In a significant leap forward for artificial intelligence processing, European researchers have demonstrated a new approach to computing that uses light instead of electricity to perform complex calculations at unprecedented speeds.
The breakthrough comes from a collaboration between teams at Tampere University in Finland and Université Marie et Louis Pasteur in France, who successfully used femtosecond laser pulses (a billion times shorter than a camera flash) directed through ultra-thin glass fibers to perform AI-like computations. What makes this achievement remarkable is both the speed and efficiency of the process - completing calculations in under one picosecond while achieving over 91% accuracy on the MNIST handwritten digit recognition benchmark, a standard test for AI systems.
"This work demonstrates how fundamental research in nonlinear fiber optics can drive new approaches to computation," explained the research leaders, Professors Goëry Genty, John Dudley, and Daniel Brunner. "By merging physics and machine learning, we are opening new paths toward ultrafast and energy-efficient AI hardware."
The system works by sending laser pulses containing multiple wavelengths through optical fibers with a cross-section smaller than a human hair. The researchers encode information by introducing relative delays between these wavelengths according to image data. As light travels through the fiber, the nonlinear interaction between light and glass transforms the spectrum in ways that preserve and process the encoded information.
Interestingly, the team discovered that optimal performance didn't come from maximizing nonlinear interactions but rather from finding a precise balance in the system's complexity. This insight could prove crucial for future development of photonic computing systems.
The researchers are now working toward building on-chip optical systems that can operate in real-time outside laboratory settings. If successful, this technology could revolutionize AI processing by dramatically reducing energy consumption while increasing processing speeds by orders of magnitude compared to current electronic systems.
The research was published in Optics Letters, with the paper titled "Limits of nonlinear and dispersive fiber propagation for an optical fiber-based extreme learning machine."