In a significant leap forward for artificial intelligence hardware, research teams from Tampere University in Finland and Université Marie et Louis Pasteur in France have successfully demonstrated how intense laser pulses traveling through ultra-thin glass fibers can perform complex AI computations at unprecedented speeds.
The collaborative study, led by Professors Goëry Genty, John Dudley, and Daniel Brunner, with key contributions from postdoctoral researchers Dr. Mathilde Hary and Dr. Andrei Ermolaev, has shown that their optical computing system can process information thousands of times faster than traditional silicon-based electronics. Most remarkably, the system achieves these speeds while maintaining accuracy comparable to conventional systems in tasks such as image recognition.
"This work demonstrates how fundamental research in nonlinear fiber optics can drive new approaches to computation," explained the research leaders. "By merging physics and machine learning, we are opening new paths toward ultrafast and energy-efficient AI hardware."
The breakthrough leverages a computing architecture known as an Extreme Learning Machine, inspired by neural networks. Instead of conventional electronics and algorithms, the system achieves computation by taking advantage of the nonlinear interaction between intense light pulses and glass. This approach addresses growing concerns about the limitations of traditional electronics, which are approaching their physical limits in terms of bandwidth, data throughput, and power consumption.
The potential applications extend far beyond academic research. As AI models continue to grow larger and more energy-hungry, this technology could help address critical bottlenecks in computing infrastructure. The researchers aim to eventually build on-chip optical systems that can operate in real-time outside laboratory settings, with applications ranging from real-time signal processing to environmental monitoring and high-speed AI inference.
This development comes at a crucial time for the computing industry, as companies like Lightmatter and LightSolver are also making significant strides in photonic computing. With Lightmatter planning to launch its M1000 platform in summer 2025 and LightSolver recently being named a 2025 Technology Pioneer by the World Economic Forum, the race to harness light for next-generation computing is accelerating rapidly.