UCSD - University of California - San Diego

03/11/2026 | Press release | Distributed by Public on 03/11/2026 09:17

Brain-inspired Device Could Lead to Faster, More Energy-efficient AI Hardware

Published Date

March 11, 2026

Article Content

A team led by engineers at the University of California San Diego has developed a new brain-inspired hardware platform that could help computer hardware keep pace with the explosive growth of artificial intelligence. By combining memory and computation on the same chip - and allowing its components to interact collectively like neurons in the brain - the brain-inspired platform improved the speed, accuracy and energy efficiency of pattern recognition in two simulated tasks: recognizing spoken digits and detecting epileptic seizures early from brain-wave recordings.

The approach could lead to the development of compact, energy-efficient hardware for smaller AI systems such as those used in wearable health monitors, smart sensors and other autonomous devices.

The work, published on March 9 in Nature Nanotechnology, falls within the field of neuromorphic computing, which aims to build machines that mimic how the brain processes information. The researchers emphasize that the technology is brain-inspired, rather than brain-like; it draws ideas from how neural networks interact but does not attempt to replicate the brain itself.

One key strategy in building brain-inspired devices is to combine memory and computation on the same piece of hardware. In most conventional computers, these functions are separate. That means data must constantly move back and forth between memory and processors. That movement consumes time and energy and has become a major bottleneck as AI models grow larger.

By taking inspiration from the brain, researchers led by Duygu Kuzum, professor in the Department of Electrical and Computer Engineering at the UC San Diego Jacobs School of Engineering, are developing new computing architectures that store and process information within the same system.

However, another bottleneck remains. Many existing neuromorphic technologies still focus on modeling individual elements of the brain - such as artificial neurons or synapses - and connecting them in predefined circuits. But in the brain, learning and computation do not arise from single neurons acting as isolated components, explained Kuzum, who is the study's senior author. Instead, they emerge from the rich, dynamic interactions of large networks of neurons collectively communicating across space and time.

Yue Zhou, a postdoctoral researcher in the Department of Electrical and Computer Engineering at the UC San Diego Jacobs School of Engineering, prepares a wafer for examination under a probe station.

To better capture this behavior, Kuzum and collaborators at Quantum Materials for Energy Efficient Neuromorphic Computing (Q-MEEN-C) - an Energy Frontier Research Center (EFRC) funded by the U.S. Department of Energy (DOE) - designed a computing platform where many nodes are physically connected through the same material and can influence one another across the network.

The platform is built from a hydrogen-doped perovskite nickelate called neodymium nickelate, a type of quantum material known for its unusual electronic properties.

When hydrogen ions are introduced into the material, they form tiny clouds beneath metal electrodes patterned on its surface. Applying voltage pulses causes the hydrogen ions to move within the material and change its electrical resistance. This motion gives the system memory-like properties. Each node can briefly retain information about recent signals, while separate programmable elements store longer-term information.

At the same time, all the nodes interact through the shared substrate beneath them. Activity at one location influences the behavior of others. This shared substrate loosely resembles the ionic fluid surrounding the neurons in the brain, where signals spread and influence nearby cells. Because of this connection, the output from any single node depends on what the rest of the network is doing. If neighboring nodes receive signals, the measured response changes. This creates collective behavior across the system, similar to communication across brain regions, explained study first author Yue Zhou, a postdoctoral researcher in Kuzum's lab at UC San Diego.

A view of a section of the wafer held in position by the probe station.

The device processes information using a strategy called spatiotemporal computing, which analyzes signals both over time and through spatial interactions across the network. Incoming signals are first converted into electrical spikes and sent into the network. Interacting nodes transform those signals into complex internal patterns that capture both timing information and network dynamics. A second layer of programmable junctions then reads those patterns and performs classification tasks.

The researchers demonstrated the approach using two simulated applications. In one, the system successfully recognized spoken digits with high accuracy. In another, it detected early signs of epileptic seizures from electroencephalogram (EEG) signals. In both cases, the system outperformed methods that relied only on time-based processing.

In the seizure-detection test, the system identified warning signals even when given only a few seconds of brain data. Because activity in one node influences others, early signals from a few channels can spread across the network and help the system detect seizures sooner.

Further, the system operates extremely quickly - on the scale of hundreds of nanoseconds - and uses very little energy, about 0.2 nanojoules per operation.

That efficiency could make the technology useful for edge AI applications, where small devices must process data locally with limited power instead of sending it to large data centers, Kuzum noted. Potential uses include wearable medical devices, smart sensors, audio processing systems and autonomous machines.

The technology is still at an early stage. Hardware demonstrations so far have focused on small-scale tasks, while larger tasks such as speech recognition and seizure detection were tested through simulations based on experimental measurements.

Future work will focus on scaling up the system; integrating it with conventional semiconductor electronics; and exploring additional applications.

Full study: Protonic nickelate device networks for spatiotemporal neuromorphic computing

This work was made possible by collaborations and funding provided by Q-MEEN-C, an EFRC funded by the U.S. DOE, Office of Science, Basic Energy Sciences, under Award No. DE-SC0019273. Senior collaborators include study co-authors Ertugrul Cubukcu, professor in the Aiiso Yufeng Li Family Department of Chemical and Nano Engineering at UC San Diego; Eva Andrei, professor in the Department of Physics and Astronomy at Rutgers University; and Shriram Ramanathan, professor in the Department of Electrical and Computer Engineering at Rutgers University. Q-MEEN-C is directed by Ivan Schuller, distinguished professor of physics at UC San Diego.

Learn more about research and education at UC San Diego in: Artificial Intelligence

A chiplet cut from the wafer containing a set of arrays from the spatiotemporal processing layer.
UCSD - University of California - San Diego published this content on March 11, 2026, and is solely responsible for the information contained herein. Distributed via Public Technologies (PUBT), unedited and unaltered, on March 11, 2026 at 15:17 UTC. If you believe the information included in the content is inaccurate or outdated and requires editing or removal, please contact us at [email protected]