Microsoft Research is trying to change how scientists look for new materials. Its two AI tools, MatterGen and MatterSim, are designed to move the process beyond screening existing candidates and toward creating entirely new materials from scratch.
The pair works as a connected system: one tool generates possible materials, and the other tests them through simulation. The goal is to help researchers design materials with specific chemical, mechanical, and electronic properties before moving deeper into real-world work.
How MatterGen Creates Candidates
MatterGen is the generation side of the system. It uses a specialized diffusion algorithm to build three-dimensional molecular structures, producing possible materials that fit user-defined constraints.
That distinction matters because the tool is not simply ranking known materials. It is proposing new structures that may meet a particular need. In materials research, that changes the starting point: instead of asking which existing option is closest, researchers can ask what kind of material should exist.
Tian Xie, Principal Research Manager at Microsoft Research, described the approach this way: "MatterGen generates thousands of candidates with user-defined constraints to propose new materials that meet specific needs," adding, "This represents a paradigm shift in how materials are designed."
The source describes this as a step forward from traditional screening methods. The promise is not that AI replaces materials science, but that it can expand the set of candidates scientists have available for investigation.
Why MatterSim Is The Other Half
Generation alone is not enough. A proposed material still has to be tested, and that is where MatterSim comes in.
MatterSim simulates how candidate materials would perform in real-world conditions. It is built to evaluate materials across a wide range of environments, including temperatures from absolute zero to 5,000 Kelvin and pressures up to 10 million atmospheres.
According to the source, MatterSim combines quantum mechanics principles with machine learning to handle complex calculations. That combination allows the tool to test material behavior computationally before scientists commit to more extensive experimental work.
Used together, MatterGen and MatterSim create a two-step AI workflow:
- MatterGen proposes new material candidates with desired constraints.
- MatterSim evaluates how those candidates may behave under demanding conditions.
- Researchers can then focus attention on candidates that appear more promising.
This is the core logic of the system. The AI is useful not just because it can generate many options, but because those options can be filtered through simulation before physical testing becomes the next step.
A Real Material Test With TaCr2O6
The system has already been connected to a real-world materials experiment. Working with the Shenzhen Institute of Advanced Technology, researchers created a new material called TaCr2O6 that MatterGen had proposed.
The synthesized material’s actual properties matched the AI’s predictions about 80 percent of the time. That result gives the project a concrete example beyond software demonstration: MatterGen proposed a candidate, researchers made it, and the outcome was compared with the AI prediction.
The source does not present that as a finished solution to materials discovery. It does show why the pairing of generation and simulation is important. A model that creates candidates must still be judged against physical results, and MatterSim is intended to support that evaluation path.
Open Access And Azure Quantum Elements
Microsoft has released MatterGen’s source code under the MIT license, along with training and fine-tuning data. That makes the generation tool more accessible to researchers and developers who want to examine or build on the work.
Both MatterGen and MatterSim are also part of Microsoft’s Azure Quantum Elements platform. There, the tools are being used to help companies develop new materials for batteries, magnets, and fuel cells.
Those application areas make the significance clear. Materials with specific chemical, mechanical, or electronic properties can matter across technologies where performance depends on what a material can do under pressure, temperature, or operational demands.
Microsoft published its research findings in Nature. The tools are also part of Microsoft’s "AI for Science" initiative, which began two and a half years ago.
What This Means For AI In Materials Research
The larger shift is methodological. MatterGen and MatterSim point to a materials workflow where AI can propose, constrain, and test candidates before a final material is synthesized and evaluated.
That does not remove the need for scientific validation. The TaCr2O6 example shows the opposite: AI predictions become meaningful when they are checked against real materials. But the process can give researchers a broader and more targeted set of possibilities to explore.
For Microsoft Research, the key claim is that new material design can become more generative. Rather than starting only with existing databases of known options, researchers can use AI to suggest structures built around specific needs, then use simulation to narrow the field.
If that workflow continues to hold up in practice, MatterGen and MatterSim could become important tools for teams working on batteries, magnets, fuel cells, and other material-dependent technologies. The central idea is simple: let AI create possible materials, then test them rigorously before deciding what deserves the next experiment.