Materials scientists have long hoped that artificial intelligence could make designing new crystals as simple as telling an AI what properties are needed. But large language models, trained almost entirely on streams of text, struggle to understand the three-dimensional atomic arrangements that define most crystalline materials. The experimental literature also skews toward a few well-studied crystal classes, so AI outputs tend to favour those materials.
To address these limitations, the MIT team created a vision-language model for crystalline materials. They trained a language model to produce textual descriptions of crystal structures using contrastive learning, a technique where the model compares similar and dissimilar examples of structures and their textual descriptions to build a unified understanding. This lets the model “see” atomic arrangements and bonding patterns, not just read about them.
“Structure is really at the heart of the materials science field,” says Ju Li, the Carl Richard Soderberg Professor in Power Engineering in the Department of Materials Science and Engineering. “If you understand the structure, you can infer quite a lot about the properties and performance of materials.”
The researchers tested the model by asking it to identify materials suitable for solid-state lithium-ion battery electrolytes. Its top results matched independently identified promising candidates. The model also learned geometric representations of material structures in ways text-only models could not.
The paper’s first author is former MIT postdoc Yang Jeong Park. Co-authors include Mayank Kumaran, Chia-Wei Hsu, and Elsa Olivetti, the Jerry McAfee (1940) Professor in Engineering in DMSE.