MIT model helps AI understand crystals in 3D to find better materials
Most solid materials are crystals, but their atomic structures are stored in special formats that ordinary language models do not read well. Researchers at MIT built a multimodal model that combines 3D atomic structures with natural language descriptions drawn from research papers and computational databases. The result is a richer picture of each material, covering its structure, properties, processing and possible uses. In a paper in Nature Communications, the team reports that the model picked better material designs than text only models. When asked for solid state lithium battery electrolytes, its top picks included candidates other researchers had already flagged as promising. Tools like this could speed up the search for materials behind future batteries and devices.