Artificial intelligence (AI) systems use a lot of energy to process information. To solve this, researchers at Université Laval developed a tiny optical chip, as thin as a hair, that moves huge amounts of data very fast while using little energy. An optical chip uses light to send information. Their work was published in Nature Photonics. Unlike older systems that only use light’s intensity, this chip also uses light’s phase. This makes the chip much more powerful.
The chip can transfer data at 1,000 gigabits per second, compared to older systems at 56 gigabits per second. This speed means it could send the data of 100 million books in about seven minutes, using just small amounts of energy. This makes the chip very energy-efficient for training AI models.
How the chip works
The chip uses devices called microring modulators, which are small silicon rings that control light to carry information. A modulator changes light to encode data. The chip has two pairs of these rings: one pair adjusts light intensity, and the other adjusts phase. This setup allows the chip to handle more data in a smaller space. Current AI data centers use thousands of processors spread over large areas. These systems can stretch for kilometers and need a lot of energy. The new chip makes processors act as if they are just meters apart, improving speed and saving energy.
This technology could be used in the future as AI needs grow. Some companies, like NVIDIA, are starting to use similar microring modulators, but only for light intensity. The researchers started working on this idea ten years ago and have now improved it significantly. They believe it could reach the market soon, making AI systems faster and more efficient. The study was led by PhD student Alireza Geravand, with contributions from other researchers. "Ten years ago, our lab laid the groundwork for this technology. Today, we're taking it to the next level," says Geravand. "Maybe in a few years, the industry will catch up, and this innovation will make its way into the real world."