Researchers at MIT have developed a groundbreaking generative AI model that significantly simplifies the process of determining the structures of crystalline materials, such as metals, rocks, and ceramics. Traditionally, scientists have relied on X-ray crystallography to uncover these structures, but this new approach offers a more efficient and versatile alternative. The model has far-reaching implications for industries relying on materials like batteries, magnets, and superconductors.

“Understanding the structure of a material is crucial for virtually any application,” explained Danna Freedman, the Frederick George Keyes Professor of Chemistry at MIT. “It’s fundamental for superconductivity, magnets, photovoltaics—essentially anything that is materials-centric.”

Crystalline materials, including metals and many inorganic solids, are composed of repeating units or “lattices” that form precise, identical patterns. MIT’s new AI model divides the structure prediction process into multiple subtasks. It begins by determining the size and shape of the lattice “box” and identifying the atoms that will occupy it. Next, the model predicts the arrangement of these atoms within the box. For each diffraction pattern, the AI generates multiple possible structures and evaluates them by comparing predicted diffraction patterns with the actual input.

“Our model is a form of generative AI, meaning it creates structures that haven’t been encountered before,” said Eric Riesel, an MIT graduate student. “We can generate a hundred different guesses and match the powder pattern to verify the accuracy. If the output matches the input, we’ve nailed it.”

The MIT research team tested their AI model using thousands of simulated diffraction patterns from the Materials Project. Additionally, they validated its effectiveness with more than 100 experimental diffraction patterns sourced from the RRUFF database, which holds powdered X-ray diffraction data for nearly 14,000 natural crystalline minerals. The model demonstrated an accuracy rate of 67% on this data.

The team then pushed the AI further by applying it to unsolved diffraction patterns from the Powder Diffraction File, which contains data on over 400,000 solved and unsolved materials. Notably, the model successfully solved more than 100 previously unsolved patterns.

In addition to solving existing diffraction patterns, MIT researchers used the AI model to discover structures for three new materials. These materials, created by Danna Freedman’s lab, formed under high-pressure conditions, forcing elements that do not typically react at atmospheric pressure to combine into new compounds. Although these new materials share the same chemical composition as their original forms, their crystal structures and physical properties are radically different.

This innovative approach could lead to the creation of new materials with unique properties, paving the way for advances in numerous fields, from energy storage to advanced manufacturing.

The generative AI model developed at MIT marks a significant step forward in materials science, offering a faster, more efficient way to predict and design crystal structures that can fuel future discoveries and innovations.

By Impact Lab