Scientists at Yale University have developed a new way to build computer chips that work more like the human brain—and, importantly, can grow much larger and more powerful than previous attempts.
This breakthrough could transform fields such as artificial intelligence, robotics, and brain-inspired computing.
The study was recently published in Nature Communications.
These special chips are known as neuromorphic chips.
Unlike traditional computer chips that process information using fixed, continuous electrical signals, neuromorphic chips mimic how neurons in the brain fire signals, or “spike,” one at a time.
Because they only activate when needed, they use far less energy and can handle certain tasks—such as processing enormous amounts of data spread across different locations—more efficiently than conventional computers.
Although neuromorphic technology has huge potential, researchers have struggled with one key challenge: scalability.
To make a very large system, many separate chips must work together smoothly. But until now, most neuromorphic chips have required a global synchronization system—a kind of master clock that forces all artificial neurons and synapses to operate in perfect step.
This ensures repeatable results but creates major limits. Everything on the chip must slow down to match the slowest part of the system, and the synchronization signal must travel across the entire chip network. As chips grow larger, this becomes more difficult, less efficient, and less reliable.
To overcome these problems, the Yale team led by Professor Rajit Manohar created a new system called NeuroScale. Instead of forcing the entire network to synchronize at once, NeuroScale uses local synchronization.
Only the groups of artificial neurons and synapses that are directly connected need to coordinate with each other. This more natural approach is closer to how the human brain actually functions.
Ph.D. candidate and lead author Congyang Li explains that NeuroScale relies on a “local, distributed mechanism” to keep the system working smoothly. Because synchronization happens only where it is needed, the system can grow much larger without slowing down.
The researchers note that the only limits on NeuroScale’s growth are the same ones that would limit a biological brain—meaning it has the potential to scale to enormous sizes.
The next step for the team is to build an actual NeuroScale chip. So far, the system has been tested through simulations and prototypes. They also plan to explore a hybrid approach that blends NeuroScale’s local synchronization with features from existing neuromorphic chips.
If successful, this technology could lead to brain-like computing systems with billions of artificial neurons, offering new levels of speed and efficiency for AI and other advanced applications.
