An engineer team led by Florida University has developed a new AI chip based on optical computing, using lasers and micro Fresnel lenses instead of traditional electronic power computing, realizing a major innovation that has increased the energy efficiency of AI computing by 10 to 100 times. The chip focuses on key computing in depth learning - folding computing, which is the step where the machine learning model consumes the most energy in image and mode recognition.

The wafer converts the machine learning data into a multi-color laser beam, and after two micron-level Fresnel lens processing, the result is to be converted back to the digital signal. Through optical wavelength multi-path reuse technology, the wafer can process multiple data streams simultaneously, greatly improving computing speed and energy efficiency. In the identification of manual digital tasks, the accuracy rate is 98%, which is comparable to traditional chips.

The project was led by Volker J. Sorger, a professor of semiconductor photonics at the University of Florida and head of researcher Hangbo Yang, and has collaborated with the Florida Semiconductor Institute, the University of California, Los Angeles and George Washington University. The results have been published in the journal Advanced Photonics and have received funding from the US Navy Research Office.

NVIDIA, the industry giant, has also begun to incorporate optical components into AI systems. This breakthrough is expected to accelerate the commercialization of photon computing in AI hardware. Professor Sorger pointed out that optical-based chips will become an important part of future daily AI chip design, opening up a new era of AI hardware.

In addition, Florida University received US$560 million in financial year 2025, with a focus on investment including AI and semiconductor technology, providing a financial backing for this type of innovative products.