Scaling deep learning for materials discovery
Problem
Discovery of inorganic crystals for applications from microchips to batteries and photovoltaics has been bottlenecked by expensive trial-and-error approaches.
Why it matters
Novel functional materials enable fundamental breakthroughs from clean energy to information processing.
What the AI did
Graph networks trained at scale on 48,000 known stable crystals reached unprecedented generalization, predicting new stable structures that escaped previous human chemical intuition.
Results
Order-of-magnitude improvement in materials discovery efficiency; discovery of 2.2 million structures below the current convex hull.
Impact
Claims an order-of-magnitude expansion in stable materials known to humanity.
Record
| Venue | Nature |
|---|---|
| Published | 2023 |
| Work type | article |
| Authors | Amil Merchant, Simon Batzner, Samuel S. Schoenholz, Muratahan Aykol, Gowoon Cheon, Ekin D. Cubuk |
| Citations | 1,471 (OpenAlex, at import time) |
| References | 69 |
| Model or method | graph networks trained at scale (deep learning) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41586-023-06735-9 |
| Open access | hybrid |
| Keywords | materials discovery, graph networks, crystal stability, deep learning, convex hull |
Importance rating
major — millions of lives, an industry or a field
Abstract claims order-of-magnitude improvement in materials discovery and expansion of stable materials known to humanity, spanning microchips, batteries, photovoltaics—industry-wide and foundational science impact.