Artificial intelligence (AI) is becoming an integral part of modern technology, powering features such as facial recognition and automatic text translation. However, as AI models grow more complex, their energy consumption also increases, posing significant challenges for energy efficiency and sustainable development. Researchers at the University of Florida have proposed an innovative solution to this problem—a new chip that uses light instead of electricity to perform one of AI’s most resource-intensive tasks. Their findings have been published in the journal Advanced Photonics.

A Breakthrough in AI Energy Efficiency

The newly developed chip is designed to carry out convolution operations—a key process in machine learning that enables AI systems to recognize patterns in images, videos, and text. Typically, these computations require substantial resources. By integrating optical components directly into a silicon chip, the scientists created a system capable of performing convolutions using laser light and microscopic lenses, which significantly reduces energy consumption and speeds up data processing.

According to Professor Folkert J. Zorker, a leading expert in semiconductor photonics and head of the research, performing essential machine learning calculations with virtually no energy loss is a major step forward for future AI systems. He emphasized that such technologies are crucial for the continued advancement of artificial intelligence.

Technological Features of the Chip

In tests, the chip prototype was able to classify handwritten digits with about 98% accuracy, comparable to traditional electronic chips. The system uses two sets of miniature Fresnel lenses—flat, ultra-thin lens analogs commonly used in lighthouses, manufactured with standard semiconductor technology. These lenses, which are about the width of a human hair, are etched directly onto the chip.

To perform convolutions, machine learning data is first converted into laser light on the chip. The light then passes through the Fresnel lenses, which carry out the mathematical transformation, after which the result is converted back into a digital signal to complete the AI task.

Associate Professor Hanbo Yan, a co-author of the study, highlighted that this is the first time such optical computations have been implemented on a chip and applied to an AI neural network.

Scalability and Advantages of Photonics

The team also demonstrated that the chip can process multiple data streams simultaneously using lasers of different colors—a method known as wavelength-division multiplexing. Yan noted that the ability to transmit several wavelengths of light through a lens at once is a key advantage of photonics.

The research was conducted in collaboration with the Florida Institute for Semiconductors, the University of California, Los Angeles, and George Washington University. Zorker added that chip manufacturers like NVIDIA are already using optical elements in some of their AI systems, which could simplify the integration of new technologies.

According to the researchers, in the near future, on-chip optics will become an essential part of every AI chip used in daily life, and optical computing for artificial intelligence will be the next stage in the evolution of the industry.