Researchers develop ultrafast laser neurons, mimicking biological functions to upgrade AI with improved speed and energy efficiency.
Researchers at the Chinese University, Hong Kong have developed a laser-based artificial neuron that mimics biological graded neurons and processes data at an unprecedented speed of 10GBaud. This innovation in photonic computing could reshape artificial intelligence by enabling faster decision-making and improved energy efficiency.
The innovation centers on a quantum-dot laser that emulates graded neurons—biological cells known for their nuanced signal transmission through continuous changes in membrane potential. Unlike spiking neurons, which rely on binary action potentials, these artificial neurons promise smoother and more efficient data processing.
“By overcoming the speed barriers of photonic spiking neurons, our laser graded neuron achieves unmatched performance,” said Dr Chaoran Huang, the study’s lead researcher. The team’s laser neuron operates without the delays typical of earlier models, thanks to the injection of radio-frequency signals into a specialised quantum-dot laser section.
The technology is designed to cater to industries requiring ultrafast decision-making and processing, including healthcare, finance, and autonomous systems. For instance, its ability to process 100 million heartbeats per second makes it invaluable for medical diagnostics, while its pattern recognition capabilities can benefit image analysis in surveillance and robotics.
Graded neurons excel in AI tasks like image recognition and pattern prediction. The researchers demonstrated the neuron’s capability by processing 100 million heartbeats and over 34 million handwritten images within a single second. “Our system could transform AI applications in areas demanding real-time decision-making,” Huang added.
This technology also forms the foundation for high-speed reservoir computing, a method ideal for analysing time-dependent data, such as speech recognition or weather forecasting. The reservoir computing system created using the laser neuron exhibited exceptional accuracy, achieving 98.4% in detecting arrhythmic heartbeats.
One of the key advantages of this approach is its simplicity and efficiency. Unlike traditional photonic neurons, which require additional lasers and modulators, the laser graded neuron offers a compact and energy-saving alternative. “Even a single laser neuron behaves like a small neural network,” Huang noted, suggesting potential scalability for complex machine-learning tasks.
Future efforts aim to cascade multiple laser neurons to simulate brain-like networks. The researchers also plan to enhance processing speeds and integrate deep reservoir computing architectures, opening possibilities for broader AI applications while maintaining reduced energy consumption.
