Researchers at Sandia National Laboratories, led by James B. Aimone, have made significant strides in neuromorphic computing that could revolutionize how mathematical problems are solved. Their recent publication in Nature Machine Intelligence details a groundbreaking approach to solving differential equations using neuromorphic hardware developed by Intel. This important work implies that the potential use cases for neuromorphic computing should be much broader than what’s being presupposed today by well-meaning computer scientists.
Use what works Aimone noted that unfortunately, the state-of-the-art computational methods have been tuned to traditional hardware, which can bloat their efficiency. He emphasized the potential of neuromorphic systems, saying, “There’s no reason to assume you can’t do something in neuromorphic computing.” Such a viewpoint emphasizes the need to pursue new pathways in the field of computational science.
The Integration of Neuroscience and Computing
The Sandia research team was singularly focused on applying the finite element method. This powerful tool of scientific computing addresses some of the world’s most complicated physical problems by simulating processes not unlike those occurring in the human brain’s motor cortex. This model was then robustly deployed on the Loihi 2 platform, a neuromorphic chip developed with the goal of replicating processing found in the brain.
“Loihi 2 is well-suited for solving problems like differential equations,” noted Steve Furber, an advocate of neuromorphic approaches. The chip includes a novel architecture specifically optimized to solve partial differential equations. This groundbreaking feature underscores its versatility for varied uses in numerous branches of science.
Aimone’s background in neuroscience is foundational to this research. The team’s approach stems from emulating the principles of how the human brain works. Their goal is to harness the efficiency of computing systems and real-time problem solving capabilities. “The brain is controlling muscles in response to real-time information,” explained Bradley Theilman, highlighting the complexities involved in mimicking such biological processes.
Practical Applications and Future Potential
The implications of these advancements are profound. Specifically, Aimone focused on the leaps forward in the field of artificial intelligence (AI). Yet, he pointed out that much of the new attention is still heavily focused on old computing paradigms. “We have made tremendous advances in AI, but people are building power plants,” he remarked, indicating a need to shift towards more energy-efficient solutions that neuromorphic computing can provide.
The Sandia team’s prior research on applying the Monte Carlo method on neuromorphic hardware is a great example of this transition. The research community can now harness Loihi 2’s power to address the challenges in science. They will investigate novel algebraic approaches that allow for fast and precise computation.
The finite element method has been adapted to a model of the motor cortex. This foundational research has the potential to spur revolutionary advances in engineering and applied physics. Theilman was surprised to find such a clear connection to the brain’s processing capabilities in mathematical modeling. He remembered that when writing finite element methods, working with matrices seemed very much like working with a number in a model.