UCLA Pioneers Hardware That Acts as Its Own Neural Network

Physical AI emerges from self-organizing materials

Researchers at UCLA have advanced a fundamentally different approach to artificial intelligence in which the hardware itself becomes the neural network. Rather than running software on passive silicon chips, self-organizing networks of nanowires and nanoparticles physically adapt their structure to learn and process information.

Professor James Gimzewski and research scientist Adam Stieg, both of UCLA's California NanoSystems Institute, were among the first to demonstrate this approach. Their pioneering work now anchors a comprehensive review published in Nature Reviews Physics that examines how these systems could complement cloud computing by handling sensor data locally in resource-constrained environments.

The technology collapses the traditional separation between hardware and software. There is no neural network code executing on a processor. Instead, computation occurs directly within the material as it establishes physical connections measured in billionths of a meter.

How self-organizing networks learn

The systems draw inspiration from the human brain's cortex, though researchers emphasize the metaphor has limits. Nanowires or nanoparticles function analogously to neurons, while the changing electrical connections between them resemble synapses. Frequent electrical stimulation strengthens connections, creating persistent pathways similar to memory formation. Lack of stimulation degrades connections, mimicking forgetting.

UCLA introduced nanowire networks in 2011. A separate team led by physicist Simon Brown at the University of Canterbury in New Zealand unveiled nanoparticle networks in 2013. Both architectures have demonstrated machine learning capabilities including speech and image recognition, performing these tasks in real time by exploiting the physical behavior of the network rather than executing conventional algorithms.

"Silicon-based electronics have shaped how we think about computing, but they're not the only way to do it," Stieg said, according to the California NanoSystems Institute. "In our systems, the model evolves in the physical network itself. It adapts and changes."

Why it matters

As AI models scale, they demand exponentially more energy, water, and computing infrastructure. Meanwhile, satellites, autonomous vehicles, industrial robots, and distributed sensors need to process enormous volumes of data locally with limited power and connectivity. Self-organizing physical networks could address both challenges by enabling continuous, adaptive AI that operates efficiently at the edge without constant cloud communication. This represents a potential path toward sustainable AI growth that doesn't rely solely on building larger data centers.

Edge computing applications

The technology targets edge computing scenarios where sensors generate far more data than can be efficiently transmitted or processed remotely. Satellites, for example, often collect more information than they can send to Earth, requiring compression or filtering before transmission.

Self-organizing networks could become part of the sensing apparatus itself, adapting their structure in response to incoming signals and extracting relevant information physically. This would enable AI to operate continuously in resource-constrained environments without relying on distant servers.

"Most AI treats the hardware as a passive platform for running software," Stieg said. "We're asking what becomes possible when the hardware itself is adaptive — when the material reorganizes in response to information and becomes part of the learning process."

The Nature Reviews Physics article was authored by an international team including theoretical physicist Francesco Caravelli of the University of Pisa, experimental physicist Gianluca Milano of Italy's National Institute of Metrological Research, physicist Carlo Ricciardi of the Polytechnic University of Turin, and physicist Zdenka Kuncic of the University of Sydney, who has collaborated extensively with Stieg on developing the technology.

The research received support from the U.S. Department of Energy, New Zealand's MacDiarmid Institute and Marsden Fund, and European Union research programs. Details were first reported by the California NanoSystems Institute at UCLA.