Speaking crystal AI predicts atomic arrangements to aid material discovery

Researchers at the University of Reading and University College London have unveiled CrystaLLM, a new AI tool capable of predicting atomic arrangements in crystal structures. This innovative approach could significantly accelerate the discovery of materials for applications ranging from solar panels to advanced computer chips.

CrystaLLM applies principles similar to AI chatbots, learning the "language" of crystals by analyzing millions of existing crystal structures. The study, published on December 6 in Nature Communications, highlights how this technology can transform material science by simplifying a traditionally complex process.

Dr. Luis Antunes, the lead researcher and recent University of Reading PhD graduate, explained: "Predicting crystal structures is like solving a complex, multidimensional puzzle where the pieces are hidden. Crystal structure prediction requires massive computing power to test countless possible arrangements of atoms.

"CrystaLLM offers a breakthrough by studying millions of known crystal structures to understand patterns and predict new ones, much like an expert puzzle solver who recognises winning patterns rather than trying every possible move."

Simplifying Predictions for Novel Materials

Traditional methods for predicting crystal structures rely on computationally intense simulations of atomic interactions. CrystaLLM simplifies this process by treating crystal descriptions, stored in Crystallographic Information Files, as textual data. The AI model "reads" these descriptions and predicts subsequent patterns, learning rules of chemistry and physics without direct instruction.

This unique approach allows CrystaLLM to generate realistic crystal structures, even for materials it has not encountered before.

To assist researchers, the team has launched a free online platform integrating CrystaLLM into crystal structure prediction workflows. This tool is expected to expedite the development of next-generation technologies, such as improved batteries, efficient solar cells, and faster computing hardware.

For further exploration of the AI model, visit here.

Research Report:Crystal structure generation with autoregressive large language modelling