A major breakthrough in material design! Microsoft released a big innovation model, and the accuracy rate increased by 10 times!
Image source: Generated by AI
Microsoft has released MatterGen, an innovative large model dedicated to inorganic material design.
MatterGen’s infrastructure is based on a diffusion model, which can gradually optimize atomic types, coordinates and periodic lattices to quickly generate different types of new inorganic materials. For example, in the energy field, MatterGen can generate a new cathode material for lithium-ion batteries.
By adjusting the atomic type, introducing some transition metal elements with special electronic structures, and accurately determining their position coordinates in the crystal lattice, a crystal lattice with a unique microstructure can be developed to continue the battery life and life.
Compared with traditional discovery methods,MatterGen can more than double the proportion of stable, unique and novel materials produced, and bring the resulting structure nearly 10 times away from its DFT local energy minimum。Therefore, MatterGen is of great help to high-tech fields such as electric vehicles, aerospace, and electronic chips.
Maybe many friends are a little confused about this new field, so let’s give a simple and easy-to-understand explanation. If you want to build a house, the traditional method is like picking from existing house plans, which may not suit our particular needs.
Now that you use MatterGen to build a house, you can simply say that you want a five-bedroom and one living room. You need a gym, an e-sports room, two small bedrooms, a master bedroom. It’s best to have a small garden outside the house. The overall structure of the house adopts Chinese style, and it is best to add some dragons ~ phoenixes to the walls.
This means thatMatterGen discovers complex inorganic materials through the diffusion process and performs more refined decomposition and generation。We can gradually explore and build the most suitable material combination and structural layout according to these specific requirements we have given.
Start with the type of atoms, just like selecting building materials with different textures and characteristics; then carefully determine the coordinate positions of these atoms in space and accurately place each piece of masonry; finally build a perfect periodic lattice and build a solid and unique house frame.
In fact, many netizens were confused after reading this. Please treat me like a 5-year-old child and explain this technology ~
I know AI is changing everything and will eventually happen. But I didn’t expect it to come so soon.
Look at it discovering some amazing superconductors, improving its computing performance, which in turn enhances its ability to discover more superconducting materials, and further improves its computing power, and so on and over again… you know. Imagine AI is optimizing everything. The critical critical mass has been achieved.
There could be a revolution in battery cell additives, which have been discussed and in demand in the field in recent years. Based on images provided by Microsoft, it looks like a model that could also help produce cathode active materials.
Thanks to this model, AGI has been implemented.
It’s time for AI to solve global warming.
Is this equivalent to the AlphaFold model in the materials world?
Introduction to MatterGen Architecture
In the MatterGen model, the diffusion process is the core mechanism for forming crystal structure. The inspiration for this process comes from the phenomenon of diffusion in physics, in which particles move from areas of high concentration to areas of low concentration until uniform distribution is achieved. In the context of material design, the diffusion process is cleverly adapted to generate an orderly, stable crystal structure from a completely random initial state.
The diffusion process begins with a random initial structure that has no physical meaning, just a random distribution of atoms in space. Then,MatterGen gradually reduces the “noise” in this initial structure through a series of iterative steps, bringing it gradually closer to a true crystal structure。This process is not a simple random change, but is strictly guided by the laws of physics and the principles of material science.
At each iteration, MatterGen fine-tunes the atom’s type, coordinates, and lattice parameters. These fine-tuning are based on a predefined distribution of physical motivations, which means that the model takes into account the actual physical properties of the crystal material, such as bond lengths between atoms, bond angles, and lattice symmetry when adjusting atomic positions and types.
For example, coordinate diffusion respects the periodic boundaries of the crystal and adjusts the position of atoms through a wrapped normal distribution to ensure that atoms do not leave the periodic structure of the crystal. Lattice diffusion takes a symmetrical form, with the mean distribution being a cubic lattice, and the average atomic density comes from training data, which ensures that the resulting lattice structure is both stable and physically meaningful.
The equivariant fractional network is another key component in the MatterGen model, responsible for learning how to recover the original crystal structure from the diffusion process. The design of this network is based on an important physical principle-isotropy.
Equivariant refers to the property of a system that maintains certain properties unchanged under certain transformations. In crystal materials, this means that the properties of the material remain unchanged under operations such as rotation, translation, etc.
Equariferous fractional networks can output equariferous fractions of atomic types, coordinates, and lattices by learning patterns in the data. These fractions represent the “mismatch” of each atom and lattice parameter in the current structure, that is, their deviation from the ideal crystal structure.
By calculating these fractions, the network guides the model on how to adjust atomic and lattice parameters to reduce noise in the structure and bring it closer to a stable crystal structure. This is also one of the important reasons why MatterGen can improve accuracy and ideality rates.
To increase the flexibility of the model,MatterGen has added an adapter module to fine-tune different downstream tasks, and can change the output of the model based on a given property label。(It’s the tailor-made feature we talked about in that case)
The adapter introduces an additional set of parameters at each layer of the model that can be adjusted based on task-specific property labels. During the fine-tuning process, these parameters are optimized to make the structure generated by the model better meet the requirements of the specific task.This design not only improves the adaptability of the model, but also reduces the amount of labeled data required for fine-tuning, because the model does not need to learn the characteristics of each task from scratch, but can be adjusted based on pre-training.。
For example, when designing a new type of battery material, a model may be needed to focus on the material’s electrical conductivity and ion diffusivity; while when designing a new type of catalyst, a model may be needed to focus on the material’s surface activity and selectivity. The adapter module allows the model to adjust its generation structure strategy based on these different requirements.
Currently, Microsoft has published the research in Nature and has been recognized by many technology celebrities. It can be compared to the AlphaFold series of protein prediction models that Google won the Nobel Prize in Chemistry last year.