Solving Bottlenecks with Light: SK hynix Unveils Blueprint for Next-Generation AI Infrastructure 'CPO'

Paper on CPO Technology Roadmap Published with International Researchers

Offering Solutions to the Limitations of Traditional Copper Wiring

SK hynix has unveiled a technology roadmap for CPO (Co-Packaged Optics), a next-generation light-based connectivity technology designed to address new bottlenecks in the artificial intelligence (AI) era, looking beyond high-bandwidth memory (HBM).

Seunghoon Hong, Team Leader at SK hynix (left), and Gyusang Lee, Professor at the University of Virginia, are discussing research on next-generation optical interconnect (CPO) technology. SK hynix Newsroom

On August 20, SK hynix announced via its newsroom that it has published a paper outlining the future direction of CPO, a next-generation photonic interconnect technology for AI and high-performance computing (HPC), in the global journal "Nature Electronics," together with a team of international researchers.

The research was co-led by Seunghoon Hong, Team Leader for AI Infrastructure at SK hynix, and Gyusang Lee, Professor at the University of Virginia (UVA), as corresponding authors. The collaborative study included researchers from the University of Illinois Urbana-Champaign (UIUC), Nanyang Technological University (NTU), Massachusetts Institute of Technology (MIT), and Yonsei University.

In ultra-large AI cluster environments, computational performance is experiencing explosive growth, tripling every two years, while bandwidth increases only by a factor of 1.4, creating a limiting factor known as the "Bandwidth Wall." Especially when data moves between racks, traditional copper wiring suffers significant signal attenuation and power consumption over longer distances, causing additional problems.

To address this, the research team proposed CPO technology, which integrates optical transceivers (TRx) within the same package as the processor. By minimizing the propagation of high-speed electrical signals inside the package and connecting data through light, CPO simultaneously achieves high bandwidth density, enhanced energy efficiency, and signal integrity.

The paper forecasts that CPO technology will evolve in the long term toward an "Optics-Centric" architecture, in which optical connectivity is extended to the "memory interface" level. By directly linking memory and processors through optical interposers, multiple AI accelerators can share a massive memory pool and respond nimbly to the scaling up of AI models.

Gyusang Lee, Professor at the University of Virginia, emphasized, "The key is to co-design memory devices, controllers, optical components, and packaging as a unified system," highlighting the significance of collaboration with SK hynix.

Seunghoon Hong, Team Leader at SK hynix, stated, "As AI systems become larger and more complex, the importance of photonic interconnects will only increase. As a memory company, we aim to go beyond supplying a single product by continuously expanding our role as a next-generation partner that strengthens the overall system competitiveness of our customers."

This content was produced with the assistance of AI translation services.