Once again, we have a natural affinity for the proposition of taking advantage of electronics, such as higher order modulation, to better enable optics by squeezing more data onto a single wavelength. However, by the time, Inphi is ready to ramp up production fully on its four-level pulse amplitude modulation devices, both InnoLight Technology and Kaiam will have sold a large quantity of their optical-centric componentry to both hyperscale Data Center (DC) operators as well as to OEM providers. Hopefully, within this time, Kaiam will have joined its direct competitor in filing for an IPO (assuming a certain amount of stabilization in the stock market) by stressing the former, sexier types of customers, “Google Plus,” forgive the pun. So, in effect, to whatever extent Inphi has success with PAM-4, for all intents and purposes, it will perhaps indefinitely, likely be more of a situation akin to itself and the optic combo being ships just passing in the night.
Kaiam states that in moving to 25G per lane in 100G transceivers with its hybrid packaging technology that the cost of the optics goes down and there is domination with the expenditure on electronics – the biggest BOM outlays in its 100-gig QSFP include the high-end CDRs, the drivers, and the TIAs. So, there is the assertion that the use of multiple wavelengths becomes cheap enough that sophisticated electronics are unnecessary. In addition, Kaiam has shown (in contrast to the findings of one or more large optical components vendors) that PLCs can have a smaller footprint than free-space optics, and consequently, it has an easier time making everything fit together.
Certainly, there is not a better endorsement in the industry than Google investing in a vendor’s products given the company’s thoroughly comprehensive process in finding the lowest cost solutions. Nevertheless, it and other buyers may be resorting to these optical-based devices because they desperately need considerably less expensive components, and PAM-4 is not commercially available, yet.
Moreover, there is an inclination in the industry that components engineered to the very unique network requirements of a Google, which is the case with both Innolight and Kaiam, will necessarily mean widespread market adoption in all DCs. (SDN is a good example of this expectation that may not turn out to be the case.) While we believe Kaiam when it tells us that its product will eventually be adaptable to any environment, at a minimum, the ability to use PAM-4 in both copper and fiber situations will be a major advantage in a lot of cases.
Inphi contends that its internal costing models demonstrate that PAM-4 is 20 to 30 percent cheaper at 2x50G (2x25 each) than a 4x25G approach. The same 25G bandwidth components are being leveraged (optics, fiber plant, QSFP cages, etc.) at the same 28GBaud rate as NRZ, while achieving twice the speed. Customers that care about publicizing such matters can legitimately state that they will be using 50G before the development of a standard for that rate. (Inphi has also introduced a 50G single laser/lambda product.)
Naturally, the actual cost for PAM-4 needs to be determined based on volume production of the final specifications resulting from feedback from potential users sampling the chip, which clearly has a high level of sophistication. In addition, one has to assume that there is room for technology from both Kaiam and Innolight to come down a cost curve. On the other hand, competition from other PAM-4 suppliers will drive pricing lower as well.
Regarding the possible limitations on length with PAM-4 resulting from link budgets and SNRs, Inphi tells us that the situation is much different than when attempts were made to use the encoding in the development of the 100GbE standard. The supplier says that there has been a lot more data based on optical links working that has resulted in confidence in getting to lengths much higher than 500 meters. Of course, there can be a huge difference in performance with any kind of technology in the laboratory as well as in field trials, compared with actual deployment of working systems.
In addition to the optimism about eating into the 4x25G NRZ market, Inphi is bullish about it prospects for using PAM-4 as a single 40G for single mode fiber instead of a 4x10G QSFP CWDM solution based on NRZ. It is another indication of the market needing to totally rid itself of the widely used template (which has been around for years) that 40G is on its way out. Yet, our intuition suggests that a single 40G laser approach will not inherently be cheap.
[written by Mark Lutkowitz]