Engineers at the University of California San Diego have developed a new type of computer hardware inspired by how the human brain processes information.
The device could help solve one of the biggest challenges in artificial intelligence today: creating faster and more energy-efficient systems that can run on small devices such as wearable health monitors, smart sensors, and autonomous machines.
The research, published in Nature Nanotechnology, describes a computing platform designed to mimic certain principles of brain activity.
Instead of separating memory and processing like traditional computers do, the new system combines both functions within the same device.
This design reduces the need for data to constantly move between different parts of a computer, which normally consumes large amounts of time and energy.
Modern computers rely on a structure where memory and processing units are separate. Every time a program runs, information must travel back and forth between these parts.
As artificial intelligence systems grow larger and more complex, this constant movement has become a major bottleneck.
The brain, however, handles information very differently. In the brain, memory and processing occur together within networks of neurons.
The new platform was developed by a team led by Duygu Kuzum, a professor of electrical and computer engineering at UC San Diego. The work belongs to a field called neuromorphic computing, which aims to build machines that borrow ideas from the brain’s structure and behavior. The researchers emphasize that their system is inspired by the brain rather than trying to copy it exactly.
Many earlier neuromorphic designs focused on creating artificial versions of individual neurons or synapses and linking them together in fixed circuits. But the brain’s real power comes from large networks of neurons interacting continuously with one another. Learning and decision-making emerge from these complex interactions across space and time.
To better capture this idea, the researchers built a system where many computing nodes interact with each other through the same physical material. The platform uses a special quantum material called hydrogen-doped neodymium nickelate. When hydrogen ions are introduced into the material, they form small clusters beneath metal electrodes on its surface. Electrical pulses cause these ions to move, which changes the material’s electrical resistance.
This movement allows the system to store information for short periods, giving it memory-like behavior. At the same time, the nodes are connected through a shared substrate beneath them. Signals from one part of the system can influence activity elsewhere, creating collective behavior across the network.
The device processes information using a method called spatiotemporal computing. This means it analyzes signals both over time and across different locations within the network. Incoming signals are first converted into electrical spikes. The network then transforms these signals into complex patterns that capture both timing information and interactions among nodes.
The researchers tested the platform on two simulated tasks. In one experiment, the system recognized spoken digits. In another, it detected early signs of epileptic seizures using brain-wave recordings from electroencephalograms. In both cases, the system showed improvements in speed, accuracy, and energy efficiency compared with methods that only analyze signals over time.
The system also operates extremely quickly, responding in hundreds of nanoseconds while using very little energy per operation. This efficiency could make it especially useful for “edge AI,” where devices must process information locally rather than sending it to large data centers.
Although the technology is still in its early stages, the researchers hope future work will expand the system to handle larger tasks and integrate it with conventional semiconductor electronics.
