McGill University researchers have developed a light-detecting nanoscale structure that mimics how a neuron processes information. The neuron-like behaviour emerges from the materials themselves, reducing the energy demand associated with similar devices that rely on circuits or software.
Instead of capturing data first and processing it elsewhere, the device senses and interprets light in the same place, similarly to how the eye processes visual information.
The researchers say the discovery could increase the efficiency of such vision-based technologies as artificial retinas and smart optical sensors. It could also transform how artificial neural networks (ANNs), a foundation of machine learning, are built.
"In our paper, using unique materials and nanostructure, we made for the first time a device that can closely mimic the neuron dynamics we'd see in a biological context," said Songrui Zhao, lead author and Associate Professor of Electrical and Computer Engineering.
Layered device responds to light
The researchers built the device by engineering layers of atoms using a technique called molecular beam epitaxy. They then exposed it to light with different colours, intensities and timing patterns, measuring how the electrical signals inside the material changed in response.
By analyzing these signals over time, they showed that the device can combine incoming inputs, store information briefly and trigger a response once a certain threshold is reached. This resembles how a single neuron processes information, demonstrating that such behaviour can emerge directly from the physics of the material, rather than from software or complex circuitry.
"By carefully engineering the layers, we created a device with a tunable response to light, which forms the basis for emulating how a single neuron behaves," Zhao said. "We were able to design the flow of electrical current to produce the behaviour we wanted."
Building neural networks from the ground up
Because ANNs are built from many connected neurons, the device could offer a new way to construct these systems, the researchers said.
"A single artificial neuron is like a cell you can use as a building block, allowing us to construct networks from the bottom up," Zhao said. "It's a bit of a crazy idea - to create something like a biological system using an inorganic material."
Such an approach could lead to more efficient forms of information processing, with potential applications in such areas as advanced computing.
Zhao said future studies will expand the device's light response range and performance and explore applications such as data encryption, where processing information directly at the sensor could improve security.
About this study
"Nanowire photodetectors: path to single physical artificial neurons," by Yunqiu Chen, Milad Fathabadi, Mohammad Fazel Vafadar and Songrui Zhao, was published in Nanoscale.
This research was funded by the Natural Sciences and Engineering Research Council of Canada and the Fonds de Recherche du Québec - Nature et Technologies.
