Facebook: Co-Packaged Optics Conundrum

While we have recently pointed out that co-packaged optics will not be the great enabler of Silicon Photonics (SiPh), the concept of markedly shortening the distance between the switch and discrete optics could potentially be useful. As the problems in using SiPh are inherent to the technology itself, it has the additional disadvantage of the area available on the chip for a co-packaged approach being quite constrained. Although Facebook is a big fan of the concept, one gets the impression that there has been an inadequate realization on the part of the company that is not business as usual anymore when it comes to optical development. The unmistakable rupture of the ecosystem was almost exclusively caused by the actions of the hyperscale data center operators, including specifically by the social media giant itself, with its infamous, indiscriminate, and irrational call for an unsustainable dollar per gigabit goal. So, Facebook’s present commands to the market concerning R&D expenditures as it relates to 400G (with co-packaged optics as a possible subset) as well as other pressing matters, will not necessarily receive anything resembling an automatic response anymore. In addition to providing substantially more financing, the hyperscale players will now be forced to take responsibility for the re-creation and the running of an adequate ecosystem to support next-generation gear in order to keep their internal networks at a state-of the-art level.

Regarding being in denial about the current situation, and relating specifically to co-packaged optics, Facebook not only nonchalantly “called on networking vendors to accelerate development in this area,” but evidently wants the various types of industry players to reconcile answerability as to matters like the testing and the yield, whether they be optics, ASIC, etc. people in putting together this costly laminate. In this new world, we anticipate Facebook will have to take a proactive role in the allocation of assignments, and in dealing with the inevitable finger pointing.

Another drawback of co-packaged optics solution is that it is a classic case of the “chicken and the egg.” The cost needs to come down, but it will only happen with the necessary volume. The hyperscale folks will likely be facing a similar situation as Verizon with NG-PON2 in going off-budget to ensure the designing of a high first-cost device.

There is certainly risk in putting a lot of money into a substrate involving co-packaged optics, and there might be inadequate yield, at least at the beginning, making it even more expensive. Another threat to such investment is that the copper people still have some cards to play in this game. While the disadvantages with copper are that it does not have the density and flexibility of fiber, it will be cheap, and it has the potential for pushing off co-packaging for some time.

Thus, going as far back to 2010, with its version of chip scale optical packages/modules, IBM somewhat mitigated the chance for these negative consequences by doing joint design work using DARPA funding. Yet, given the impact of the anti-US military stance of just a small minority of people at a least some of these large “cloud” corporations, it would obviously not be a viable option for them.  However, DOE-sponsored projects to share the cost is a possibility.

Naturally, Facebook is also having some financial difficulty in general at the moment. It is unlikely that the Street would take a favorable view of huge optical investments, despite the necessity, again the reason to reasonably expect going off-budget. The other result, at least for some of the optical component suppliers, particularly the chip players, is that they can gain increased leverage in their future dealings with these dominant customers.

As always fibeReality does not recommend any securities, and this writer does not invest in any companies being analyzed by us.

To follow us on our totally separate, quick update company blog, which is exclusively on fibeReality’s LinkedIn page, please click here.

[written by Mark Lutkowitz]