Artificial intelligence systems like ChatGPT require huge amounts of computing power—and with that, massive amounts of energy.
But researchers at the Center for Optics, Photonics and Lasers (COPL) in Canada may have just found a way to make these systems much faster and far more energy efficient.
The team has developed a groundbreaking optical chip that uses light not only to send data faster than ever before, but also while using very little energy.
The chip is as thin as a strand of hair, but it can transfer information at an incredible speed of 1,000 gigabits per second—about 20 times faster than current technology.
For comparison, that’s like transferring the contents of 100 million books in less than seven minutes.
What makes this chip different is how it uses light. Most current systems only use the brightness, or intensity, of light to send information.
But the COPL team has figured out how to use the phase of light as well—essentially tracking how the light wave shifts as it moves. This added dimension allows the chip to carry far more information at once.
At the heart of this technology are microring modulators, tiny loops made of silicon that control how light is used to carry data.
The system uses two pairs of these microrings—one pair to control light intensity and the other to adjust the phase. This dual approach unlocks much higher speeds while keeping the chip extremely small and efficient.
According to Alireza Geravand, a Ph.D. student and the lead author of the study published in Nature Photonics, this leap in speed doesn’t come at the cost of energy.
In fact, sending that enormous amount of data would use only about four joules of energy—barely enough to warm a single milliliter of water by one degree Celsius.
This innovation could reshape how data centers—especially those that train AI models—are designed.
Currently, these centers use thousands of processors that must communicate over long distances, which leads to high energy use and space requirements.
With this optical chip, those same processors could act as if they were right next to each other, making the entire system faster and much more compact.
Geravand says that while this work builds on research started 10 years ago, it’s now reaching a stage where it could be used in real-world applications. Companies like NVIDIA have already begun using parts of this technology, though only for simpler light-based systems.
If industry catches up, this ultra-fast, energy-saving chip could soon become a key part of the future of AI.
Source: KSR.
