The MATRIX AI Consortium at the University of Texas at San Antonio (UTSA) announced this week it received a $4 million grant from the National Science Foundation to fund “The Neuromorphic Commons (THOR)” project.
The THOR project is a multi-university initiative giving researchers access to large-scale heterogeneous neuromorphic computing hardware systems.
“Our vision is to foster interdisciplinary collaborative research on the neuronal foundations of biological intelligence, covering the full spectrum from perception, decision making, and continual learning in the physical world,” the project’s website says.
UTSA says the THOR project is expected to catalyze a transformation in algorithm design, hardware and software co-design, and neuromorphic applications, similar in scale to the impact seen when HPC systems became accessible to the engineering research community. The system will be accessible for research in various domains including artificial intelligence and machine learning, physics, life sciences, and computational neuroscience.
Dhireesha Kudithipudi, THOR Principal Investigator and Director of the MATRIX AI Consortium, said the group plans to design a national hub for open access large scale neuromorphic platforms through industry partnerships.
“The field is at a pivotal moment and ensuring access to a broader group of researchers is critical at this stage. This initiative reflects a community-driven approach, shaping a framework designed by and for the community,” she said in a release.
The THOR team also plans to develop training and education materials to cover the fundamentals of neuromorphic learning algorithms and systems and to make these resources available through open platforms.
Neuromorphic computing is an architecture that uses hardware and algorithms inspired by the human brain, particularly the neocortex, where high-level functions like spatial reasoning, sensory perception, and language take place. Neuromorphic systems use intricate spiking neural networks of chips using artificial neurons and synapses to process information and solve problems. The networks simulate how biological neurons transmit information through discrete spikes over time, allowing the system to process temporal patterns in data.
Interest in neuromorphic computing has fluctuated in recent times, especially with the push for quantum, as well as the current GPU-driven hardware landscape that has supercharged the capabilities of classical computing and taken us into the exascale era. But spiked neural networks still hold promise for low energy and low latency computing which could boost several technologies, including AI, where these techniques can enhance machine learning algorithms with more efficiency and flexibility. Large manufacturers are on also board: IBM released a neuromorphic chip in 2023 called NorthPole that the company claims is 25x more energy efficient than current chip technology. And earlier this year, Intel announced the Hala Point system in collaboration with Sandia National Laboratories which is powered by 1,152 of its own Loihi 2 neuromorphic processors.
NSF Program Director Andrey Kanaev said the agency’s award for the THOR project is crucial in advancing the NSF’s mission to drive innovation and broaden access to research resources. “By making bio-inspired computing resources available to a wider community of researchers in computer science, neuroscience, and computational physics, this project will contribute to democratizing access to advanced tools and fostering breakthroughs in energy-efficient, resilient AI through neuromorphic computing,” he said.
The core team of researchers driving this interdisciplinary collaborative effort include Dhireesha Kudithipudi, Principal Investigator, University of Texas San Antonio; Catherine Schuman, Co-Principal Investigator, University of Tennessee Knoxville; Gert Cauwenberghs, Co-Principal Investigator, University of California San Diego; and Vijay Janapa Reddi, Senior Personnel, Harvard University.
