At an IEEE Photonics Society event in Bangkok, Dr. Daniel Kuchta from NVIDIA was invited to share his perspectives on the development of Co-packaged Optics (CPO) technology. He stated that CPO has moved past the volatile early stages of development and entered a steady growth phase with gradual improvement.
IBM launched the first commercial CPO system in its P775 platform back in 2010, marking the start of the technology’s 15-year evolution. Since then, the concept of CPO has gradually entered public view. In 2018, Rockley introduced Topanga, the first silicon photonics-based CPO system, which further boosted interest in the technology. A key milestone came in 2021 when the Optical Internetworking Forum (OIF) released a draft standard for CPO, lifting the technology to its first historical peak.
CPO aims to address the growing bandwidth challenges in High-Performance Computing (HPC) systems. Theoretically, CPO enables higher-radix switching (supporting flatter network architectures), increased node bandwidth, frees up electrical package pins for additional DRAM channels, and allows network decoupling. However, implementing CPO has proven more difficult than anticipated. The biggest hurdle lies in building a viable ecosystem—related ASICs are prohibitively expensive, there is no one-size-fits-all solution, and pay-as-you-grow scalability remains unavailable. More importantly, CPO still faces numerous practical challenges, including manufacturing processes, product reliability, packaging and testing, maintenance, and standardization. Meanwhile, CPO’s competitor—pluggable solutions—continues to advance, with 100G per lane now achievable, and Linear Optics (LPO) also offering power reduction capabilities.
Nevertheless, Dr. Kuchta concluded that CPO technology has entered a stable development phase since NVIDIA’s major announcement at this year’s GTC event, with ELSFP emerging as the most significant achievement of this round. The future development of CPO will require solving interoperability issues, with a target power consumption of 1pJ/bit. After the event yesterday, I met Dr. Kuchta downstairs and told him I was eager to see if his prediction of CPO entering a growth phase in the next few years would materialize. Throughout his presentation, he first discussed the development trends of AI data centers and HPC, all to illustrate the demand for CPO. He specifically noted that the 2022-2023 macroeconomic crisis in the United States, which led to layoffs in many technology roles, actually presented an opportunity for CPO development (likely due to the retained talent pool in the field). I agree with two key points from Dr. Kuchta: first, never underestimate the technological advancement potential of copper cables; second, the coexistence of multiple solutions is the way forward.
Additionally, Dr. Prong from Lumentum Thailand pointed out at the conference that photonics technology is helping Thailand catch up with the new S-curve of economic growth. Thailand’s top priorities are building a photonics industry ecosystem and nurturing talent—and efforts are already underway. Indeed, the ability to attend a conference on silicon photonics and CPO in Thailand is itself a testament to the progress of this ecosystem.