Researchers from Keio University in Japan have developed an advanced plastic optical fiber (POF) technology that promises to revolutionize short-range communications in data centers dedicated to artificial intelligence (AI).
This innovative approach, led by Professor Yasuhiro Koike and speaker Kenta Muramoto, enables data transmission at speeds of up to 106.25 Gbps per core, a vital capacity to address existing bottlenecks in AI infrastructure.
The new extrusion technique, which facilitates the manufacturing of multicore fibers in a single step, dramatically reduces production costs and complexity by 10 to 100 times compared to traditional glass fibers.
100 times cheaper than fiberglass
This advancement not only optimizes manufacturing but also offers greater adaptability to various data center designs. The tests conducted confirmed that, after 30 meters of transmission at 106.25 Gbps PAM4, a high signal integrity was maintained, with minimal degradation and bit error rates (BER) reduced to 1/100,000 compared to glass fibers.
The non-uniform internal structure of the multicore POF plays a crucial role in reducing light coherence, acting as a mechanism for volumetric noise reduction throughout the fiber. Researchers stated that this type of fiber has properties that allow for effective diffusion of optical coherence, thereby improving the quality of transmission.
The multicore optical fiber system, which includes circular versions with 61 cores and rectangular versions with 4 coress, has demonstrated high manufacturing reliability and stable transmission across all cores.
With the growing demand for efficient and high-performance interconnections, Keio’s multicore plastic fiber technology could be key to unlocking a faster and more scalable AI infrastructure in the coming years. This advancement will be presented at the Optical Fiber Communication (OFC) 2025 conference, highlighting its importance in the telecommunications world.