Andrew Schmitt presented at the ECOC (European Conference for Optical Communications) in Dusseldorf, Germany during the week of Sept. 19. No time was wasted during the trip as Andrew fit in well over 30 scheduled meetings with component and equipment suppliers and end users of optical hardware.
The meetings were very productive and we gained new insight in several areas:
- Huawei’s plans to transition to 100G CFP-DCO technology and what will follow
- Updates on vendor selection and R&D in coherent DSPs
- Details on the ongoing ramp in CFP2-ACO and the development of this new market
- Overview of 400G market development
- More insight into Nokia’s China Mobile optical design win
Huawei CFP Transition
Huawei’s WDM equipment design is evolving to utilize pluggable CFP based 100G technology, and the use of non-pluggable hardware is being de-emphasized. Huawei’s transition will ripple through the supply chain and warrants detailed examination. The notes that follow are a result of direct discussions between Cignal AI and Huawei, as well as other suppliers.
Huawei will use the CFP-DCO format for 100G coherent interfaces, and has imminent plans to introduce this format into its WDM long haul and metro equipment. Huawei will build its own CFP modules in house to support the transition, including it’s internally designed DSP. According to Huawei, Acacia will not be used as a CFP module or DSP supplier.
Other vendors we spoke with indicated that both Neophotonics and Oclaro have a role in supplying components used in the design of Huawei’s CFP-DCO module. It’s possible that Huawei may be using discrete silicon photonics components as part of this CFP design, but the company is more likely planning to use this technology in a more compact future design – such as an in-house CFP2-DCO. Huawei engineers view silicon photonics as a key enabler of the higher density opto-electronics required to make CFP2-DCO modules a reality.
The majority of 100G ports shipped by Huawei in 2017 are still expected to be built with discrete electronics, not pluggable CFPs. That said, the company expects that all of the 2017 port shipment growth will come from CFP-DCO while production with discrete components will remain flat during this period. The conclusion is that suppliers wishing to experience Huawei shipment growth must be designed into this CFP.
Future CFP2-DCO designs are of great interest to Huawei, though the company does not anticipate that the technology will experience volume production until 2018. Samples are expected to be available from a vendor (most likely Acacia) within a few months. Huawei intends to evaluate CFP2-DCO against equivalent CFP2-ACO technology next year in order to determine which provides the greater advantage.
We were disappointed not to learn more about ZTE’s plans for CFP-DCO in 2017. Unlike Huawei, ZTE currently lacks the technical abilities to build its own CFPs. If ZTE must source CFPs externally, presumed supplier Acacia would likely benefit, since its revenue is higher per port for CFP modules vs. DSP only sales.
Coherent DSP Updates
Huawei is continuing to work with NTT Electronics (NEL) to supply DSPs, though they are less reliant on NEL’s support than in the past. According to one source at Huawei, all of the company’s 100G coherent production will now use Huawei’s own internal DSP. Note that sources outside of Huawei dispute this assertion and Cignal AI cannot independently confirm it, but it appears that Huawei’s internal DSP will account for the majority of the company’s DSP volume over the next 12 months.
As for NEL the company is actively investing in next generation DSPs and has several designs in the pipeline. These designs include a lower power 200G capable DSP as well as another design which is expected to operate at 400G.
One interesting comment made to us by a vendor purported that Huawei will sometimes strike agreements with component suppliers in order to have exclusive access to a particular component within China. There is precedent for this type of arrangement, and it appears possible that Huawei has struck an exclusive arrangement with NEL as an external DSP supplier. Anecdotal evidence seems to support this conclusion – ZTE has now moved 100G production exclusively to Acacia-based hardware, and away from NEL.
It appears that Acacia has failed to made headway at Huawei with its DSP, despite the superior performance it offers. Acacia’s lack of success at Huawei could be a reflection of the fact that ZTE is such a large customer of Acacia, and concern of this fact by Huawei. It may also be attributable to a reluctance on the part of Huawei to deploy three different DSP types (i.e., in-house, NEL and Acacia). Our conversations with Huawei engineers indicated that they are most interested in Acacia silicon photonics technology for CFP2-DCOs in 2018.
As we look at other equipment vendors, we understand that Cisco has significantly scaled back its internal DSP team following the selection of Acacia for the NCS4000 system selected by Verizon (see Acacia IPO Thoughts). We believe that Cisco is using Oclaro as its primary 200G CFP2-ACO supplier and has shipped 200G line cards to Verizon. If so, this would imply that the Oclaro CFP2-ACO is working successfully with the Acacia DSP. This is notable as Coriant is also using the Acacia DSP with another component vendor’s ACO module, and needs this to work in order to also ramp production of its own G30 DCI product.
CFP2-ACO Production Status
Through a careful triangulation of information from multiple sources, Cignal AI estimates that 6,000 CFP2-ACO modules have shipped to date (as of September 2016). Of this number, Oclaro accounts for almost all of the volume, and we suspect that the majority of Oclaro’s production has been shipped to Cisco, which is ramping up the module format across several platforms.
Fujitsu is shipping 100G capable ACO modules, but is experiencing light demand, as virtually all applications for this format require 200G capability. Fujitsu and competitor Finisar are racing to bring 200G capable modules into production, but customers tell us that these two vendors are 6 months away from achieving substantial production volume in this format. End customers universally agree that Fujitsu has the lead over Finisar in ramping 200G ACO production, an important development as these modules are exceptionally supply constrained at the moment.
Neophotonics is also continuing its pursuit of a 100G only CFP2-ACO based on traditional discrete components. The company is of the opinion that this brute-force approach with existing high volume components will yield a reliable and low cost 100G-only solution. Acacia is now sampling silicon photonics-based CFP2-ACO modules, but these modules are considered to be very early; or, in the words of one vendor, “pre-alpha samples.”
Overall, vendors indicated to us that they are reasonably happy with the 200G CFP2-ACO technology, but they are frustrated with the constrained supply. Oclaro remains the sole source for 200G CFP2-ACO volume production for the next 6 months. Cisco, Coriant and Ciena are the primary consumers of this technology and they are leading the rollout of cheaper 200G in metro and DCI applications. Our conclusion is that production volume in this format is growing rapidly and we forecast that cumulative shipments may breach 10000 modules by end of year.
Huawei does not expect to use CFP2-ACO in 2017. It will re-evaluate the solution next year as part of a transition to some type of CFP2 coherent pluggable format. Note that despite not having a targeted product, the company is currently pursuing the design of silicon photonics components suitable for an in-house CFP2-ACO at its Belgian R&D center.
The Emergence of 400G Technology
The optical networking industry is beginning to focus its R&D efforts on the next speed, which is 400G coherent, as well as higher density 200G coherent.
While equipment vendors are expected to first deploy 400G with non-pluggable discrete based solutions, the consensus is that 400G ACO modules are the likely outcome in the longer-term. This transformation is likely to entail upgrading CFP2-ACO operational speeds from 200G to 400G. Such a speed upgrade is required to keep ACO technology ahead of the 200G DCO modules that will sample in 2017 and then reach production in 2018. We suspect component vendors are currently pursuing 400G ACO R&D and would not be surprised to see announcements in 2017 coinciding with the availability of new 400G DSPs.
Details on 400G DSP development are hard to come by, though we understand that Acacia, NEL, and Clariphy are all actively investing in this area. Cignal AI also expects Ciena to introduce a new DSP in 2017 if it intends to continue its ongoing internal development. Nokia recently announced a new DSP which has a wide range of speed and reach capabilities. Given all of this interest, 400G DSP equipment and component plans should reach the public within the next 12 months–most likely sooner.
Density upgrades are of interest as well. New module types such as O-SFP (CFP4 size, 16W) have been proposed and should provide the needed format for 200G with a 2x improvement in density. All of these 200G and 400G coherent improvements present a rising challenge to the proposals for non-coherent 100G using PAM-4 that are promoted by Inphi and a new entrant to the arena, Ranovus, which had a compelling demonstration at the show.
Finisar, Inphi, Broadcom, Xilinx, among others, are now supplying prototype components and modules for 400G CFP8 short reach modules – the initial format for 400G Ethernet inside the datacenter. Real volume is still a far point on the horizon, but we expect to see a number of product announcements at OFC in March 2017.
Nokia CFP2-ACO and China Mobile
We had to do a little digging in order to better quantifythe China Mobile win that was announced by Nokia (see Nokia 2Q16). This win was somewhat remarkable for the fact that it employed CFP2-ACO technology at a Chinese customer – until this was announced all Chinese carriers were strongly against the use of this format. All our information indicates that this win does not truly represent 30% of China Mobile’s 100G coherent deployments; rather, it is just 30% of some fraction of China Mobile’s deployments that we cannot yet quantify.
It is also worth noting that Nokia’s product isn’t a true pluggable CFP2-ACO, but instead discrete components on a non-pluggable daughter card. Cignal AI’s conclusion is that Nokia’s win wasn’t quite of the magnitude assumed by most, and it does not dictate a mandate for CFP2-ACO technology at China Mobile, which is China’s largest consumer of 100G coherent technology.
The meetings and conversations we had during ECOC yielded some great insight on what we can expect for the market in 2017 and beyond. Beyond the tactical information provided, these interactions allowed us to sharpen our forecasts for worldwide 200G ports and total Chinese coherent demand in 2016/7 – insight we will share in our next article.
We also gained a better understanding of the rollout of the upcoming DCI platforms which will constitute the majority of 200G growth in the coming year. Cignal AI thanks the many suppliers and end customers who shared their market developments and observations with us, and we look forward to sharing our Active Insight with you.