Peter Hirschfeld is a member of the team including Richard Hennig, Jason Gibson and Ajinkya Hire whose work was one of six inventions recognized by UF Innovate | Tech Licensing.
AI-accelerated workflow for superconductors recognized by UF Innovate
Faster. Smarter. Revolutionary.
These three words capture the impact of an invention from Professor Richard Hennig in the Department of Materials Science and Engineering, Distinguished Professor Peter Hirschfeld in the Department of Physics, and their graduate student collaborators Jason Gibson and Ajinkya Hire at the University of Florida.
It was one of six inventions recognized Wednesday as UF Innovate | Tech Licensing celebrated the achievements of UF innovators at its eighth annual Standing InnOvation.
The event honored 796 innovators, all of whom contributed in some way to 435 new technologies disclosed, 130 licenses, 130 issued patents, or the 9 new startup companies. Eric Wang, Ph.D., was named Innovator of the Year for his work on all aspects of neuromuscular and neurological diseases, as well as RNA biology.
The discovery by Hennig and Hirschfield related to superconducting materials, which conduct electricity without resistance. This work has traditionally been slow and resource heavy. This invention introduces a workflow that applies artificial intelligence to make the discovery process faster and more efficient.
Built around a deep learning system called the Bootstrapped Ensemble of Equivalent Graph Neural Network, or BEE-NET, the model predicts critical temperatures of superconductors using data on crystal structures and phonon density of states. It is trained on different data types, including crystal structure only and coarse phonon density of states, and evaluated with loss functions like mean squared error, weighted mean squared error, and Earth Mover’s Distance to improve accuracy and reliability.
The invention combines BEE-NET with machine learning and physics-based simulations to filter materials from large databases based on formation energy, bandgap, critical temperature, and energy stability. The most promising candidates are then refined using Density Functional Theory simulations. The workflow has identified over 700 potentially stable superconductors. Two of these materials, Be2HfNb2 and Be2Hf2Nb, have been successfully synthesized and confirmed in the lab.
By combining AI and materials science, this invention brings us closer to next-generation superconductors.
