MatterGen - Microsoft's inorganic material generation model
MatterGen is an innovative generative model from Microsoft, specifically designed for inorganic materials. Based on a unique diffusion process, it progressively refines atomic types, coordinates, and periodic lattices to generate stable and diverse inorganic materials spanning the periodic table...
What is MatterGen?
MatterGen is an innovative generative model from Microsoft, specifically designed for the design of inorganic materials. Based on a unique diffusion process, it progressively refines atomic types, coordinates, and periodic lattices to generate stable and diverse inorganic materials spanning the periodic table. MatterGen can be fine-tuned to meet a wide range of performance constraints, such as chemical composition, symmetry, magnetic properties, electronic properties, and mechanical properties. Compared to previous material generation models, MatterGen excels in generating stable, unique, and novel materials, producing structures that more closely approximate the local energy minimum of the DFT (Digital Functional Theory). Within a given DFT property computational budget, MatterGen can find more materials that meet extreme performance constraints.
MatterGen's main functions
- Generate stable and diverse inorganic materialsIt can generate various inorganic materials across the periodic table, and the generated materials have high stability, uniqueness and novelty.
- Satisfying a wide range of performance constraintsBased on fine-tuning, materials that meet specific constraints such as chemical composition, symmetry, magnetic properties, electronic and mechanical properties can be generated, such as magnetic materials with high magnetic density, semiconductor materials with specific band gaps, and superhard materials with high bulk modulus.
- Reverse Material DesignThis method directly generates material structures based on target performance constraints, breaking through the limitations of traditional screening methods based on known materials and greatly improving the efficiency of finding new materials.
MatterGen's technical principles
- diffusion modelCrystalline materials are generated based on a diffusion model. The diffusion model generates samples based on a reversed fixed destruction process implemented using a learned fractional network. For crystalline materials, a customized diffusion process is defined that considers their unique periodic structure and symmetry, destroying and denoising the atomic type, coordinates, and periodic lattice respectively.
- Fractional NetworkA pre-trained fractional network is used to jointly denoise atom types, coordinates, and lattice on a large, stable material structure dataset. The fractional network outputs fractional values for noise removal without learning symmetry from the data.
- Adapter moduleThis introduces an adapter module to fine-tune the score model on an additional dataset with performance labels. The adapter module is an adjustable component injected into each layer of the base model, which can change the model output based on the given performance labels, thereby guiding the generation of target performance constraints.
- Dataset: Pre-trained using the large and diverse dataset Alex-MP-20, which contains 607,683 stable structures recomputed from the Materials Project and the Alexandria dataset.
MatterGen's project address
- Project official website:https://www.microsoft.com/en-us/research/blog/mattergen
- Technical Papers:https://www.nature.com/articles/s41586-025-08628-5
Application scenarios of MatterGen
- Energy storageIt is used in the design of new battery materials, such as high-capacity lithium-ion battery cathode materials and high-performance solid electrolytes, to improve the energy density and power density of batteries.
- catalytic: Develop highly selective catalysts for the synthesis of specific chemicals in petrochemical and fine chemical industries, as well as for the treatment of automobile exhaust in environmental catalysis, to improve reaction efficiency and environmental friendliness.
- Carbon captureThe goal is to design materials that efficiently adsorb carbon dioxide and catalytic materials that convert carbon dioxide into useful chemicals, thereby achieving carbon recycling and contributing to environmental protection.
- Electronic materialsTo develop new semiconductor materials and high-performance magnetic materials for use in manufacturing high-performance electronic devices and to promote the development of electronic technology.
- superhard materials: Develop superhard materials for use in cutting tools and wear-resistant coatings to improve the wear resistance and corrosion resistance of mechanical parts, with applications in aerospace, automotive and other fields.