Chip destruction speed, energy barrier to AI communication

Scientists at Laval University have created an optical chip that is no thicker than human hair, which can transmit data at 1,000 gigabits per second, fast enough to deliver 100 million books in seven minutes while consuming a milliliter of water’s warm energy.

The chip solves a key bottleneck in AI systems where the eager data center requires a lot of information to flow between processors. Current AI systems (such as ChatGpt) require huge computing resources, but the connection between processors is often a limiting factor.

Beyond Light Intensity: The Stage of Utilizing Speed

Traditional optical communications mainly rely on different light intensities to encode information. The innovation of the Laval team also adds the second dimension by manipulating the stage of light, which is actually its timing moves. This dual approach allows the system to package more data into each optical pulse.

“We increased from 56 gigabits per second to 1,000 gigabits per second,” said the lead author of the study. The study, published in Natural Photonics, shows how combining intensity and phase modulation produces unprecedented levels of performance in ultra-compact devices.

The chip uses a micropore modulator – a micro ring silicon device that can manipulate light to encode information. Think of them as microscopic traffic controllers of photons, guiding light signals with significant accuracy while occupying minimal space.

Key Performance Indicators

• speed: 1,000 Gigabit transmission rate per second
• Energy efficiency: Only 4 Joule to transmit data worth 100 million books
• size: The mobile component is only 100 micrometers wide
• distance: Successfully spread over 80 kilometers

The impact of the real world on AI infrastructure

Modern AI data centers depend on thousands of processors, communicating like neurons in the brain. Each processor measures only a few millimeters, but the support infrastructure extends to kilometers. This brings both physical and energy challenges.

This new technology essentially allows these processors to communicate as if they are only a few meters apart, greatly reducing physical footprint and energy requirements. For AI training, this could mean the difference between waiting hours and the complex model processing information.

Companies like Nvidia have included basic versions of microporous modulators, although limited to intensity modulation. The advancement of the Laval team in adding phase modulation represents a significant leap in technical capabilities.

Ten years of development

This breakthrough was a foundation established in the same laboratory ten years ago. Geravand noted that despite their labs setting up fundamental work a decade ago, today’s achievements represent taking technology “to the next level.”

The study involves a detailed study on how these micro-zone devices behave when transmitting complex optical signals. The team found that by embedding a pair of microscopes into a professional configuration, they could eliminate the undesirable frequency distortion of previously limited performance.

The chip achieves what researchers call the “documentary coastline bandwidth density”, which is more than 5 mm per millimeter per second, a measure of how much data can flow through a given chip area. This efficiency measure proves crucial to future AI systems where space and energy maintain a premium.

As demand for AI computing continues to grow, innovations like this kind of optical chip may be crucial to maintaining technological advancements without overwhelming the global energy grid. Researchers expect their technology to reach commercial markets in the next few years.

Related