AI data centers consume electricity at a growing rate, and finding a sustainable way to run them has become urgent. A new scientific review proposes one path: building chips that mimic how the human brain works.
The study comes from researchers at the University of Ottawa, Columbia University and Hanyang University. Its lead author is Dr. Kannan Udaya Mohanan, a postdoctoral fellow in Professor Chang-Hyun Kim's group. Published in Advanced Functional Materials, the work examines so-called "post-CMOS" neurons, devices built from emerging materials that go beyond conventional silicon chip architectures.
The core idea is to make hardware resemble the brain, with neuromorphic circuits that physically reproduce how neurons sum incoming signals and turn them into a response. The researchers argue these devices could sharply reduce the energy needed compared with conventional processors, which separate memory and computation and constantly shuttle data back and forth.
The review does not describe a finished product but a set of candidates and open problems. Moving from the lab to large-scale production remains to be demonstrated, as does compatibility with semiconductor manufacturing processes already in use. The authors themselves point to scalable circuits as the necessary condition for the approach to become industrial.
Possible applications extend beyond data centers. According to the review, the same hardware family could serve low-power wearable devices, personalized health technologies and more natural human-machine interfaces.
The topic fits into a broader race. In recent months several projects have tried to lower the power draw of AI systems, from photonic chips to neuromorphic prototypes tested on specific tasks. What sets this review apart is its ambition to treat the brain as a model for the entire chip architecture, not just a single component.
No comments yet. Be the first!