2023 R&D 100 Award for SuperNeuro: An Accelerated Neuromorphic Computing Simulator

To address the current limitations and inefficiencies that limit large-scale neuromorphic computing, researchers at ORNL created SuperNeuro, a Python-based open software that provides AI practitioners with brain-like simulators that are fast and scalable on central and graphics processing platforms. Using matrix-based and agent-based modeling approaches, SuperNeuro allows for different workloads and provides the option of simulating the user’s own spiking mechanisms in a human-interpretable manner.

Compared with existing simulation platforms, SuperNeuro provides an indispensable neuromorphic software with increased adaptability, leveraging GPU computing to provide superior performance for neuroscience, spiking neural networks, or SNNs, and general-purpose computing workloads. Easy integration with other tools for SNN optimization opens the possibilities for codesign of neuromorphic circuits. SuperNeuro can be up to 300 times faster than other simulators for small sparse networks and up to 3.4 times faster on large sparse and dense networks.

Funding for this project was provided by DOE Office of Science.

ORNL’s Prasanna Date, Shruti Kulkarni and Chathika Gunaratne co-led the development. Research contributors included, ORNL’s Robert Patton, Mark Coletti and Thomas Potok.