Co-packaged Optics standards activity have moved CPO from the realm of discussion closer to real-world adoption: (1) COBO has established a CPO working group. (2) CPO specifications have materialized in the form of the CPO Collaboration Joint Development Forum’s CPO Module product requirements document (PRD). And (3) there is an implementation agreement (IA) from the Optical Internetworking Forum (OIF).
These steps — particularly OIF’s high-profile entrance into the space, and its IA — point towards commercialization of CPO production and signal that the industry is taking CPO seriously as an answer to the data growth testing the limits of network technology currently in use, and as a component in emerging AI and 5G infrastructure.
CIR analyzed the participation of industry actors to predict the timing of CPO adoption in its “Co-Packaged Optics Markets: 2021-2025” report. Among the important firms actively involved in the evolution of CPO are Microsoft, Facebook, IBM, Huawei, Juniper, and many others.
We note that some of the firms active in the CPO space seem very bullish about the prospects of standards. While we are too for the most part, we think there will be a healthy proving period for CPO before it goes mainstream. It is not just that the number of data interconnections requiring 800G and above are quite limited, but also, our research suggests that CPO will take some serious selling to data center managers, who we suspect will be reluctant to use a non-pluggable solution after decades of pluggable and the convenience it offers.
We have predicted that the market for CPO this year will be under $4 million, but we expect it to reach over $1 billion by the end of the decade as CPO prices decline and the market comes to accept non-pluggability. Pluggable transceiver modules, like QSFP-DD and OSFP, may potentially support 800G with some adaptations and even beyond, but each speed increase adds to power and signal-integrity problems. Once these issues become unmanageable, CPO will likely stand out as a more feasible option for large data centers, video centers, control centers for autonomous vehicles and so forth. This will open up market opportunities for companies offering viable CPO products.
The CPO IA: OIF announced its CPO Framework Project, a working group including hyperscalers as well as optical and switch companies, in November 2020. The group’s strategy will be to first study CPO, then assess the technology to determine what IAs are required. OIF unveiled the framework’s first project — a 3.2T CPO Module IA — at its Q121 Technical and MA&E Committees meeting late last February.
The industry has clearly articulated the need for Co-packaged Optics standards to address power constraints, new architectures, and applications and OIF is well-adapted to dealing with such issues.
COBO’s CPO working group: The Consortium of On-Board Optics (COBO) set up its own CPO working group in December 2020. COBO’s pivot toward CPO should not surprise anyone who has followed its trajectory. Founded by Microsoft and Facebook to promote the interoperability of onboard-optics — a kind of intermediate station en route to CPO — it never succeeded in building any interesting market-ready products.
COBO’s CPO working group is focusing on openness standards that allow end users multi-vendor choice in the chips and optics used. It will also work towards specifications around optical connectivity and external lasers, among other things. It will be collaborating with the Institute of Electrical and Electronics Engineers (IEEE) for the development of Ethernet specifications.
COBO will also collaborate with OIF, with whom it shares a history. COBO is focusing largely on what is outside the co-packaged assembly, while OIF is at work on what’s inside, and overall interoperability. COBO Chairman, Brad Booth, has said there are twelve possible specification items needed to create an interoperable ecosystem. Some of the specifications already exist, with many referencing the OIF CEI-112G XSR specification for the electrical interface.
As we see it, COBO’s current stance on CPO is a concession to reality. The original story for COBO was as standardized version of the “integrated optics” of yesteryear. But it now seems that COBO did not go far enough in its design innovations. Even COBO’s proponents now see COBO as a training standard. But the goal of COBO specification is to have a standardized way to approach on-board optics, an approach that has been around for so long prior to standards.
One question yet to be answered is, how will CPO fit into the established supply chains for transceivers, if it will at all? The skills needed to build CPO are different in many ways than those needed for garden variety transceivers.
With working groups like COBO and, especially, OIF now at work producing standards, we can expect CPO to become a major influence on design for big networking equipment. Engineers will now get exposure to CPO as well as experience building and deploying it, and CPO products will become a more visible choice for data centers, with adoption increasing as prices gradually go down. This shift will have major implications for a wide swath of manufacturers.