1005 Alcyon Dr Bellmawr NJ 08031
+1 484-446-2566
support@bmcpros.com
Artificial intelligence models that turn text into images can also be used to generate new materials. Generative materials models from companies such as Google, Microsoft and Meta have borrowed their training data to help researchers design tens of millions of materials over the past few years.
However, these models are working hard when designing materials with exotic quantum properties such as superconductivity or unique magnetic states. Too bad, because humans can use help. For example, after a decade of research, only a dozen candidates have been identified for a class of materials that may revolutionize quantum computing (called quantum spin liquids). Bottlenecks mean fewer materials are the basis for technological breakthroughs.
Now, MIT researchers have developed a technology that enables popular generative materials models to create promising quantum materials by following specific design rules. Rules or constraint models can create materials with unique structures that produce quantum properties.
“The models of these large companies generate materials optimized for stability,” said Mitda Li, a professor of career development at MIT in 1947. “Our view is that this is not usually how materials science is going. We don’t need 10 million new materials to change the world. We just need a very good material.”
The method is today Natural materials. The researchers applied their technology to generate millions of candidate materials composed of geometric lattice structures related to quantum properties. From this pool, they synthesize two actual materials with exotic characteristics.
“People in the quantum community really care about these geometric constraints, such as the kagome lattices of two overlapping, inverted triangles. We created materials with kagome lattices, because these materials can mimic the behavior of rare earth elements, so they have a high technical importance,” Li said.
Lee is the senior author of the paper. His MIT co-authors include PhD students Ryotaro Okabe, Mouyang Cheng, Abhijatmedhi Chotrattanapituk and Denisse Cordova Carrizales; Postdoctoral Manasi Mandal; undergraduate researchers Kiran Mak and Bowen Yu; visiting scholar Nguyen Tuan Hung; Xiang Fu ’22, PhD’24; Tommi Jaakkola, professor of electrical engineering and computer science, is a branch of the Institute of Computer Science and Artificial Intelligence (CSAIL) and the Institute of Data, Systems and Society. Other co-authors include King Wang of Emory, Weiwei Xie of Michigan State University, YQ Cheng of Oak Ridge National Laboratory and Robert Cava of Princeton University.
Steering model impact
The properties of a material depend on its structure, and quantum materials are no different. Some atomic structures are more likely to produce exotic quantum performance than others. For example, lattices can be used as a platform for high-temperature superconductors, while other shapes called kagomes and Lieb Lattices can support the creation of materials that may be useful for quantum computing.
To help a popular generative model called diffusion models produce materials that conform to specific geometric patterns, the researchers created Scigen (the abbreviation of structural constraint integration in generative models). Scigen is a computer code that ensures that user-defined constraints are adhered to in each iteration generation step. With Scigen, users can follow any generated AI diffusion model geometric rules when generating materials.
The AI diffusion model generates structures that reflect the structure distribution found in the dataset by sampling from its training dataset. Scigen blocked generations that didn’t match structural rules.
To test SCIGEN, the researchers applied it to a popular AI material generation model called DIFFCSP. They are equipped with SCIGEN models to generate materials with unique geometric patterns called Archimedean Lattices, which are collections of 2D lattice tiles of different polygons. Archimedean Lattices can lead to a range of quantum phenomena and have been the focus of many researches.
“Archimedean Lattices produce quantum spin liquids and so-called flat bands, which can mimic the properties of rare earths without rare earth elements, so they are very important,” said Cheng, co-author of the work. “Other Archimedean lattice materials have large pores and can be used for carbon capture and other applications, so it is a special range of materials. In some cases, there is no known material for that lattice, so I think it’s really interesting to find the first material that fits the lattice.”
The model generated over 10 million materials with Archimedean Lattices. One million of these materials survived screening for screening stability. The researchers then used a smaller sample using a supercomputer in Oak Ridge National Laboratory and conducted detailed simulations to understand the performance of the underlying atoms of the material. The researchers found that 41% of the structures were magnetic.
From this subset, researchers synthesized two previously undiscovered compounds Tipdbi and Tipbsb in XIE and Cava’s Labs. Subsequent experiments show that the predictions of AI models are largely consistent with the properties of actual materials.
“We wanted to discover that new materials might be possible by incorporating these structures that are well-known to produce quantum properties,” said Okabe, the first author of the paper. “We already know that these materials with specific geometric patterns are interesting, so it’s natural to start with them.”
Accelerate material breakthrough
Quantum spin liquids can unlock quantum computing by enabling stable, error-resistant Qubits that can serve as the basis for quantum operation. However, quantum spin liquid materials have not been confirmed. Xie and Cava believe Scigen can speed up the search for these materials.
“There are a lot of searches for quantum computer materials and topological superconductors, and these are all related to the geometric patterns of the materials,” Xie said. “But the experimental progress is very, very slow,” Kava added. “Many of these quantum rotating liquid materials are bound to be: they must be in a triangular lattice or kagome lattice. If the material meets these constraints, quantum researchers will be excited; this is a necessary but insufficient condition. Therefore, by producing many similar materials, it can immediately provide experimental subjects with hundreds or thousands of mass studies, thus making the quantification high for the quality calculation of the material in a tailored manner.
“This work proposes a new tool that leverages machine learning to predict which materials have the desired geometric patterns,” said Drexel University professor Steve May. “This should speed up the development of previously undeveloped materials for next-generation electronic, magnetic or optical technology applications.”
The researchers stressed that experiments are crucial to assess whether AI-generated materials can be synthesized and how their actual properties are compared with model predictions. Future work of Scigen may incorporate other design rules into generative models, including chemical and functional limitations.
“People who want to change the world about the properties of materials care more about the material than the stability and structure of the material,” Okabe said. “Through our approach, the proportion of stabilizing materials has dropped, but it opens a door to produce a large pile of promising materials.”
This work was partially supported by the U.S. Department of Energy, the National Center for Scientific Computing for Energy Research, the National Science Foundation and the Oak Ridge National Laboratory.