials discovered by GNoME could enable breakthrough innovations. Priority areas include solid-state electrolytes for safer batteries, layered materials for electronics, and photovoltaic absorbers for more efficient solar cells.
In a related project, DeepMind also worked with a robotic lab at Berkeley to produce the predicted materials. With little human input, the robotic system autonomously executed over 500 experiments to create 41 entirely new stable compounds.
GNoME showcases the immense potential when AI is applied to accelerate scientific discovery. With further advances in predictive modeling and automated experiments, the pace of materials innovation could dramatically increase - unlocking technologies to benefit both society and the environment.
Sources:
[1]https://deepmind.google/discover/blog/millions-of-new-materials-discovered-with-deep-learning/
Sources