A Breakthrough Technology Set to Revolutionize Ultra-Fast, Eco-Friendly AI
- Ritambhara K
- Jul 16
- 2 min read
Researchers at Université Laval have developed a tiny optical chip that can transmit data at an impressive speed of 1,000 gigabits per second—all while using less energy.
The chip, as thin as a hair, relies on two pairs of ring microresonators to achieve unparalleled performance.
Artificial intelligence systems like ChatGPT consume a significant amount of energy to function. To help address this issue, researchers at the Centre for Optics, Photonics and Lasers (COPL) have developed a cutting-edge optical chip capable of transferring massive amounts of data at lightning speed. Despite being as thin as a strand of hair, the chip delivers exceptional energy efficiency.
Their breakthrough, published in Nature Photonics, relies on light to transmit information. But unlike conventional systems that only use the intensity of light, this new chip also takes advantage of the light's phase—or its time shift—adding a whole new layer to the signal.
This added dimension enables the chip to perform at an entirely new level, all while staying compact. “We’ve gone from speeds of 56 gigabits per second to 1,000 gigabits per second,” said PhD student Alireza Geravand, the study’s lead author.
Transferring the equivalent of 100 million books in just 7 minutes—that's the kind of speed this new optical chip makes possible. The research team sees enormous potential for accelerating AI training. “At 1,000 gigabits per second, you could move all the training data—equivalent to over 100 million books—in under seven minutes, just enough time to grab a coffee,” explains the researcher. Even more impressive, the entire transfer would only use about 4 joules of energy—the same amount it takes to heat one milliliter of water by a single degree Celsius.
This breakthrough is made possible by ring microresonators—tiny silicon structures that manipulate light to carry information. The system uses two pairs of these rings: one set handles light intensity, while the other encodes the phase, adding depth and efficiency to the data transmission.
READ ALSO: ANYbotics Unveils Advanced Gas Leak and Presence Detection Feature for ANYmal Inspection Robot
Modern AI data centers rely on tens of thousands—or even hundreds of thousands—of processors that communicate with one another much like neurons in the human brain. Each processor is just a few millimeters in size, but when you connect that many of them together, the supporting infrastructure quickly becomes massive—and so does the energy required to keep it all running. “You end up with a system that stretches for miles,” says PhD student Alireza Geravand. The team’s new technology changes that, allowing processors to communicate with the speed and efficiency of systems just meters apart. That’s a major advantage as AI demands continue to rise.
This innovation could become part of mainstream industry in the near future. Companies like NVIDIA have already begun using microresonators, though current designs only use light intensity.
“Ten years ago, our lab first demonstrated this concept. Now, we’ve taken it a step further,” Geravand adds. “In a few years, industry may catch up, and this technology could make its way into real-world applications.”
Comments