A team of engineers led by the University of Florida has recently developed a new AI chip based on optical computing. Using lasers and micro-Fresnel lenses instead of traditional electronic power, this significant innovation improves AI computing energy efficiency by 10 to 100 times. This latest breakthrough offers a new solution to energy-intensive AI computing.
The chip focuses on a key computational step in deep learning called “convolution,” which is at the heart of how AI interprets photos, videos, and even written language. This step is also the most energy-intensive step in machine learning models used for image and pattern recognition.
The new optical AI chip design addresses this problem by integrating the lasers and micro-lenses directly onto the circuit board, enabling the chip to perform these calculations with significantly less energy and at a higher speed. In early trials, the chip achieved approximately 98% accuracy in recognizing handwritten digits, comparable to the performance of traditional electronic chips.
A Leap in AI Efficiency
“Performing critical machine learning calculations at near-zero energy is a leap forward for future AI systems,” said study leader Dr. Volker J. Sorger, the Rhines Professor of Semiconductor Photonics at the University of Florida. “This is critical for continuing to expand AI capabilities in the coming years.”
“This is the first time anyone has put this type of optical computing on a chip and applied it to AI neural networks,” said Dr. Hangbo Yang, a research associate professor in the Sorgers group at the University of Florida and co-author of the study.
The project, led by Volker J. Sorger, the Rhines Professor of Semiconductor Photonics at the University of Florida, and researcher Hangbo Yang, collaborated with the Florida Institute for Semiconductors, the University of California, Los Angeles, and George Washington University. The results were published in the journal Advanced Photonics on September 8 and were funded by the U.S. Office of Naval Research. This interdisciplinary engineering achievement demonstrates the tremendous potential for innovation at the fusion of optics and electronics.
The prototype chip uses two sets of miniature Fresnel lenses fabricated 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 images or other pattern recognition tasks, is converted to laser light on-chip and passed through a lens. The result is then converted back into a digital signal to complete the AI task.
Advantages of Optical Computing
This lens-based convolution system is not only more computationally efficient but also reduces computation time. Using light instead of electricity has other benefits. Sorger’s team designed a chip that can process multiple data streams in parallel using different colored lasers.
“We can pass multiple wavelengths, or colors, of light through the lens simultaneously,” said Hangbo Yang. “This is a key advantage of photonics.”
Leading chipmakers like Nvidia, following the latest trends, are already integrating optical components 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 be a key part of every AI chip we use every day,” said Sorger, who is also the deputy director of strategic initiatives at the Florida Semiconductor Institute. “Optical AI computing will be next.”
Notably, the University of Florida raised $560 million in fiscal year 2025, focusing on investments in AI and semiconductor technologies, providing strong financial support for such innovative products. The latest report shows that industry giants such as Nvidia have also begun to integrate optical components into AI systems. This breakthrough is expected to accelerate the commercialization of photonic computing in AI hardware and open up new paths for the next generation of electronic systems and engineering applications.
