400 GigE is the first Ethernet standard not to incorporate copper cable. There is some talk about implementing 400 GigE for ultra-short reach links on a board or even board-to-board. But for the typical data center applications – rack-to-rack and beyond – it’s fiber all the way. An open question is how will the fiber be implemented? The options here being field termination and active optical cables.
Can AOCs happen at 400 Gbps?
The facts of the matter seem to point in both directions. A few years ago – Finisar, Molex and TE Connectivity – seemed to be aggressively pursuing 400 Gbps AOCs but their prototypes in this space have been deemphasized or dropped.
From what we have heard at the OFC trade show there is now a faction claiming that 400 GigE AOCs will never happen. Apart from the usual complaints that one hears against all AOCs (long-term durability and the difficulty with pulling AOCs through data center walls), we suppose that the biggest gripe against 400 GigE AOCs would be that current optical integration technology is simply not up to reliably building the large numbers of 400 GigE AOCs that the data center will ultimately need. CIR is more bullish than this, noting that:
- A lot of investment is being thrown at optical integration at the present time. It is hard for us to imagine that the state of the art in optical integration won’t reach to creating reliable 400 GigE AOCs within a few years; that is exactly when data center managers start to need such cable assemblies
- AOCs are now an immensely popular way of implementing fiber in data centers. Even large cloud providers such as Microsoft are making considerable use of AOCs in their latest data centers, a trend that one might not have expected five years ago or so.
Based on these positives, CIR remains quite bullish on high-data rate AOCs including at the 400 GigE level. In our recent analysis and ten-year forecast of the AOC market, we predicted that by 2020 sales of 400 GigE will have reached $820 million, although admittedly one reason for such a large number is that for the foreseeable future, 400 GigE AOCs will be extremely expensive – just as 100 GigE AOCs were at this stage in their product evolution
Cabling Requirements for 400 GigE
But the 400 GigE AOCs of the near future will be different structurally than the AOCs today. In particular, we are expecting a new wave of single-mode AOCs to start appearing. Until now, single-mode AOCs were implemented mostly by one company – Molex – reflecting that Molex’s liking for single-mode platforms more generally and also its adoption of the Luxtera technology for its early AOC products.
This time it is, as CIR sees it, different. Initially we expect a lot of 400 GigE to be implemented over MMF, including AOCs. After all, according to the IEEE standards for 400 GigE, implementations of this standard can be created on MMF up to 100 meters using sixteen parallel strand of 25 Gbps fiber (400GBASE-SR16).
But after that everything is SMF in the 400 GigE standards. From 500 meters to 2 km, 400 GigE will be implemented over four strands of 100 Gbps SMF (400GBASE-DR4). Beyond that SMF will still be used but with DWDM.
While over 2km is beyond the market usually considered viable for AOCs (there are actually a few exceptions to this rule), we think that the dominance of SMF will begin to take hold of the zeitgeist in the data center as far as media is concerned. While SMF is something of an oddity at the moment, SMF will inevitably become mainstreamed in the next few years, including in AOC design. In any case, how long will it be before a 16 x 25 MMF solution for less than 500 meters appears inelegant and shorter reach SMF AOCs begin to be implemented?
The bottom line for AOCs CIR therefore believes is that five years from now, many of them will be implemented on single-mode fiber and they will be sold into a data center market where SMF is no longer an eccentricity.
AOCs and the MSA Zoo
AOC makers are like arms dealers when it comes to MSAs. That is, they sell to all sides – every new MSA presents an opportunity for them to create a new kind of AOC. So the fact that the 400 GigE MSAs are in a state of flux might be considered a very good thing from the perspective of AOC firms. (Note: Transceiver firms are always saying for the next generation of standards there won’t be so many MSAs, but these promises are always broken!).
The bottom line then is – CIR believes – that a variety of MSAs will start to be implemented in a new breed of AOCs. CFP will probably be a popular option initially but there is also the CDFP MSA and the very new OSPF MSA to be considered.
CFP and 400 GigE: The way that the CFP group sees things, the appropriate form factor for Gen 1 400 GigE would be to use four CFP 4 modules; CFP 4 modules are the smallest of the three defined CFP modules. A CFP16 arrangement has emerged during 2016.
According to the CFP story around 2020-21 second generation of 400 GigE will emerge and will be based around a 50-Gbps signaling/laser core technology using NRZ. The module would be the CFP2. Essentially, this second generation 400 GigE will be squeezing the last drop out of the existing CFP technology.
CDFP: The CDFP form factor is already used in proprietary interfaces to interconnect (1) groups of high performance servers and (2) switch and router chassis. What CDFP has going in this regard is that its design philosophy is essentially the “small format pluggable” concept which made SFP+ popular in the 10 Gbps environment and QSFP in the 40 Gbps and 100 Gbps environment.
The CDFP module also provides a number of advantages including high-density; CDFP’s supporters say that it will be the densest high-speed 16-lane interface.
OSFP: Recently there has been an announcement that a new 400- Gbps MSA is in the works. This is the Octal Small Form Factor Pluggable (OSFP) optical transceiver module. The modules will support reaches from the data center to metro networks and, according to the MSA members, will be “slightly wider and deeper” than a QSFP module. And there seems to be no word as to whether OSFP will support AOCs. The small size and comparatively low power consumption of the OSFP aims to meet the requirements of large-scale data center operators. OSFP should be able to support 800 GigE and perhaps beyond, but no word yet whether OSFP AOCs will ever arrive on the market.
Based on what CIR is seeing and the analysis above, we think that 400 GigE AOCs will become a real product category quite soon and we would not be surprised to see 400 GigE AOCs to be the “flavor of the month” at say OFC 2018. And we expect that what these cable assemblies will look like will be mostly single mode with CFP and CDFP inside.