TDK has found a way to speed up data transfer between chips by an order of magnitude – optics will eliminate the bottleneck of AI systems

Japanese company TDK has demonstrated a “breakthrough” optical data transmission technology with a response time of 20 picoseconds. The “spin photodetector,” which combines optical, electronic, and magnetic elements, will replace existing semiconductor-based data transmission methods. According to TDK, the new technology will speed up data transmission by an order of magnitude, which will eliminate the main bottleneck holding back the growth of generative AI.

Image source: unsplash.com

Currently, data is exchanged between chips using electrical signals, but the increased volumes of information processed by AI require a shift to optical technology. “Data transfer is the biggest bottleneck for AI, not the performance of the GPU chip,” says Hideaki Fukuzawa, senior manager of TDK’s Next-Generation Product Development Center. “Because we can overcome many of the current bottlenecks, we believe this technology will be a game changer for the AI ​​and data center industries.”

Testing of the optical data transmission technology developed by TDK has confirmed its effectiveness. According to Professor Arata Tsukamoto of the University of Tokyo, “the spin photodetector has remarkable prospects from both a scientific and technological point of view.” After additional testing, TDK plans to provide working samples of the technology to its customers by the end of March 2026 and begin mass production within the next three to five years.

Despite the immaturity of the technology and the need to create a full-fledged ecosystem together with chip developers, TDK is confident in the advantages of the proposed solution. The company emphasized the relative cheapness and high energy efficiency of its spin photodetector, as well as a wide range of applications, including in augmented and virtual reality headsets and high-speed image sensors.

The new device is part of a market for photonic integrated circuits that technology research group IDTechEx predicts will grow more than tenfold over the next decade to $54.5 billion due to the needs of generative AI.