Conditional generative models enable targeted exploration of MAX phase design space

Researchers at UCL, AWE and collaborators have developed an artificial intelligence approach to accelerate the discovery of new MAX phases, a family of layered materials used in applications ranging from energy technologies to advanced coatings. MAX phases are also precursors to MXenes, an exciting class of two-dimensional materials with potential uses in energy storage, catalysis and electronics.

The team adapted a large language model, CrystaLLM-π, to generate candidate crystal structures and, crucially, steer its search towards regions of chemical space most likely to yield desirable materials. By introducing simple design targets into the model, the researchers doubled the rate at which promising new structures were discovered compared with conventional generation approaches. Follow-up quantum mechanical calculations confirmed the stability of several previously unknown candidates.

The work demonstrates how generative AI can be used not just to explore materials space, but to target specific properties, providing a scalable route to faster materials discovery and design

Authors: Jamie Swaine; Cyprien Bone; Prakriti Kayastha; Matthew T. Darby; Ewan Galloway; Keith T. Butler