Light is fast. Blindingly so. That speed is why fiber optics carry the internet. But for a computer trying to process information, that same speed is a headache. You cannot hold onto light. You cannot pause it. You cannot make it wait its turn. Until now.
Researchers from Seoul National University and the University of South Korea have built a chip that fixes this. It slows light on command.
This isn’t just a theoretical trick. It is a programmable photonic chip designed for the heavy lifting of future AI servers. If optical computing is to replace power-hungry electronics, it needs memory. It needs buffers. It needs the ability to delay a signal without losing it.
The team, led by Professors Namkyoo Park, Sunkyu Yu, and Xianji Piao, published their findings in Advanced Science. They argue that the current bottleneck in optical computing is not speed. It is control.
Why traditional optical chips lack flexibility
To understand the breakthrough, you have to look at the hardware. Most photonic integrated circuits (PICs) use structures based on coupled-resonator-induced transparency, or CRIT.
Here is the mechanism. CRIT uses interference between multiple optical resonators. It lets specific frequencies of light pass while slowing them down. This creates the “slow light” effect needed for buffering.
But there is a catch.
Conventional CRIT devices are static. Once they are manufactured, their operating characteristics are locked in. The frequency range? Fixed. The delay duration? Fixed. If you need a longer delay, you need a different chip. If you need to change the frequency, you need new hardware.
For data centers running generative AI, this rigidity is expensive. It makes systems complex. It increases development time. It forces engineers to build separate components for signal synchronization, optical buffering, and frequency conversion.
Fixed optical hardware is a significant barrier for AI servers processing vast quantities of information in real time.
Two couplers create programmable light
The Seoul-based researchers changed the physics.
They stopped treating the bright and dark modes of the CRIT system as separate issues. Instead, they treated them as one unified degree of freedom. Then, they added two adjustable loop couplers.
This small structural change did a lot of work.
By treating the interference between modes as a single design parameter, the team made the resonator structures reconfigurable after fabrication. The chip is no longer static. It is programmable.
Theoretical analysis showed that these two loop couplers control everything. The width of the passband? Adjustable. The shape of the transmission? Reconfigurable. The delay? Dynamic.
You can change the speed of optical pulses while the circuit is running. You can convert light between frequencies without adding extra specialized components. The signal delay adjusts without sacrificing performance.
This is what makes it a programmable photonic chip. It behaves more like software than traditional hardware.
Real-world reliability in silicon nitride
Does it work in practice? Or just in a simulation?
The team tested the design using three-dimensional electromagnetic simulations. They targeted a silicon nitride ($Si_3N_4$) platform. This material is common in commercial photonics.
They didn’t just test the happy path. They simulated real-world manufacturing nightmares. Material loss? Included. Differences in resonator quality? Included. Backscattering and thermal crosstalk? Included. Phase errors in the couplers? Included.
The results were steady. The design operated reliably under realistic conditions.
This matters because it means the technology can move from paper to production. A single circuit architecture can now combine signal synchronization, variable delay lines, and optical buffers. All on one chip.
One chip replaces many components
If this technology scales, it changes the economics of optical computing.
Right now, signal processing requires many discrete components. Each one adds cost. Each one adds size. Each one consumes power.
A programmable optical chip consolidates these functions. It can change signal speed and switch tasks on the fly. Think of it as a software-defined system for light.
The benefits are clear:
- Lower energy consumption in data centers.
- Smaller, less expensive optical communication equipment.
- Faster processing for autonomous driving and quantum technologies.
Professor Namkyun Park noted the significance. This research proposes a new design principle. It allows the flow of light to be reconfigured as needed. The goal is large-scale programmable photonic circuits based on silicon photonics.
Co-first authors Dr. Seungkyn Park and Ph.D. student Beomjoo Chae emphasized the shift in perspective. Reinterpreting conventional photonic resonator physics can unlock new functionalities. They plan to move toward practical device implementation and experimental validation.
The reference for this study is “Fully Programmable Slow Light Based a Spinor Representation of General Coupled-Resonator-Transparency,” published in June 2026 in Advanced Science (DOI: 10.10002/advs.6378). Support came from the Ministry of Science and ICT.
The implications are broad. The method could apply beyond CRIT to many resonator-based photonic circuits. It suggests a future where light is not just transmitted, but designed. Controlled. Slowed.
We have long treated light as something that escapes. Now, we can catch it.
