Neuromorphic computing, in other words, tries to replicate the architecture and functionality of the human brain. Now, it is starting to emerge as a powerful tool with much broader potential than computer scientists previously recognized. Recent developments reveal that this technology, traditionally used for brainlike applications such as real-time sensor data processing, is capable of solving complex mathematical problems. Scientists and engineers from Sandia National Laboratories and the University of Manchester are leading the way to understand these capabilities.
That’s the promise of neuromorphic computing, according to Steve Furber, an emeritus computer scientist at the University of Manchester. He argues that its real-world applications could go much further than those original uses. The technology’s potential to solve previously infeasible mathematical problems could redefine how scientists approach computational challenges, fundamentally changing the rules of the game.
Advances in Neuromorphic Hardware
To their credit, the Sandia team really demonstrated their chops with an impressive implementation of the Monte Carlo method on neuromorphic hardware. This method made random sampling easier for numerical calculation. Brad Theilman and Felix Wang, both researchers in neuromorphic computing, recently worked on a neuromorphic computing core at Sandia. They focused instead on its extraordinary ability to reproduce or solve differential equations.
Intel’s Loihi 2, a neuromorphic chip directly inspired by the brain, has recently proven successful for solving partial differential equations. Loihi 2 proves adept at addressing sophisticated challenges, particularly those making use of the finite element method. This approach is foundational in the larger technical domain of scientific computing. These developments are a sign that neuromorphic systems can do better than traditional approaches, particularly when it comes to energy efficiency.
“There’s no reason to assume you can’t do something in neuromorphic computing.” – James B. (Brad) Aimone
As a visual artist, Aimone’s background in neuroscience colors his artistic practice. He laments that even with the huge advances in artificial intelligence, we still have scientists building power plants instead of realizing the construction of neuromorphic systems. These contributions, combined with the energy efficiency of neuromorphic hardware, could yield benefits beyond approaches that use standard computation methods optimized for conventional hardware.
Energy Efficiency and Mathematical Problem Solving
The human brain is the most efficient computer ever created, operating on approximately 10 watts of power. Problems which have traditionally required sophisticated techniques such as the finite element method are handled smoothly and gracefully. Neuromorphic systems, such as Loihi 2, seek to achieve this same kind of efficiency. The finite element method has long been a staple in scientific computing, allowing for the approximation of solutions to complex equations.
Neuromorphic computing’s event-based design enables it to process information in a more fluid, instantaneous manner, much like the way the brain interprets and responds to real-time data. Theilman noted the complexity involved in these computations, using the example of a brain controlling muscles to respond to real-time information:
“It’s a complex problem. The brain is controlling muscles in response to real-time information to make contact with the ball.” – Bradley Theilman
Neuromorphic hardware such as SpiNNaker, which is based on ARM, provides really cool opportunities for future work. It’s especially powerful for modeling diffusion of heat or other substances.
The Future of Neuromorphic Computing
As researchers unlock what’s possible with neuromorphic computing, its potential uses will multiply by orders of magnitude. Now the main thrust of these efforts is improving existing methodologies and demonstrating that this technology can indeed solve complex mathematical challenges.
Conventional architectures ruled the day in computational heavy lifting. Today, these types of tasks can be effectively handled by neuromorphic frameworks. Aimone emphasizes this potential:
“It’s worth looking deeply at any kind of mathematical problem.” – James B. (Brad) Aimone
The implications of these recent advances would catapult a wide variety of fields—including artificial intelligence, engineering, and more—to unprecedented heights. By focusing on where neuromorphic systems can and should do things better, the scientific community would be tapping into a fundamental new source of efficiency and capability.