By Ferris Lipscomb, NeoPhotonics
Here’s a prediction for OFC 2018: 400G is now considered “slow,” so get ready for “1.2T” -- as in 1.2Terabits per second transmission, which is the next big threshold in optical communications. One Terabit is 1000 Gigabits, which in turn is 1000 Megabits. So a Terabit is a trillion bits. The reason people are now turning to Terabit transmission is that data centers demand it, and a new generation of electronics and optical components are making it feasible.
Web-scale internet content providers (ICPs) require ever increasing data capacity both within and between mega-data centers. In satisfying this need, the ICPs really care about only two things: cost per bit and density. Actually, they only care about the cost per bit. Density is important because lower density costs more. There are several kinds of density: Spectral density refers to the amount of optical bandwidth in a fiber that is required to transmit a given amount of information bandwidth. The higher the spectral density achieved, the more information can be transmitted in a single fiber, leading to fewer fibers needed between point A and point B.
Since for longer distances of 80km and up, running a fiber, including the right-of-way, trenching, conduit and cabling over long distances is expensive. The fewer fibers going long distances, the cheaper it will be. There is also “rack density”, which also comes in many forms, but is basically the amount of equipment required in the data center to transmit, receive and control the information traffic. More space requires more building area, air conditioning and electrical power, and therefore, more cost.
So, as long as there is traffic to use it, higher speeds and smaller sizes are always desirable. But there are limits to what can be achieved. For instance, the advent of 400ZR meant a 400G coherent transponder in a compact DD-QSFP package for 80 km datacenter interconnects. This is indeed a high speed and high density interface, but it also presents some severe challenges and limitations. To fit a full 80 km coherent DWDM transponder in the size constraint of the DD-QSFP, which after all was designed for electrical and optical transceivers going less than 2 km, a new generation of ultra-compact optical components must be developed, and will likely not be available until 2019.
Furthermore, to fit within the tight 12 – 14 W power budget of a 400G DD-QSFP coherent transponder a new generation of DSP is required that not only moves to the 7 nm CMOS node, but also dispenses with any non-essential capability. This new generation of DSP will not be available until sometime in early 2019 and comes with some limits to performance. The 400ZR DD-QSFP would be limited to 400G using 64 Gbaud and 16 QAM and would also be limited to Datacenter Interconnect distances of 80 km or so. It would not be able to address higher speeds or longer distances for metro and long haul applications. Some DSP vendors have been talking about adding extra chromatic dispersion compensation to cover metro/regional distances in a 400ZR transceiver. However, due to the limited forward-error-correction (FEC) coding gain defined by OIF, the coverage cases of metro/regional networks might be limited. Also, a larger form factor such as CFP2 would be needed.
Here is where the 1.2T approach enters. High speed, 64 Gbaud coherent modulators and coherent receivers along with ultra-narrow linewidth tunable lasers capable of supporting 64 QAM are now available, but in form factors that cannot fit into the DD-QSFP. Furthermore, DSPs using the 16 nm CMOS node are becoming available in early 2018 that can achieve 600G per wavelength using 64 Gbaud and 64 QAM. They do have higher power consumption than the future 400ZR DD-QSFP, but can still achieve the “rack density” needed by using small “daughter cards” within the equipment, rather than pluggable transponders located at the equipment face plate. The power is thus dissipated over a larger area.
And some of the new DSPs are powerful enough that two wavelengths can be operated using a single dual-core DSP to make a card with 2x600G for a total of 1.2 T. This has the same spectral density as 600G, but better “rack density”. Furthermore, the DSPs and FECs are powerful enough that the same card and optics can be used for metro and long-haul distances, albeit for lower, but still fast, speeds. For example, the card could be programmed to do 2x400G over metro distances using 64Gbaud and 16 QAM or 2x200G over long-haul distances using 64 Gbaud and QPSK.
Another key advantage of a dual-wavelength 1.2T daughter card is that it can be used in an arbitrary metro or long-haul system, which has a pre-determined optical signal-to-noise ratio (OSNR) requirement, by dialing up the transmission speed continuously until the system performance limit is reached. For example, in a specific metro link, instead of operating at 2x400G, a 1.2T daughter card can operate at 2x483G to obtain an extra capacity of 83G per wavelength.
The main drawback to the 1.2T daughter card versus the 400G DD-QSFP is that pluggability is sacrificed. The card, or transponder, must be deployed when the equipment is installed and cannot be added later if needed. However, for many DCI applications the desire is to deploy as much bandwidth up front as possible, so a lack of pluggability is tolerable. And the 1.2T daughter card has the compelling advantage that it can be available in 2018 rather than 2019 or 2020. Therefore, OFC 2018 figures to mark the “true feasibility” of Terabit level transmission and it will be interesting to see the moves and announcements the industry will make there in support of this next step in optical networking.
Posted: 23 February 2018 by
Ferris Lipscomb, NeoPhotonics
| with 0 comments