Researchers at the University of Florida have developed a new chip that uses light instead of electricity to perform one of the most power-hungry tasks in artificial intelligence. Reported studies Advanced Photonicsdetailing systems that dramatically reduce energy usage and speed up processing.
As AI systems become the centre of technology, their electricity consumption poses a major challenge to sustainability. This new chip provides a potential solution by integrating optical components directly into the silicon chip to process core AI functions.
How the optical chip works
The chip is specially designed to perform convolution operations, a fundamental process of machine learning that allows AI systems to detect patterns of images, videos and text. These operations usually require a large amount of computing power.
The new system performs these convolutions using laser light and a microscope lens. Machine learning data is first converted to laser light on the chip. This light passes through two sets of miniature Fresnel lenses, a flat, ultra-thin version of the lens found in the lighthouse. This is etched directly into the tip and becomes narrower than human hair. These lenses perform mathematical transformations and the results are returned to the digital signal to complete the AI task.
“Performing critical machine learning calculations with energy close to Zero Energy is a step forward for future AI systems. This is important to continue to expand AI capabilities in the coming years.”
said research leader Volker J. Sauger, a professor at the University of Florida.
Performance and Benefits
In tests, the prototype chip successfully categorized handwritten numbers with an accuracy of approximately 98%, comparable to traditional electronic chips.
An important advantage of the Photonic approach is that it allows multiple data streams to be processed simultaneously. Using lasers of different colors, known as wavelength multiplexing, allows the chip to process multiple operations simultaneously. “It can have multiple wavelengths or colors of light passing through the lens at the same time,” says Hangbo Yang, an associate professor of research and co-author. “That's an important advantage of photonics.”
Future integration
This study was conducted in collaboration with the Florida Semiconductor Research Institute, UCLA, and George Washington University. Researchers noted that major chipmakers like NVIDIA are already using optical elements in some of their AI systems, which could encourage integration of this new technology.
“In the near future, chip-based optics will become an important part of all AI chips we use every day,” Sauger said. “And then there's optical AI computing.”
