Artificial intelligence models that turn text into images can also help generate new materials. Over the past few years, generative material models from companies such as Google, Microsoft, and Meta have utilized training data to help researchers design tens of millions of new materials.
However, these models are difficult when it comes to designing materials with exotic quantum properties such as superconductivity and unique magnetic states. That’s a shame because humans can use help. For example, after a decade of research into a class of materials that could revolutionize quantum computing, known as quantum spin liquids, only dozens of material candidates have been identified. Bottlenecks mean that there are fewer materials to serve as the basis for technical breakthroughs.
Currently, MIT researchers have developed a method that allows popular model of material to create promising quantum materials by following specific design rules. Rules, or constraints, manipulate the model to create materials with unique structures that cause quantum properties.
“These large companies’ models produce materials optimized for stability,” says Mingda Li, a 1947 career development professor MIT class. “Our perspective is not usually how materials science advances. We don’t need 10 million new materials to change the world. We need really good materials.”
The approach is explained in a paper published by today Natural materials. Researchers applied techniques to generate millions of candidate materials consisting of geometric lattice structures related to quantum properties. From that pool they synthesized two real materials with exotic magnetic properties.
“People in the quantum community really care about these geometric constraints, like two overlapping upside down triangle Kagome lattices. They created materials with Kagum lattices because these materials can mimic the behavior of rare earth elements, says Lee.
Li is a senior author of the paper. His MIT co-authors include doctoral students Ryotaro Okabe, Mouyang Cheng, Abhijatmedhi Chotratanapituk and Denisse Cordova Carrizales. Postdoc Manasi Mandal; undergraduate researchers Kiran Mack and Bowen Yu; Visiting Scholar Nguyen Tuan Hung; Xiang Fu ’22, PhD ’24; Professor Tommy Cocola of Electrical Engineering and Computer Science, an affiliate of the Institute of Computer Science and Artificial Intelligence (CSAIL) and the Institute of Data, Systems and Society. Additional co-authors include Yao Wang from Emory University, Wei Wei Xi from Michigan State University, YQ Chen from Oak Ridge National Laboratory, and Robert Kava from Princeton University.
Maneuver the model for impact
The properties of materials are determined by their structure, and quantum materials are no exception. Certain atomic structures are more likely to cause exotic quantum properties than other atomic structures. For example, square lattices act as platforms for high temperature superconductors, while other shapes known as Kagome and Lieb Lattices can support the creation of materials that are useful for quantum computing.
To help a popular class of generative models known as diffusion models generate materials that conform to specific geometric patterns, researchers have created Scigen (short for structural constraint integration of generative models). Scigen is computer code that ensures that the spreading model adheres to user-defined constraints at each iterative generation step. Scigen allows users to provide geometrical rules for generative AI diffusion models as they follow when generating materials.
AI diffusion models work by sampling from training datasets to generate structures that reflect the distribution of structures found in the dataset. Scigen blocks generations that do not match the structural rules.
To test Scigen, researchers applied it to a popular AI material generation model known as DiffCSP. They produced a material with a unique geometric pattern known as the Archimedean Lattices on the Scigen-equipped model. This is a collection of 2D lattice tilings of different polygons. Archimedean’s lattice can lead to a variety of quantum phenomena and is the focus of many researches.
“Archimedean Lattices are extremely important because they produce quantum spin liquids and so-called flat bands that can mimic the properties of rare earths without rare earth elements.” “Other archimede lattice materials have large pores that can be used for carbon capture and other applications, making them a special collection of materials. In some cases, there are no known materials in that lattice, so I think it’s really interesting to find the first material that fits that lattice.”
This model generated over 10 million material candidates with Archimedean Lattices. One million of these materials survived screening for stability. Using the Oak Ridge National Laboratory supercomputer, researchers took small samples of 26,000 materials and performed detailed simulations to understand how the underlying atoms of the material worked. Researchers found magnetism in 41% of these structures.
From that subset, researchers synthesized two previously undiscovered compounds, Tipdbi and Tipbsb, in the Xie and Cava labs. Subsequent experiments showed that the predictions of the AI model were in close agreement with the properties of the actual material.
“We wanted to discover new materials that could have a great potential impact by incorporating these structures known to cause quantum properties,” says Okabe, the first author of the paper. “We already know that these materials with certain geometric patterns are interesting, so it’s natural to start with them.”
Accelerate material breakthroughs
Quantum spin liquids can unlock quantum computing by enabling stable, error-resistant qubits that serve as the basis for quantum operation. However, no quantum spin liquid materials have been identified. Xie and Cava believe that Scigen can accelerate searching for these materials.
“There’s a big search for quantum computer materials and topological superconductors, all of which are related to the geometric patterns of the material,” says Xie. “But the experimental progress was very slow,” Kaba added. “Many of these quantum spin liquid materials are subject to constraints: they must be in a triangular or kagum lattice. If the material meets those constraints, quantum researchers will be excited. It is necessary but not sufficient.
“This work presents a new tool that utilizes machine learning, which can predict which materials will have specific elements in the desired geometric pattern,” says Steve May, a professor at Drexel University who was not involved in the study. “This should speed up the development of previously unexplored materials for next-generation electronic, magnetic or optical technology applications.”
Researchers emphasize that experiments are still important to assess whether AI-generated materials can be synthesized and how actual properties compare to model predictions. Future work on Scigen can incorporate additional design rules into generative models that contain chemical and functional constraints.
“People who want to change the world care more about material properties than material stability and structure,” says Okabe. “Our approach reduces the stable material ratio, but it opens the door and produces many promising materials.”
This work was supported in part by the U.S. Department of Energy, the National Center for Energy Research and Science Computing, the National Science Foundation, and the Oak Ridge National Laboratory.
