Researchers have created an ultra-fast artificial neuron that can perform high-speed computing and AI tasks such as pattern recognition. (Image credit: Chaoran Huang, The Chinese University of Hong Kong)
Scientists have unveiled a new type of artificial neuron based on lasers that replicates the functionality of a biological nerve cell. This neuron could significantly speed up high-speed computing and artificial intelligence (AI), the researchers claim.
Artificial neurons mimic nerve cells by activating when they reach a certain threshold of information. When a biological neuron receives enough of the desired data, it produces an electrical impulse to communicate with neighboring neurons. Likewise, artificial neurons process and transmit computational information only after receiving a certain amount of corresponding electronic information.
Existing artificial neurons, known as photonic spiking neurons, respond to input signals with all-or-none spikes. However, the way the neurons receive these signals creates a time gap after each spike during which they cannot respond to new input. This short reset period limits the speed at which computations can be performed using artificial spiking neurons.
The new artificial neurons, by contrast, transmit information through “graded” signals of varying intensity. In a new study published online Dec. 19, 2024, in the journal Optica, the scientists used a system of graded neurons to overcome the speed limitations of spiking neurons. Like biological graded, or “non-spiking,” neurons, the laser system generated increasingly stronger output signals in response to successive stimuli, eliminating the need for the same reset period as spiking neurons. As a result, the new artificial neuron transmitted data up to 100,000 times faster than its spiking counterparts.
The researchers integrated a graded neuron into a reservoir computing system, a type of artificial neural network that processes time-dependent data. They used the system to analyze 700 heartbeats for arrhythmia. The reservoir processed these heartbeats at a rate of 100 million beats per second, significantly faster than spiking neural networks can. The new system identified arrhythmic patterns with more than 98% accuracy. In a separate experiment, the system analyzed and classified handwritten digits at a rate of nearly 35 million digits per second with 92% accuracy.
“With its outstanding memory effects and excellent information processing capabilities, a single laser graded neuron can function as a small neural network,” said study co-author Chaoran Huang, an engineer at the Chinese University of Hong Kong, in a commentary. “Thus, even a single laser graded neuron, without additional complex connections, can perform machine learning tasks with high efficiency.”
Combining multiple graded neurons could provide even more powerful computing capabilities. “In this work, we used a single laser-graded neuron, but we believe that cascading multiple such neurons will further unlock their potential, similar to how billions of neurons in the brain work together in networks,” Huang said.
“Our technology can accelerate AI decision-making in mission-critical applications while maintaining high accuracy,” Huang added. “We hope that implementing our technology in edge computing devices — which process data close to its source — will lead to faster, smarter AI systems that better meet the needs of real-world applications with lower power consumption in the future.”
Sourse: www.livescience.com
