Steve Furber, a distinguished computer scientist emeritus at the University of Manchester, has announced significant advancements in the realm of neuromorphic computing. His research extends the landmark work of a pod of researchers and practitioners with the Battleground Collaborative at Sandia National Laboratories. That same team eventually introduced the Monte Carlo approach to solving differential equations. As our recent breakthrough demonstrates, neuromorphic systems have the potential to be truly remarkable. They can now solve highly sophisticated mathematical problems at power levels commensurate to that of the human brain.
The University of Manchester has long been associated with Furber’s pioneering work in neuromorphic computing, which emulates the brain’s architecture and functioning. Notably, the human brain operates at an astonishingly low power consumption of approximately 10 watts, making it a prime model for efficient computing. The Sandia neural network team has pioneered this project by developing a neural net-FEM model for motor cortex. Recreating this model on Intel’s Loihi 2 neuromorphic computing core was driven to success by researchers Brad Theilman and Felix Wang.
Building on Previous Work
Furber’s team has achieved some success in harnessing ARM-based SpiNNaker hardware to model heat diffusion. This accomplishment is a demonstration of the incredible potential versatility of neuromorphic systems. This new body of work takes those methodologies even farther. More broadly, it’s a demonstration of how amenable neuromorphic computing can be to a variety of mathematical challenges.
Furber points out that this research is a natural extension of the Sandia team’s previous work. Since they had already used the Monte Carlo method implemented on neuromorphic hardware, this was a natural next step. The Monte Carlo method Monte Carlo methods are especially valuable in tackling complicated mathematical problems through the simulation of random variables.
“It’s worth looking deeply at any kind of mathematical problem,” – James B. (Brad) Aimone
Aimone’s visionary thoughts are indicative of the broader promise of neuromorphic computing. His biggest point is that there are no fundamental barriers to using these systems for highly complex mathematical operations.
Neuromorphic Computing and Real-Time Applications
In practice, what sets neuromorphic computing apart is its prioritization on applications with brain-like capabilities. These systems have unique strengths in processing real-time sensor data, an increasingly important capacity across disciplines and industries, from robotics to artificial intelligence. The combination of real-time data processing capabilities and high-fidelity mathematical models provide a unique opportunity to innovate.
Theilman elaborates on the complexities involved in neuromorphic computing, stating, “It’s a complex problem. The brain is controlling muscles in response to real-time information to make contact with the ball.” This complexity illustrates the deep relationship between neuromorphic systems and biology itself. Consequently, these systems allow the development of a wider range of more computationally efficient and effective strategies.
The Sandia team’s use of FEM on Loihi 2 serves as a perfect example of the problem neuromorphic hardware is well-suited to tackle—partial differential equations. This technique fulfills the computational requirements of tasks that require rapid responses to changing data streams. That’s most useful for applications where the environment is constantly changing.
The Future of Neuromorphic Systems
As progress marches on in this exciting new field, researchers are excited about the future applications of neuromorphic computing. Aimone expresses confidence in the field’s potential: “There’s no reason to assume you can’t do something in neuromorphic computing.” This feeling is shared throughout the scientific community as researchers push the boundaries of what’s possible with the new computational capabilities.
With every step made in this field, the divide between classical computing and cognitive computing closes in. More than computing efficiency, these developments have allowed like all other technologies being pushed by their advocates, they promise to transform any industry that relies on smart real-time data analysis and adaptive response.