Experiment

MatterGen

Designing new materials is often a slow, resource-intensive process. MatterGen, now available in Azure AI Foundry, offers a new approach: a generative diffusion model that creates candidates for inorganic materials directly at the atomic level. We first announced MatterGen when it published in Nature earlier this year, serving as an exciting precursor to today’s model launch in Azure AI Foundry.

MatterGen can generate crystal structures unconditionally, or researchers can guide generation by targeting properties such as bulk modulus, band gap, chemical system, or magnetic density. This enables scientists to quickly explore stable, novel, and unique material candidates tailored to their research goals.

Trained on high-quality open datasets like the Materials Project and Alexandria, MatterGen has shown strong performance even in cases where training data is scarce for extreme property values. The model represents a step forward in how AI can accelerate R&D for new materials in fields such as energy, electronics, and manufacturing. 

Experiment with MatterGen in Azure AI Foundry and reshape the way we discover the building blocks of the future.

Resources: GitHub – microsoft/mattergen – Model in Foundry Catalog.