Chinese scientists have developed a laser-based artificial neuron that operates at unprecedented speeds, potentially revolutionizing artificial intelligence applications. The research team, led by Chaoran Huang from the Chinese University of Hong Kong, published their findings in the journal Optica.
The newly developed "laser graded" neuron processes data at a speed of 10 GBaud, which is approximately one billion times faster than natural neurons. The breakthrough could lead to significant advancements in AI systems and advanced computing due to its ultrafast data processing speeds and low energy consumption.
"Our laser graded neuron surpasses the speed limits of current photonic versions of spiking neurons and has the potential for even faster operation," said Huang, as reported by Science Daily. The laser neuron emulates the functions, dynamics, and information processing of biological graded neurons, providing superior speed and accuracy.
The research team created a reservoir computing system using the developed laser neurons, demonstrating exceptional performance in AI tasks such as pattern recognition and sequence prediction. The system detects arrhythmias with a 98.4% accuracy rate and processed 100 million heartbeats per second.
"With powerful memory effects and excellent information processing capabilities, a single laser-graded neuron can behave like a small neural network," Huang stated, according to La Stampa [https://www.lastampa.it/salute/2024/12/20/news/neurone_artificiale_laser_velocita_luce-423898883/]. This means that even without complex connections, a single laser graded neuron can perform machine learning tasks with high performance.
Most laser-based artificial neurons developed so far have been spiking photonic neurons, which have limitations in response speed, can suffer from information loss, and require additional laser sources and modulators. The new laser graded neuron overcomes these limitations by simulating the operation of graded neurons, using very fast laser light pulses to process signals in a precise and continuous manner.
To achieve faster performance, the researchers injected radiofrequency signals into the saturable absorption section of the quantum dot laser, allowing them to avoid delays that limit the response speed of photonic spike neurons. They designed high-speed radiofrequency pads for the saturable absorption section, enabling a faster, simpler, and more energy-efficient system.
"Therefore, even a single laser graded neuron without additional complex connections can perform machine learning tasks with high performance," Huang explained, as reported by Scienze Notizie. The neuron-like nonlinear dynamics and fast processing speed make the laser graded neuron ideal for supporting high-speed reservoir computing and provide more effective use in artificial intelligence applications.
The reservoir computing system demonstrated superior success in tasks like image classification due to its high-speed data processing capability. It shows excellent pattern recognition and sequence prediction across various AI applications.
"In this work, we used a single laser graded neuron, but we believe that cascading multiple laser graded neurons will further unlock their potential, just as the brain has billions of neurons working together in networks," Huang said, according to Science Daily. The team is working to improve the processing speed of the laser graded neuron while developing a deep reservoir computing architecture that incorporates cascaded laser graded neurons.
"We expect that integrating our technology into edge computing devices, which process data close to its source, will enable faster and smarter AI systems that better serve real-world applications with reduced energy consumption in the future," Huang added, as reported by La Razón.
Biological neurons are divided into two basic types: graded neurons, which encode information through continuous changes in membrane potential for subtle and precise signal processing, and spiking neurons, which transmit information using all-or-nothing action potentials, creating a more binary form of communication. The laser graded neuron simulates the operation of graded neurons, providing superior speed and accuracy.
This article was written in collaboration with generative AI company Alchemiq