Ultra-thin optical chip moves data energy-efficiently at record speeds

The race to power artificial intelligence is heating up, and scientists are turning to light for faster data transfer. At the Centre for Optics, Photonics and Lasers, researchers created an ultra-fast, energy-saving optical chip. This chip is no wider than a human hair and moves data at record speeds. It could transform AI systems like ChatGPT, which require massive amounts of information.

The new chip doesn’t just rely on light's intensity—the traditional method for optical data transfer—it also uses its phase. Phase refers to the shift in a light wave, similar to how two ocean waves might crest at slightly different times. By adding phase to the equation, the researchers unlocked a whole new dimension of control. This enabled the chip to carry more data, more efficiently, and across longer distances without losing signal quality.

Alireza Geravand, a PhD student at Laval University and lead author of the study, describes the jump in performance as massive. “We’re jumping from 56 gigabits per second to 1,000 gigabits per second,” he said. That’s nearly a 20-fold increase.

To put that into perspective, this chip could transfer the data equivalent of 100 million books in less than seven minutes. And it would only take four joules of energy to do so—enough to heat just a single millilitre of water by one degree Celsius. It’s a striking comparison that shows just how far data transfer technology has come.

At the heart of the breakthrough is a silicon-based structure known as a microring modulator. These are tiny ring-shaped circuits that can manipulate light to carry information. Their compact design makes them perfect for fitting onto chips, and their energy efficiency makes them well-suited for large-scale computing systems.

Microring modulators (MRMs) have been studied before, praised for their small size and low energy use. However, they come with limitations. Most notably, they suffer from problems like frequency chirp—where the color of the light shifts during transmission—and dynamic nonlinearity, which disrupts signal quality. Because of these challenges, MRMs have mostly been restricted to simpler, less efficient communication methods that only detect light intensity.

But this new study takes a fresh look at MRM dynamics and discovers a way to overcome these obstacles. The research team found that while the intensity and phase modulations in MRMs are distinct, they are also connected. This connection made it hard to use MRMs in advanced, high-order modulation systems—which are essential for the fastest and most efficient data transfers.

The solution? Embedding two MRMs into a Mach–Zehnder interferometer using a push–pull design. This method created a stable, bistable phase response and allowed for amplitude modulation without any frequency chirp. As a result, the signal stayed clean and strong, even at very high speeds.

Building on this innovation, the team designed an ultra-compact modulator that can handle in-phase and quadrature modulation (commonly called IQ modulation). This technique uses both amplitude and phase to encode data, which helps squeeze even more information into a smaller signal.

The chip was built using a photonic process that works with standard silicon manufacturing tools. This means it could be mass-produced easily. It reached symbol rates up to 180 gigabaud and data rates over 1 terabit per second. These speeds were achieved using just 10.4 femtojoules of energy per bit.

With this chip, data centers running advanced AI models could work much faster and with far less energy. Today’s data centers can stretch for kilometers and use thousands of processors to mimic the interconnectedness of a human brain. Each processor may be only a few millimeters wide, but the wiring and power needed to keep them connected is massive. Geravand notes, “You end up with a system that’s kilometres long.” But with this chip, those connections can behave as though they are only meters apart.

The demand for powerful computing systems is growing fast, especially as AI tools become more complex. Training these systems involves massive datasets and long processing times. The new chip from COPL could slash that time while drastically cutting energy use.

Companies like NVIDIA are already exploring the use of microring modulators in their data systems, though current designs still rely only on intensity modulation. The technology introduced by Geravand’s team pushes those limits much further by introducing phase control in a stable and efficient way.

The researchers believe it won’t take long for the industry to catch up. “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.”

As the need for faster, more efficient data transfer continues to grow, especially in areas like AI, autonomous vehicles, and high-performance computing, this new chip might soon become a critical piece of the puzzle.

Research findings are available online in the journal Nature Photonics.

Note: The article above provided above by The Brighter Side of News.

Like these kind of feel good stories? Get The Brighter Side of News' newsletter.