by Martin Zirngibl, Finisar CTO
400G needed for explosive bandwidth growth everywhere in the network
It seems like 100G pluggable modules were introduced just yesterday but they are already running out of steam for leading-edge applications. 2019 will be the year of the early roll-outs of 400G across the entire network, for short-reach links such as server-to-top-of-rack switches in data centers, for leaf-to-spine connections inside data centers, for data center interconnection links (DCI) and DWDM metro networks.
But there is no one-size-fits-all solution
Although 400G will appear almost simultaneously throughout the network, there is no one-size-fits-all solution. Cost, power and footprint per bit need to be carefully tailored for each application, requiring a diverse set of technologies to optimally address the various distances and capacities. There is a strong trade-off between power consumption, footprint and cost of transponders versus their performance in reach and fiber capacity, which translates into the cost of the link. For instance, an electrical copper link based on direct attach copper cables (DAC) consumes on the order of 1W/100G and costs a few dollars but it can only reach three meters. On the other end of the spectrum, an optical transmitter transponder for long haul (several thousand km) burns up to 30-40W/100G and costs about three orders of magnitude more than the above copper link. A network provider must carefully choose the interconnection technology in their network. If the technology has too much performance, then there is a penalty to be paid in terms of power and cost. However, a transponder with low performance may constrain the scalability of future network expansions.
Multimode fiber (MM) will increasingly replace copper for high volume server to TOR links
There are three types of optical interconnection technologies today: optical multimode, optical single-mode direct detect, and optical signal-mode coherent. Optical multimode transponders can be as low as 2W per 100G power consumption. They are based on Vertical Cavity Emitting Lasers (VCSEL) which are directly modulated. VCSEL-based multimode technology is very cost-effective since VCSELs can be processed and tested on wafer scale and the alignment tolerances of MM is very forgiving, allowing low-cost packaging technology. The main disadvantage of MM is their short reach: At 400G it is typically 30-70m, limited by modal dispersion. Multimode links can come in the form of parallel fibers, where there is typically one fiber per 50G lane capacity; or wavelength division multiplexed channels, where there are 4 lanes of 50G per fiber, the latter reducing the amount of fiber four times.
It is expected MM links will become the “new copper” and therefore their volumes will explode. Indeed, the great majority of links in a data center are between the servers and the top-of-rack switch and currently, they are based on electrical copper cables. Their volume dwarfs that of all the other optical links in a data center. But as SERDES speeds increase to 100G, the reach of copper links will be significantly curtailed, thus many of these high-volume links will need to become optical. This transition represents a significant opportunity for optical multimode VCSEL technology and they will be even more ubiquitous in all data centers.
For links beyond 100m, SiP and InP are competing for lowest cost solution
Once reach needs to extend beyond 100 meters at 400G, single mode optics is required. There are several technology options: Silicon Photonics (SiP), direct modulated lasers (DML) and externally modulated lasers (EML). It is generally accepted in the technical community that SiP-based transponders are best suited for parallel fiber, with one 100G lane per fiber, because this technology allows sharing of a single laser source amongst multiple modulators, each one creating a 100G lane. The sweet spot for parallel-fiber solutions is around 500m applications, although technically SiP can go longer reaches, the cost of the parallel fibers starts to outweigh the transponder cost savings. For links that need to reach 2km or 10km, most network providers, therefore, prefer duplex fiber, reducing the amount of fibers fourfold. Since either four or eight separate wavelengths are now required, the above described laser sharing is no longer an option thereby making DMLs and EMLs appealing. However, SiP solutions are also competing in this space and the jury is still out about which technology will eventually dominate.
Coherent transponders will dominate the DCI and metro applications
Of course, the bandwidth explosion will ripple through data center interconnection, metro and access networks as well. It is the consensus in the technical community that for distances >30km and rates of 100G and above coherent technologies offer the best trade-offs between cost of the transponders versus cost of the fiber plant. One big advantage of coherent is that the fiber plant does not have to be engineered, because a coherent transponder can undo most of the transmission impairments on its own. Coherent transponders can also bridge very lossy links without need of optical amplification, making them attractive even for access networks.
In 2019, Finisar expects to ship 400G products across all reaches. We are now sampling single mode parallel 400G modules based on internal Silicon Photonics (shown at ECOC 2018), supporting up to 500m links. For longer reaches (2km) and duplex fiber, 400G FR8 pluggable modules based on InP components are also available. Direct detect at 400G can even go much longer; our recently demonstrated QSFP-DD “eLR8” module with directly modulated 50G lasers can transmit data up to 30km through unamplified grey optics links. For coherent optical links we have developed what we believe is the highest integrated, lowest power coherent optical components on the market. The integrated tunable transmitter receiver assembly (ITTRA) integrates all the optical and control functionality into a single gold box. These very small form factor components are designed to fit into future coherent pluggable modules such as QSFP-DD, OSFP, CFP2, and CFP4. Its bandwidth of 40GHz will enable capacities of about 600Gb/s per single wavelength, something that was unimaginable only a few years ago. The low power, cost and footprint profile of the ITTRA will make coherent competitive for new non-unamplified applications that have been, up to now, the undisputed domain of direct detect technology.
Only the future will tell where the boundaries between multimode, single-mode direct detect, and coherent lies. There are powerful new technologies such a Silicon Photonics, low-power CMOS, chip-on-glass packaging technology and high-volume applications in data centers and 5G Mobile networks that will push the boundaries in unknown directions. What is clear is that Finisar has all the tools and technologies in-house to offer products that are cost and performance competitive for any of the emerging optical interconnection needs.
