New light-based chip boosts power efficiency of AI tasks 100 fold

A team of engineers has developed a new kind of computer chip that uses light instead of electricity to perform one of the most power-intensive parts of artificial intelligence — image recognition and similar pattern-finding tasks.

Using light dramatically cuts the power needed to perform these tasks, with efficiency 10 or even 100 times that of current chips performing the same calculations. Using this approach could help rein in the enormous demand for electricity that is straining power grids and enable higher performance AI models and systems.

This machine learning task, called “convolution,” is at the heart of how AI systems process pictures, videos and even language. Convolution operations currently require large amounts of computing resources and time. These new chips, though, use lasers and microscopic lenses fabricated onto circuit boards to perform convolutions with far less power and at faster speeds.

In tests, the new chip successfully classified handwritten digits with about 98% accuracy, on par with traditional chips

“Performing a key machine learning computation at near zero energy is a leap forward for future AI systems,” said study leader Volker J. Sorger, Ph.D., the Rhines Endowed Professor in Semiconductor Photonics at the University of Florida. “This is critical to keep scaling up AI capabilities in years to come.”

“This is the first time anyone has put this type of optical computation on a chip and applied it to an AI neural network,” said Hangbo Yang, Ph.D., a research associate professor in Sorger’s group at UF and co-author of the study.

Sorger’s team collaborated with researchers at UF’s Florida Semiconductor Institute, the University of California, Los Angeles and George Washington University on study. The team published their findings Sept. 8 in the journal Advanced Photonics.

The prototype chip uses two sets of miniature Fresnel lenses using standard manufacturing processes. These two-dimensional versions of the same lenses found in lighthouses are just a fraction of the width of a human hair. Machine learning data, such as from an image or other pattern-recognition tasks, are converted into laser light on-chip and passed through the lenses. The results are then converted back into a digital signal to complete the AI task.

This lens-based convolution system is not only more computationally efficient, but it also reduces the computing time. Using light instead of electricity has other benefits, too. Sorger’s group designed a chip that could use different colored lasers to process multiple data streams in parallel.

“We can have multiple wavelengths, or colors, of light passing through the lens at the same time,” Yang said. “That’s a key advantage of photonics.”

Chip manufacturers, such as industry leader NVIDIA, already incorporate optical elements into other parts of their AI systems, which could make the addition of convolution lenses more seamless.

“In the near future, chip-based optics will become a key part of every AI chip we use daily,” said Sorger, who is also deputy director for strategic initiatives at the Florida Semiconductor Institute. “And optical AI computing is next.”