AI Predicts Hundreds of Thousands of New Stable Materials, Speeding Up Discovery
A new AI program has predicted the stability of hundreds of thousands of previously unknown inorganic crystals. This massive discovery expands our catalog of known stable materials by ten times, offering a huge leap in material science. It provides scientists with a roadmap to create new, useful materials for various technologies.
These newly predicted materials could lead to breakthroughs in everything from better batteries and electronics to new medical devices and sustainable energy solutions.
Discovering new materials usually means slow, expensive trial and error. A deep-learning system predicted the thermodynamic stability of millions of hypothetical crystal structures, identifying a large set predicted to be stable enough to exist.
A portion have since been synthesized by autonomous and human labs, validating the predictions. Candidates span potential battery electrolytes, superconductors, and catalysts.
The work reframes materials discovery as a search problem AI can accelerate dramatically, shrinking the gap between an idea for a material and a sample on a bench.
🔬 Read the technical version
AI predicts hundreds of thousands of previously unknown stable crystalline materials
A graph neural network predicted the stability of vast numbers of candidate inorganic crystals, expanding the catalog of known stable materials by an order of magnitude and pointing experimentalists toward promising compounds to synthesize.