Researchers at UC San Diego have unlocked a method for controlling how a material “learns” and “forgets,” using vanadium dioxide to construct a neuristor, a device designed to mimic brain function. By observing the material while applying electricity with a newly refined coherent X-ray diffraction technique, the team discovered that jamming the flow of electrons enables learning, while restoring free-flow erases memory, a dynamic likened to the movement of passengers on a crowded platform.

“The research provides an exciting avenue toward developing energy-efficient devices that mimic neurons that can learn and forget at the device level,” said Nicolò D’Anna, a postdoctoral researcher at UC San Diego. The findings, published September 24, 2026 in Nature Communications, offer a pathway toward energy-efficient, brain-inspired computing hardware.

Phase Transitions Enable “Learning” and “Forgetting” in Neuristors

Vanadium dioxide is the core material in the newly controlled neuristors, artificial devices designed to replicate neuronal function, allowing researchers to manipulate the transition between fluid and jammed states within the material. This control stems from observations made using a refined coherent X-ray diffraction technique, which enabled visualization of the neuristor’s internal dynamics while electrical current was applied. The team discovered a strategy to make the material learn and forget a specific state, opening an avenue toward developing energy-efficient brain-inspired hardware.

This mechanism draws a compelling parallel to pedestrian movement on a crowded platform; when few people are present, movement is unhindered. But increased density creates interconnectedness, where one person’s actions influence others. Similarly, within the neuristor, “jamming-controlled stochasticity in metal-insulator switching”, as the researchers describe it, alters electron flow and establishes a memory state. Restoring free electron movement reverses this process, mirroring the erasure of a memory. The findings, published in Nature Communications on September 24, 2026, demonstrate a pathway toward building hardware that more closely mimics the efficiency and adaptability of the human brain.