Researchers at the Chinese University of Hong Kong have created a laser-based artificial neuron that mimics the functionality of biological graded neurons, achieving a processing speed of 10 GBaud—about a billion times faster than biological counterparts. This innovation promises to transform artificial intelligence (AI) and computing, offering ultrafast data processing and energy efficiency. Unlike conventional spiking neurons that operate on an all-or-nothing response, this laser graded neuron relies on nuanced changes in signal processing, making it ideal for applications requiring precise data handling.
The team, led by Chaoran Huang, has demonstrated that their technology can process extensive datasets quickly, including 100 million heartbeats or nearly 35 million images in just one second. This capability could enhance decision-making in time-sensitive scenarios and improve AI systems integrated into edge computing devices.
The laser graded neuron overcomes the limitations faced by existing photonic spiking neurons, such as speed delays and additional resource requirements, by utilizing a new approach with radio frequency signals. Its performance in high-speed reservoir computing showcases its potential in AI tasks like pattern recognition and sequence prediction. Future work aims to cascade multiple neurons to enhance processing speed and functionality, mimicking the complexity of biological neural networks.
