Generative AI has revolutionized how we create text, images and code. How about new materials? We at Microsoft Research #AI4Science are thrilled to announce MatterGen: our generative model that enables broad property-guided materials design.
The central problem in materials science is to discover materials with desired properties. Traditionally, it has been done by first finding novel materials and then filtering down based on the application. This is like trying to generate the image of a cat by first creating a million different images and then searching for the one with a cat.
MatterGen is a diffusion model that can instead directly generate novel materials with desired property conditions – including chemistry, symmetry, and material properties – similar to how DALL·E 3 tackles image generation.
MatterGen outperforms a previous SOTA model (CDVAE) in generating 2.9 times more stable and novel structures, and produces structures that are 17.5 times closer to the energy local minimum. It also outperforms screening in proposing high bulk modulus candidate structures, and improves upon substitution and random structure search when targeting a particular chemical system.
We believe MatterGen is an important step forward in AI for materials design. Our results are currently verified via DFT, which has many known limitations. Experimental verification remains the ultimate test for real-word impact, and we hope to follow up with more results.
None of this would be possible without the highly collaborative work between Andrew Fowler, Claudio Zeni, Daniel Zuegner, Matthew Horton, Robert Pinsler, Ryota Tomioka, and myself, our amazing interns Xiang Fu, Aliaksandra Shysheya, and Jonathan Crabbé, as well as Jake Smith, Lixin Sun and the entire AI4Science Materials Design team. We are also grateful for all the help from the Microsoft Research AI4Science team and Microsoft Azure Quantum team.
#artificialintelligence #generativeai #materialsscience #ai4science #microsoft #ai #materials