Current artificial intelligence performance relies heavily on energy-intensive computing architectures that are difficult to miniaturize. A team of researchers is now proposing a radically new approach: memristic artificial neurons capable of communicating with each other through light, paving the way for ultra-compact and highly efficient three-dimensional neural networks.
In a recently published article in Nature Electronics, the researchers demonstrate that it is possible to design photonically interconnected neural networks based on nanoscale memristic devices. These components mimic the functioning of biological neurons while overcoming the limitations of conventional electronic technologies.
This breakthrough is the result of an international collaboration in which the Carnot Interdisciplinary Laboratory of Burgundy (ICB, CNRS / University of Burgundy Europe) participated, through the work of Alexandre Bouhelier.
Nanoscale Blinking Neurons
The artificial neuron developed is based on a silver/PMMA/silver memristive junction (metal-insulator-metal). Its operation relies on the formation and rupture of conductive filaments at the atomic scale. When a critical number of electrical signals is reached, the neuron emits a light pulse, hence the term “blinking neuron.” This photonic emission is a major advantage: it allows information to be transmitted without complex electrical wiring or bulky readout circuits. The device thus acts simultaneously as a neuron, synapse, and optical emitter, with an extremely small footprint of 170 nm × 240 nm.
Toward Three-Dimensional and Scalable Neural Networks
Thanks to these blinking neurons, researchers have designed a three-dimensional neural network interconnected by light, capable of processing complex information. The network notably achieved an accuracy of 91.51% for a speech recognition task using the Google Speech dataset. Furthermore, a dense matrix of artificial neurons, spaced only one micrometer apart, enabled the classification of handwritten digits from the MNIST dataset with an accuracy of 92.27%.
A Promising Advance for Neuromorphic AI
These results open up significant prospects for the development of neuromorphic systems, that is, hardware architectures inspired by the workings of the human brain. By combining extreme miniaturization, low power consumption, and photonic communication, these memristive neurons could contribute to the next generation of artificial intelligence systems: faster, more compact, and more durable.
Zhou, Y., Fang, Y., Gisler, R. et al. Photonically linked three-dimensional neural networks based on memristive blinking neurons. Nat Electron (2026).
https://doi.org/10.1038/s41928-025-01529-5
