Neuromorphic computers modeled after the human brain can now solve the complex partial differential equations behind physics simulations—a breakthrough once thought possible only with massive energy-hungry supercomputers, researchers at Sandia National Laboratories announced.
In a paper published February 14 in Nature Machine Intelligence, computational neuroscientists Brad Theilman and Brad Aimone describe a novel algorithm enabling neuromorphic hardware to tackle PDEs, the mathematical foundation for modeling fluid dynamics, electromagnetic fields, and structural mechanics.
"For decades, experts believed neuromorphic computers were best suited for pattern recognition or accelerating neural networks," Theilman explained. "They weren't expected to excel at rigorous mathematical problems like PDEs. We've proven they're shockingly good at it."
The breakthrough could pave the way for the world's first neuromorphic supercomputer, dramatically reducing the energy costs of scientific computing. Traditional supercomputers consume enormous power to solve these equations, while neuromorphic systems use brain-like spiking neural networks that operate far more efficiently.