The Speed of Light Is a Problem

Light moves fast enough to carry enormous amounts of data — but in computing, that speed is also a problem. Computers need to hold, delay, and synchronize signals, something that is trivial with electrons but nearly impossible with photons moving at 300,000 kilometers per second.

Now, researchers at Seoul National University and the University of Seoul have built the first programmable photonic integrated circuit that can slow light whenever needed. The breakthrough, published in Advanced Science, could fundamentally change how data centers and AI servers handle information.

How It Works

The chip uses a phenomenon called coupled-resonator-induced transparency (CRIT), where interference among multiple optical resonators selectively allows light within a specific frequency range to pass while reducing its speed. Previous CRIT devices were locked into their operating characteristics after fabrication — reconfiguring them required building an entirely new chip.

The Seoul team solved this by adding two adjustable loop couplers to the design. This lets engineers reconfigure the resonator structures after fabrication, controlling the width and shape of the passband along with the delay and transmission properties of signals in real time.

"This research is significant in that it proposes a new design principle that allows the flow of light within photonic integrated circuits to be reconfigured as needed," says Professor Namkyoo Park, co-corresponding author of the study.

Why It Matters

The timing could not be more urgent. The explosive growth of generative AI has sent computing demand soaring. Conventional electronic semiconductors face persistent constraints: heavy power use and limited data transfer speeds. Optical computing promises to process information far more rapidly with far less energy — but the inability to buffer light has been a critical missing piece.

A single programmable optical chip could perform multiple tasks now handled by separate components: signal synchronization, variable delay lines, optical buffers, and frequency conversion. For AI servers crunching massive models, that translates to lower energy bills, smaller hardware footprints, and faster processing.

Beyond Data Centers

Three-dimensional electromagnetic simulations confirmed the design could be built on a silicon nitride (Si₃N₄) photonic platform — a mature, commercially viable manufacturing technology. The researchers tested for material loss, resonator quality variations, backscattering, and thermal crosstalk, and found the design held up under realistic conditions.

If commercialized, the technology could eventually benefit autonomous driving systems, next-generation communications, and quantum technologies — any field requiring extremely fast information processing.

"We plan to expand this technology toward large-scale programmable photonic integrated circuits based on silicon photonics and photonic AI technologies," Park added. The team is now working toward practical device implementation and experimental validation.

Sources