Google DeepMind's new AI has developed over 700 materials for EV batteries, solar cells, and more

Google DeepMind's new materials are being tested in the A-Lab using robotics (Image: Marilyn Sargent / Berkeley Lab)

Google DeepMind's new AI is developing numerous materials, including for EV batteries, solar cells, computer chips and much more. In addition, the new A-Lab laboratory is taking material development to a next level by combining robotics and machine learning.

Nicole Dominikowski (translated by Nicole Dominikowski), Published 12/16/2023 🇩🇪 🇫🇷 ...

Google DeepMind's new AI, GNoME (graphical networks for material exploration), has identified the structures of approximately 2.2 million new materials. This includes roughly 380,000 stable materials that could drive future technologies, such as next-generation electric car batteries, solar cells, computer chips, and superconductors. Various researchers around the world are now producing and experimentally testing 736 of these. DeepMind has identified 528 promising lithium-ion battery conductors that could help make batteries more efficient.

While materials play a very critical role in almost any technology, we as humanity know only a few tens of thousands of stable materials.

- Dogus Cubuk, materials discovery lead at Google DeepMind

GNoME has expanded the number of stable materials known to humankind to 421,000 (Image: DeepMind Google)

While the use of AI to develop new materials has become common, GNoMe stands out for its scale and precision. Chris Bartel, assistant professor of chemical engineering and materials science at the University of Minnesota, notes that GNoMe was trained with a significantly larger amount of data than comparable projects.

Hardware, especially when it comes to clean energy, needs innovation if we are going to solve the climate crisis. This is one aspect of accelerating that innovation.

- Kristin Persson, leader of The Materials Project at Berkeley Lab

Researchers often spend years developing materials based on existing structures in the hope of discovering new combinations. Thanks to the deep learning tool, this research can now be accelerated. The Lawrence Berkeley National Laboratory, together with Google DeepMind, has published two articles in the journal Nature. One article describes how AI predictions can be used for autonomous material synthesis.

Repeat for rounds of active learning (Image: DeepMind Google)

However, a significant issue persists: New materials often take a long time to reach the commercial stage.

If we can reduce this to five years, that will be a big improvement.

- Dogus Cubuk

The newly autonomous A-Lab at Berkeley Lab is investigating the real-world utility of new materials. In just 17 days, the lab was able to carry out 355 experiments and successfully synthesize 41 out of 58 proposed compounds. This is significantly faster than the time it would take human researchers.

If you’re unlucky, it can take months or even years. Most students give up after a few weeks. But the A-Lab doesn’t mind failing. It keeps trying and trying.

- Kristin Persson