A disruptive technological breakthrough is injecting powerful momentum into the future development of artificial intelligence (AI).
On September 15th, a team of engineers led by the University of Florida successfully developed a new AI chip based on optical computing. This chip is expected to increase the energy efficiency of AI computing by 10 to 100 times, providing a revolutionary solution to the challenges of growing computing power and energy consumption.
Artificial intelligence technology is currently developing at an unprecedented pace, with deep learning models achieving remarkable success in fields such as image recognition and natural language processing. However, these achievements are driven by the enormous consumption of computing resources. Traditional electronic chips face severe energy efficiency bottlenecks when performing deep learning operations, especially when processing “convolution” calculations—the core step in AI’s recognition of images, videos, and language. As the scale of models continues to expand, data center energy consumption has skyrocketed, driving up operating costs and placing enormous pressure on the environment. Therefore, finding a more efficient and energy-efficient computing method has become a pressing need in the technology community.
The innovation of the University of Florida team targets this pain point. They abandoned the traditional computing model that relies on the flow of electrons and instead used lasers and micro-Fresnel lenses to perform key AI computing tasks. Fresnel lenses, a thin two-dimensional lens commonly found in lighthouses, are miniaturized and integrated directly onto the chip’s circuit board. When machine learning data (such as image information) is converted into an on-chip laser beam and passes through these micro-lenses, the physical properties of light enable complex convolution operations to be performed with extremely low energy consumption. The calculation results are then converted back into digital signals, completing the entire AI task.
The latest breakthrough of this technology lies in its true “chip-ization” of optical computing from a theoretical perspective and its application to artificial intelligence neural networks. Research leader Dr. Volker J. Sorger of the University of Florida noted, “Performing critical machine learning calculations at near-zero energy represents a quantum leap for future AI systems.” This assessment highlights the technology’s enormous potential. In early experiments, this optical AI chip achieved approximately 98% accuracy in recognizing handwritten digits, comparable to traditional electronic chips while achieving a quantum leap in energy efficiency.
In addition to its energy efficiency advantages, photonic computing also offers improvements in speed and parallel processing capabilities. Compared to electronic signals, optical signals travel much faster and do not interfere with each other. The research team cleverly exploited this advantage, designing a system that allows multiple wavelengths or colors of laser light to simultaneously pass through the same lens, enabling the parallel processing of multiple data streams.
Dr. Sorger asserted, “In the near future, chip-based optics will become a key component of every AI chip we use daily. Optical AI computing will be next.” The success of this new optical computing chip is not only a triumph in materials science and photonics, but also a profound transformation of the entire AI hardware ecosystem.
