Microsoft released MatterGen, a new AI model that is able to discover new chemical materials
Researchers have made the code for the AI model available on GitHub.
2 min. read
Published on
Read our disclosure page to find out how can you help Windows Report sustain the editorial team. Read more
Microsoft Research has introduced a new AI tool called MatterGen, which opens up a fresh approach to materials discovery. Rather than screening through existing candidates, MatterGen directly generates novel materials based on specific requirements. This allows for the efficient exploration of materials beyond those already known.
MatterGen is based on a diffusion model, an AI concept that has already seen success in other areas, such as image generation. The model generates new materials by adjusting their 3D geometry, element composition, and arrangement in periodic space. It then can be fine-tuned to create materials with specific properties, such as certain chemistry, crystal symmetry, or electronic and mechanical characteristics.
However, the AI model can also generate materials with properties not seen in known materials, potentially leading to the discovery of unexpected new materials. According to the researchers, the model is particularly good at generating materials with a combination of properties, such as a specific bandgap and a high mechanical strength.
It was designed for researchers to use as part of a broader materials discovery pipeline. For instance, MatterGen doesn’t generate detailed atomic structures of materials, which are necessary to calculate properties like how well a material can conduct electricity or absorb light. Instead, MatterGen generates simplified representations of materials that can be used to screen through many candidates quickly.
The researchers behind MatterGen have released a paper describing the model and made the code available in open source, along with tools for evaluating the performance of materials generation models. At launch, MatterGen supports the generation of materials that consist of elements in the periodic table. Still, the researchers plan to expand the tool to more complex materials like molecules and 2D materials like graphene.
Microsoft Research has also validated MatterGen’s capabilities by using the model to suggest the structure of a novel material, tantalum chromium oxide, which the company then successfully synthesized in the lab.
The Redmond-based tech giant has been extensively invested in AI focused on science advancements, so MatterGen should not surprise anyone. Last year, for example, Microsoft started the TRAIN initiative in Europe, intending to use AI to improve healthcare services, Including easier access to hospitalizations, doctors, and treatments.
What’s your take on this?
User forum
0 messages