Artificial intelligence has become a central part of modern technology, running everything from facial recognition on smartphones to medical imaging, online translations, and self-driving cars.
But there’s a problem: today’s AI systems use enormous amounts of electricity.
As models grow larger and more powerful, the energy they consume is straining electrical grids and raising concerns about sustainability.
Now, researchers at the University of Florida have unveiled a new computer chip that could dramatically reduce this power demand.
The chip uses light, rather than just electricity, to perform one of AI’s most energy-hungry operations: finding patterns in data such as images, video, or text.
By combining optical components with traditional electronics, the new chip can perform these tasks with 10 to 100 times greater efficiency than current chips.
The study, published in Advanced Photonics, was led by Volker J. Sorger, the Rhines Endowed Professor in Semiconductor Photonics at the University of Florida.
According to Sorger, the breakthrough represents “a leap forward for future AI systems,” making it possible to keep scaling up capabilities without overwhelming the world’s power supply.
At the heart of this advance is the way the chip performs what are called convolution operations.
These are the mathematical building blocks that allow AI systems to recognize patterns—such as distinguishing a dog from a cat in a photo, or identifying the features of handwritten numbers. Normally, these operations demand vast amounts of processing power, but the Florida team has shown that using light to perform them is far more efficient.
The chip works by integrating microscopic lenses directly onto a silicon wafer, built using the same standard methods already used to make semiconductors. These are flat Fresnel lenses, inspired by the giant glass lenses once used in lighthouses but shrunken to a scale narrower than a human hair.
When laser light passes through the lenses, they naturally perform the mathematical transformation needed for a convolution. The results are then converted back into electronic signals, completing the AI task.
In testing, the prototype chip successfully classified handwritten digits with about 98% accuracy—matching the performance of conventional electronic chips, but with a fraction of the power consumption.
Even more impressively, the team demonstrated that the chip could handle multiple streams of data at once by shining different colors of laser light through the lenses. This method, called wavelength multiplexing, takes advantage of one of light’s key strengths: the ability to carry multiple signals simultaneously without interference.
“This is the first time anyone has put this type of optical computation on a chip and applied it to an AI neural network,” explained Hangbo Yang, a research associate professor at UF and co-author of the study.
The work was carried out in collaboration with the Florida Semiconductor Institute, UCLA, and George Washington University. Sorger noted that big chipmakers such as NVIDIA already use some optical components in their products, which could help speed up the adoption of this new technology.
“In the near future, chip-based optics will become a key part of every AI chip we use daily,” Sorger predicted. “Optical AI computing is next.”
If this vision becomes reality, the next generation of AI may not only be smarter and faster but also far greener—powered by light instead of electricity alone.
