MIT researchers build a multimodal model that links crystal structures to text descriptions

MIT researchers created a multimodal AI model that combines 3D atomic structures with natural language descriptions of crystals from experimental papers and computational databases.

In a Nature Communications paper, the researchers say the model identifies better material designs than text-only models, such as candidates for solid-state lithium-ion battery electrolytes that were independently identified as promising.

The researchers say the model also learns geometric representations of material structures that text-only models cannot.