MIT links crystal structures and language in a new materials-search model

MIT researchers have developed an AI model that combines the three-dimensional arrangement of atoms in crystals with textual descriptions. The aim is to help scientists search for materials with desired properties without relying only on published text.

According to MIT's October 18 announcement, the findings were published in Nature Communications. Training uses descriptions linking structure, properties, processing and potential applications, including information from computational databases.

In a test involving solid electrolytes for lithium-ion batteries, the suggestions included materials independently identified as promising. The authors report an advantage over the text-only model used for comparison.

This is a research result for selecting candidates, not an announcement of a finished new battery. The practical suitability of suggested materials requires further testing.