Neuromorphic hardware solving engineering problems.

Finite element problem calculations In the future, computing will not be made of increasingly smaller chips, but of machines that think like a human brain, that idea has just taken a decisive step towards reality. National Laboratories Sandia Researchers from the United States demonstrated a new algorithm capable of using neuromorphic hardware to solve partial differential equations. These equations are one of the most fundamental and heavy-duty types of calculations in modern science and are behind climate simulations, fluid dynamics, materials behavior, and even nuclear models. Until now, it was believed that neuromorphic computing was good only at pattern recognition or machine learning, but inadequate for complex mathematics.

This new work breaks precisely that barrier. Current supercomputers consume colossal amounts of energy and with the exposure of artificial intelligence that demand continues to grow. On the other hand, neuromorphic computers are directly inspired by the human brain, which solves problems at a scalable level consuming just a few Watts. Externally they look like common computers, internally they are radically different. Its circuits imitate biological neural networks working in parallel, continuously and extremely efficiently. Neuroscientists Bradley Theilman and James Aimone developed an algorithm that preserves the dynamics of the brain's cortical networks and at the same time solves partial differential equations. The connection between these two worlds, neuroscience and applied mathematics, was only perceived now, more than a decade after the creation of the original neural model. The implications go far beyond mathematics, if the brain is, in essence, a computational system, neurological diseases can be seen as failures of biological computing and understanding how calculations occur in neuromorphic architectures can open new avenues for studying Alzheimer's, Parkinson's and other brain disorders.