NYCU × Semiconductor Research Center Reshaping Data Transmission Architecture for the AI Era

As demand for AI computing power grows at an almost exponential rate, the semiconductor industry faces a formidable physical bottleneck. While computing engines (GPUs/XPUs) accelerate like high-performance supercars, the data transmission pathways connecting them remain as congested as narrow one-way streets.

To break through this bandwidth wall, tech giants globally have placed high expectations on silicon photonics and co-packaged optics (CPO). In a landmark collaboration, Distinguished Professor Chi-Wai Chow of the Department of Photonics at National Yang Ming Chiao Tung University (NYCU) and the Semiconductor Research Center have successfully established an optoelectronic measurement platform supporting single-channel speeds up to 112 GBaud (224 Gb/s PAM4). This platform lays the groundwork for Taiwan’s supply chain to evolve from a follower to a leader in heterogeneous integration and talent cultivation.

“The computing power of AI servers is growing exponentially. If transmission speeds cannot keep up, chip performance cannot be fully realized,” explained Professor Chow. A speed of 112 GBaud or 224 Gb/s means a single optical channel can transmit 224 billion bits of data per second, roughly equivalent to transferring the contents of six single-layer DVDs in just one second. Current mainstream 1.6T optical modules are built from eight channels, each operating at 200 Gb/s.

However, achieving such extreme speeds presents two major challenges for Taiwan. First, the industry has long depended on overseas validation resources; only by establishing a domestic measurement platform can it independently develop and verify next-generation components. Second, at ultra-high frequencies, signal attenuation and noise interference remain the developer’s worst nightmare.

Silicon photonics measurement is a high-barrier integration effort. Professor Chow acknowledged that building a 112 GBaud platform requires high-frequency probes, high-speed signal generators, oscilloscopes, and mechanically stable optical tables. In this process, the Semiconductor Research Center acted as a crucial industry-grade technology catalyst, supplying international-standard high-speed measurement equipment. By combining advanced digital signal processing (DSP) with upgraded hardware, the team successfully overcame challenges such as uneven channel frequency response and system noise, ensuring the reliability of measurement data.

Strategic Deployment of Heterogeneous Integration and Virtual Fabs

In addition, the Center and NYCU partnered with the Eindhoven University of Technology (TU/e) in the Netherlands to develop indium phosphide (InP) chips, driving transmission speeds to 320 Gb/s. Because silicon cannot efficiently emit light, introducing InP, a direct bandgap material, through heterogeneous integration is seen as the key to meeting the demands for high-efficiency light source and low-power operation in AI data centers. NYCU has also linked with international foundries to create a virtual fab environment, positioning itself as a hub for innovation.

“We have accumulated 20 years of tape-out experience worldwide, allowing us to understand the design tolerances of different foundries,” Professor Chow explained. This expertise allows NYCU to serve as a bridge, translating Hon Hai’s industrial needs into specifications that a foundry can actually fabricate and refining design models through measurement feedback.

The most lasting impact of the collaboration is on talent development. Students participating in this project gained direct exposure to industry standards and cutting-edge design platforms. Professor Chow emphasizes that Taiwan’s semiconductor success stems from a generation of engineers fluent in processes, devices, and systems. Through the collaboration with the Hon Hai Research Institute, the students developed the ability to solve real-world engineering problems even before entering the workforce. This enduring advantage secures Taiwan’s long-term global leadership.

We are witnessing the third wave of the AI revolution. The role of AI and optical communication is shifting from long-haul cables that connect cities to short-reach interconnects within data centers and ultimately to direct chip-to-chip links. The academic-industrial collaboration between NYCU and the Semiconductor Research Center represents Taiwan’s decisive advantage at this pivotal turning point.