I am writing this blog article in continuation of my previous blog – “Future-proof FPGA platforms the longevity and upgradability”. In this blog entry, I discuss the available pluggable front port technology, complementing the flexible properties of the described future-proof FPGA-based SmartNICs.
OFC 2016 – What’s hot and what’s not in the current optical business?
The annual highlight in the optical business is the OFC (Optical Fiber Communication) Conference, this year held in Los Angeles, California with 580 exhibiting companies and more than 13000 attendees, for more details go to ofcconference.org
Whereas the European counterpart, ECOC, seems to serve a narrower audience, with an optical module almost as the highest integration level, OFC manages to bring the complete value chain to “the table”, ranging from the vendor of small specialized optical subcomponents to several router manufactures, test equipment vendors, and even service providers, all under the same roof… very efficient for the attendees. This year OFC had two hot topics that one way or the other impacted the majority of the exhibiting companies… Read on!
Can you pronounce “PAM4”?
PAM4 is hot… NRZ is not. One of last OFC’s hot debate topics was how to break the per channel 25Gbps “barrier” required in order to enable efficient backplane- and inter device communication, supporting next generation line rates of 400Gbps and beyond. The conference even witnessed a vote among the audience at one of the plenary sessions with 4 different implementation options on vote. This year, three of the options were already history as the industry had already adopted the PAM4 option.
All stakeholders in the sourcing value chain, ASIC-, FPGA-, Optical-, Connector- and Test Equipment manufacturers made a great effort of showcasing that they are ready for the new PAM4 area, are mastering the technology and are ready to support it. The PAM4 technology has certainly made a massive stage entry; now it is time to prove its value and provide return on investment. Personally, I will be revisiting the technology having worked with the PAM4 technology at startup Enigma Semiconductor back in the mid-2000s. Here, we were using the Rambus RaSerX transceiver hard IP in our backplane switch chip set, which included PAM4 technology. The Enigma Semiconductor use-case did not succeed, but for a lot of other reasons, which is why I am certainly ready to give the technology another chance.
In my mind, PAM is an interim solution that fills a gap until something more scalable comes along. PAM4 will not scale to PAM8 or even PAM16 for general use cases. For short haul deployment, such as USR/XSR/VSR, PAM8 might have a future, but it is very unlikely that it will be suitable for MR/LR applications such as backplane deployment. With the current high level investments in PAM4 technology in ASICS, FPGA’s and test equipment, I am sure that the industry and standardization bodies will make the most of it, deploying PAM4 wherever it makes sense. One obvious application is already maturing, as 200G in the existing 4 channel form factors, CFP4/QSFP, are about to enter the market.
Optical module form factors… how many do we need?
With the presence of interest groups, such as the “Ethernet Alliance” and the “Optical Internetworking Forum (OIF)”, it is fruit for thought to observe that as many as three new optical form factors have made it past the “proof of concept” phase in the last year, competing with each other with a significant amount of overlap for coming 200G/400G appliances. Members of both “Ethernet Alliance” and OIF span several parts of the value chain, from the component vendor all the way to the service provider. One would assume that this blend of members would impact the industry with a more united industry investment in new technologies, discarding new initiatives without broad industry consensus. Nevertheless, we have never seen so many different form factor offerings competing for the favor of the applications as we do now. The microQSFP is a 4 channel design in a SFP+ like form factor, targeted for density optimized 40G/100G/200G switch and line card applications (200G through PAM4 operation). The QSFP-DD and the CFP8 form factor are both 8 channel designs, targeting 400G deployment and beyond. Whereas the CFP8 form factor is completely new with no backwards compatibility, the QSFP-DD is backward compatible supporting both QSFP, QSFP+ and QSFP28 deployment through a clever two row PCB connector on the optical module, see the picture below.
The table below summarizes the current stakeholders, categorized after industry, in the mentioned competing new form factor initiatives. The microQSFP and the QSFP-DD form factor are defined, and promoted through industry consortiums, whereas the CFP8 is defined in a MSA. Without the connector- and optical vendors to implement one or more of the new standards, they are of course left without a future. I am sure that the current situation is far from ideal for these two vendor categories. With an expectation of decreasing margins when volume has ramped, it must be tough to justify participation in all three initiatives at the same time, as is the case for several of the optical vendors in the table below.
In my point of view, it might actually be that all three form factors will receive market acceptance over time. For example, the microQSFP is suitable when a requirement for high port density is key, while the QSFP-DD is suitable when a requirement for backwards compatibility is key. The CFP8 is more appropriate in backbone applications, which require best in class bit-error-rate performance and long hauls, with high power dissipation in the optical modules. If I had to make a bet on one of the three standards, then I would bet my buck on the QSFP-DD.