MARK LUTKOWITZ CLASH OF TECH
Clash of technologies: NRZ vs PAM-4 signalling
A t ECOC 2016, there will be extensive discussion of all types of optical component solutions, which are in conflict with each other, in terms of technical applications or business opportunities, such as ACO versus DCO modules or transceivers versus on-board optics. One of the most pressing matters in the industry regards the apprehension level by vendors over the extent to which the newcomer, PAM-4, actually overtakes the incumbent, NRZ, in the marketplace. Historically, optics is driven by other industries, rather than the other way around. Therefore, as a result of an established knowledge base of PAM-4 in the electrical domain, more or less of a consensus in the industry has developed for certain apps to use that advanced modulation approach to move the optics forward. Nevertheless, in the vendor community, there still can be found some uneasiness about it possibly having marginal utility, especially with the notion that NRZ signaling could perhaps be pushed at least a little further regarding its capabilities. A key demarcation point will be the future emergence of 50G componentry. Mellanox Technologies is apparently planning to introduce an NRZ oering at that rate, a supplier, which has a track record for delivering product early, and which has secured a dominant position in the high-performance computing space. Conversely, the vendor appears to be the only significant player talking to the industry about a strategy that uses NRZ beyond 25G. So, it still remains to be seen whether it can be successful in persuading its Ethernet system customers of the superiority of its oering. In addition, there are no current indications of any
One of the most pressing matters in the industry are the implications of signalling protocol PAM- 4 overtaking incumbent NRZ, argues Mark Lutkowitz, analyst at fibeReality, LLC.
still have to be proven. Yet, with its past performance with semiconductors across the board, AMCC still needs to be taken seriously by competitors. Another factor that goes into the determination of PAM-4’s potential is the extent to which there is business beyond the unique requirements of Microsoft. In contrast to other hyperscale operators with capacity requirements at the high end, including Google, Facebook, and Amazon, the software giant built its data centers long before the others. Microsoft tends to have a good number of smaller data centers, which are in closer proximity to one another, with the goal of logically connecting all of them. In this kind of a situation, the severe signal-to-noise ratio with PAM-4 is not viewed as much of a hurdle by the user. A final potential obstacle for PAM-4 are some of the misgivings by at least certain component vendors. If it had to vote in the IEEE 802.3 standards body again, at a minimum, one of the biggest companies in the market evidently would have instead chosen NRZ. In hindsight, while probably no one disagrees that a reasonable and plausible argument can be made for PAM-4 in the electrical domain, such as with its use on copper cables, the amount of complexity involved, especially because of its newness in combination with optics, has resulted in at least a few prominent optical engineers concluding that its adoption just seems silly. Unfortunately, in general, there is a lot of groupthink that takes place at those 802.3 meetings that may or may not be based on reality – and may not necessarily reflect the plans of management back home. Nevertheless, for now, the industry has in eect made the decision to go full speed ahead with PAM-4 involving speeds above 25G. Supposedly, it is that solution’s market to lose. At the same time, there can often be a fragility with any type of perceived consensus involving a brand new technology.
technical problems or even the slightest delay in the PAM-4 ecosystem, including from the most prominent firm in the fray, Inphi, which otherwise would facilitate a greater consideration of the traditional signaling method. However, it is interesting to note that one chip manufacturer looking to enter the market does not expect an adequate ecosystem for PAM-4 for itself until 2019. An aspect to take into consideration is the exact timing of 50G I/O-based switches. Of course, when it came to 25G-based devices, the engineering diculties turned out to be quite considerable, and so any aggressive prognostications of elements at the higher rate should probably be taken with a grain of salt. Thus, there will be ample time for possible positioning of one signaling technology against the other, including from Broadcom, which has demonstrated 50G SerDes with PAM-4. The main long-term concern is that at 50G, circuit designers are discovering the electronics are starting to run out of gas. Applied Micro Circuits (AMCC) has demoed a 100G single lambda employing PAM-4, using a fairly high-speed, 56-gigabaud laser and a Lithium Niobate modulator, along with a lot of equalization. Since it is based on a simulation, and given that in optics, there are a lot of non-linearities, it is rare to see a good correlation between models and actual hardware. Moreover, even with technical feasibility, the economics
When it came to 25G-based devices, the engineering diculties turned out to be quite considerable.
| ISSUE 7 | Q3 2016
Made with <A HREF="https://flippingbook.com/online-digital-catalog" TITLE="Learn about FlippingBook for Online Catalogs">FlippingBook</A> - Online catalogs