The factitious intelligence models that turn text into images are also useful for generating recent materials. Over the previous couple of years, generative materials models from corporations like Google, Microsoft, and Meta have drawn on their training data to assist researchers design tens of hundreds of thousands of latest materials.
But in terms of designing materials with exotic quantum properties like superconductivity or unique magnetic states, those models struggle. That’s too bad, because humans could use the assistance. For instance, after a decade of research right into a class of materials that might revolutionize quantum computing, called quantum spin liquids, only a dozen material candidates have been identified. The bottleneck means there are fewer materials to function the premise for technological breakthroughs.
Now, MIT researchers have developed a way that lets popular generative materials models create promising quantum materials by following specific design rules. The principles, or constraints, steer models to create materials with unique structures that give rise to quantum properties.
“The models from these large corporations generate materials optimized for stability,” says Mingda Li, MIT’s Class of 1947 Profession Development Professor. “Our perspective is that’s not normally how materials science advances. We don’t need 10 million recent materials to vary the world. We just need one really good material.”
The approach is described today in a paper published by . The researchers applied their technique to generate hundreds of thousands of candidate materials consisting of geometric lattice structures related to quantum properties. From that pool, they synthesized two actual materials with exotic magnetic traits.
“People within the quantum community really care about these geometric constraints, just like the Kagome lattices which are two overlapping, upside-down triangles. We created materials with Kagome lattices because those materials can mimic the behavior of rare earth elements, in order that they are of high technical importance.” Li says.
Li is the senior writer of the paper. His MIT co-authors include PhD students Ryotaro Okabe, Mouyang Cheng, Abhijatmedhi Chotrattanapituk, and Denisse Cordova Carrizales; postdoc Manasi Mandal; undergraduate researchers Kiran Mak and Bowen Yu; visiting scholar Nguyen Tuan Hung; Xiang Fu ’22, PhD ’24; and professor of electrical engineering and computer science Tommi Jaakkola, who’s an affiliate of the Computer Science and Artificial Intelligence Laboratory (CSAIL) and Institute for Data, Systems, and Society. Additional co-authors include Yao Wang of Emory University, Weiwei Xie of Michigan State University, YQ Cheng of Oak Ridge National Laboratory, and Robert Cava of Princeton University.
Steering models toward impact
A fabric’s properties are determined by its structure, and quantum materials aren’t any different. Certain atomic structures are more likely to offer rise to exotic quantum properties than others. For example, square lattices can function a platform for high-temperature superconductors, while other shapes generally known as Kagome and Lieb lattices can support the creation of materials that may very well be useful for quantum computing.
To assist a well-liked class of generative models generally known as a diffusion models produce materials that conform to particular geometric patterns, the researchers created SCIGEN (short for Structural Constraint Integration in GENerative model). SCIGEN is a pc code that ensures diffusion models adhere to user-defined constraints at each iterative generation step. With SCIGEN, users can provide any generative AI diffusion model geometric structural rules to follow because it generates materials.
AI diffusion models work by sampling from their training dataset to generate structures that reflect the distribution of structures present in the dataset. SCIGEN blocks generations that don’t align with the structural rules.
To check SCIGEN, the researchers applied it to a well-liked AI materials generation model generally known as DiffCSP. They’d the SCIGEN-equipped model generate materials with unique geometric patterns generally known as Archimedean lattices, that are collections of 2D lattice tilings of various polygons. Archimedean lattices can result in a variety of quantum phenomena and have been the main target of much research.
“Archimedean lattices give rise to quantum spin liquids and so-called flat bands, which may mimic the properties of rare earths without rare earth elements, in order that they are extremely vital,” says Cheng, a co-corresponding writer of the work. “Other Archimedean lattice materials have large pores that may very well be used for carbon capture and other applications, so it’s a set of special materials. In some cases, there aren’t any known materials with that lattice, so I feel it’ll be really interesting to search out the primary material that matches in that lattice.”
The model generated over 10 million material candidates with Archimedean lattices. A million of those materials survived a screening for stability. Using the supercomputers in Oak Ridge National Laboratory, the researchers then took a smaller sample of 26,000 materials and ran detailed simulations to grasp how the materials’ underlying atoms behaved. The researchers found magnetism in 41 percent of those structures.
From that subset, the researchers synthesized two previously undiscovered compounds, TiPdBi and TiPbSb, at Xie and Cava’s labs. Subsequent experiments showed the AI model’s predictions largely aligned with the actual material’s properties.
“We desired to discover recent materials that might have an enormous potential impact by incorporating these structures which have been known to offer rise to quantum properties,” says Okabe, the paper’s first writer. “We already know that these materials with specific geometric patterns are interesting, so it’s natural to begin with them.”
Accelerating material breakthroughs
Quantum spin liquids could unlock quantum computing by enabling stable, error-resistant qubits that function the premise of quantum operations. But no quantum spin liquid materials have been confirmed. Xie and Cava imagine SCIGEN could speed up the seek for these materials.
“There’s a giant seek for quantum computer materials and topological superconductors, and these are all related to the geometric patterns of materials,” Xie says. “But experimental progress has been very, very slow,” Cava adds. “A lot of these quantum spin liquid materials are subject to constraints: They should be in a triangular lattice or a Kagome lattice. If the materials satisfy those constraints, the quantum researchers get excited; it’s a mandatory but not sufficient condition. So, by generating many, many materials like that, it immediately gives experimentalists a whole lot or hundreds more candidates to play with to speed up quantum computer materials research.”
“This work presents a brand new tool, leveraging machine learning, that may predict which materials may have specific elements in a desired geometric pattern,” says Drexel University Professor Steve May, who was not involved within the research. “This could speed up the event of previously unexplored materials for applications in next-generation electronic, magnetic, or optical technologies.”
The researchers stress that experimentation continues to be critical to evaluate whether AI-generated materials could be synthesized and the way their actual properties compare with model predictions. Future work on SCIGEN could incorporate additional design rules into generative models, including chemical and functional constraints.
“Individuals who want to vary the world care about material properties greater than the soundness and structure of materials,” Okabe says. “With our approach, the ratio of stable materials goes down, however it opens the door to generate a complete bunch of promising materials.”
The work was supported, partly, by the U.S. Department of Energy, the National Energy Research Scientific Computing Center, the National Science Foundation, and Oak Ridge National Laboratory.
