Artificial intelligence is becoming the backbone of modern technology, from facial recognition to translation apps. However, powering AI models requires a large amount of electricity, raising questions about efficiency and sustainability.
Researchers at the University of Florida believe they have found a way to tackle the issue. Their new chips use light as well as electricity to perform one of the most demanding tasks in AI.
Light-based computing breakthrough
This chip is built to handle convolutional operations, a core function of machine learning. These operations allow AI to detect images, videos and text patterns. It also consumes a large amount of computing power.
The team integrated the optical components directly into the silicon chip. Laser lights and microscope lenses provide faster, lower energy demand convolution.
“The University of Florida has a great opportunity to develop a new line of expertise in the world,” said Volker J. Sauger, Lines Award Professor of Semiconductor Photonics at the University of Florida. “This is important to continue to expand AI capabilities over the next few years.”
Tests showed that handwritten numbers classified as prototypes match traditional chips with approximately 98% accuracy.
The system relies on two sets of Fresnel lenses, a flat ultra-thin construction similar to a lighthouse. Each lens is narrower than human hair and is etched into the chip using standard semiconductor technology.
To perform convolution, the data is converted to laser light on the chip.
That light passes through a Fresnel lens that performs a mathematical transformation. The result returns to the digital signal of the AI model.
“This is the first time that this type of optical calculation has been placed on a chip and applied to AI neural networks,” says Hangbo Yang, an associate professor of research for Sauger's group and a co-author of the study.
Multiplexing for parallel processing
A chip can also process multiple data streams at once. The team achieved this using lasers of various colors. This approach is known as wavelength multiplexing.
“Light of multiple wavelengths or colors can pass through the lens at the same time, and have colors,” Yang said. “That's an important advantage of photonics.”
The project involved the Florida Semiconductor Research Institute, UCLA and George Washington University.
Sorger noted that major players such as Nvidia are already using optical components in their AI systems. This makes the new chip easier for commercial use.
“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.”
The Florida team's chips help expand AI to meet global demand by reducing energy use while maintaining high accuracy.
As technology moves beyond the lab, light-based chips could soon power many of the AI tools people rely on every day.
This study is published in the journal Advanced Photonics.
