MatterGen: Microsoft's diffusion model that designs materials on demand
In one sentence Microsoft Research publishes MatterGen in Nature: a diffusion model generating stable crystal structures conditioned on target properties (magnetism, conductivity). Experimental synthesis of a new material confirmed.
Microsoft Research publishes a Nature paper on MatterGen, a model that does for materials what DALL-E does for images: tell it "I want a magnetic, stable material with low formation energy" and it generates the candidate crystal structure, atom by atom.
It's not just simulation: the team took one of the proposed materials (TaCr₂O₆) and actually synthesized it in the lab. The measured elastic modulus is within ~7% of the prediction. Rare proof that a generative model proposed something genuinely new and physically real.
For materials discovery (batteries, semiconductors, magnets) it's a meaningful step: from "search across known candidates" (DeepMind's GNoMe) to "generate the right candidate for the given problem".
Companies
Microsoft Research
Tools
MatterGen
Tags
Sources