Nicolò D'Anna was watching electrons stumble like commuters on a packed train platform when he realized—this wasn’t chaos. It was memory. At UC San Diego, D'Anna and physicist Oleg Shpyrko were probing a tiny device made of vanadium dioxide, a material that can switch between conducting electricity and blocking it, much like neurons fire or stay quiet in the brain. What they found wasn’t just a switch—it was a system that could learn and forget, controlled by something unexpected: jamming.
Most computers today run on rigid binary logic, but the brain works differently. It learns by adjusting connections, storing patterns in a messy, dynamic way. Scientists have long wanted to build neuromorphic computers—machines that mimic this brain-like flexibility—but they’ve struggled to create hardware that can adapt without guzzling energy. The breakthrough from UC San Diego offers a new path. By using coherent X-ray diffraction, a powerful imaging method sensitive to nanoscale changes, the team watched how vanadium dioxide reorganizes itself when electricity flows through it. They saw that when the material’s internal domains—tiny regions with different electronic properties—become crowded and jammed, like people packed on a train platform, the system gets stuck in a state. That stuck state is memory. When the jam clears, the memory fades. Learning and forgetting aren’t bugs—they’re built into the physics.
The team’s neuristor, a device designed to act like a neuron, uses this jamming effect to control when it remembers a state and when it lets go. In experiments, they could reliably trigger the transition between free-flowing and jammed states by adjusting electrical pulses. This isn’t just theoretical: the study, published in Nature Communications, shows that memory emerges from the collective behavior of nanoscale domains, offering a physical basis for adaptive computing. Vanadium dioxide’s ability to switch between metal and insulator states has been known for decades, but this is the first time scientists have linked that switch to jamming-driven memory.
The implications go beyond faster or smarter computers. Neuromorphic devices based on this principle could use far less power than today’s AI hardware, which often requires massive data centers. If memory is baked into the material itself, there’s no need to shuttle data back and forth between separate memory and processing units—a major source of energy waste. This could one day lead to intelligent sensors, medical implants, or robots that learn on the fly without draining batteries.
We’re not there yet, but the physics is now clearer. As Shpyrko puts it, 'We’ve found a way to control when the system learns and when it forgets—just by managing how crowded the internal landscape gets.' In the quest to build machines that think more like we do, sometimes the answer isn’t in code, but in the quiet jam of a crowded platform.