The optics industry is approaching 50 years since the concept of Silicon Photonics (SiPh) was mentioned for the first time. During an upsurge in excitement over the technology in 2014, we were accused by one or more prominent engineering executives in the space about having an agenda, after we expressed our doubts. As its use went to higher capacity, there was a movement away from monolithic devices, and eventually the tendency was for a lot of vendors to get carried away by calling any new solution, SiPh. For the purposes of this discussion, we will confine the definition to those chips, which have a modulator. We truly believe that the last large marketing buildup for SiPh involves the use of co-packaged optics. It has heavily been promoted by Facebook, and ironically appears to be the most difficult technical challenge to pull off to date before we expect SiPh to kind of burn itself out as a pronounced portion of marketing messages.
Undoubtedly, SiPh has been successful with complex circuits like coherent receivers (situations with dual polarization and the quadrature arm) and with PSM4, with the ability to split the laser (although its pricing helped lead to a virtually unprofitable market for that device). There may also be a use case for SiPh with the potential future use of multicore fibers.
Still, as a general rule, SiPh does not offer any cost, scalability, or reliability advantages. In fact, there is a host of failure mechanisms. Whenever on-board optics take off, they can be built with discrete components. Moreover, no lasing on silicon seems to be on the short-term horizon, and even then, there are all these coupling and yield issues.
With the announcement of a new MSA group for Multimode Fiber (MMF) and other indications of a likely growth spurt for this medium at the hyperscale data center operators, it is useful at this time to look at SiPh as any kind of competitive threat. Clearly, the communication from SiPh suppliers is that they are selling engines with wonderful integration and “compatibility.” Yet, putting in a fiber pigtail/connector is a pain. There is less hardship with MMF with larger core sizes and looser alignment tolerances. In addition, SiPh packaging is labor intensive, as the last step has to be done manually in Asia, preventing high volumes. While there are vendors promoting packaging with passive alignment, this concept goes back to the 1990s, and we are not convinced, yet that there is adequate proof of its utility, especially with SiPh. Therefore, SiPh is always more expensive than MMF.
Moreover, VCSEL-based interconnect is lower power. Some SiPh suppliers do not even count the source laser in this evaluation, which can be coupled externally. Also, there is no intrinsic disadvantage by using 50Gbaud VCSELs, which will eventually enter the market. In fact, we are unaware of any supplier showing a manufactured SiPh 50Gbaud modulator – and even it happened, it is hard to imagine the avoidance of a lot of loss. With a 50Gbaud MZM interferometer, not even two kilometers can be reached.
Lastly, while our negative assessment of SiPh may appear to be good news for firms offering polymer solutions, as in the past, we are not bullish on this alternative in the present relating to Lightwave Logic. Lightwave’s approach has several technical concerns related to process compatibility, to a high VPi, and to challenges with high temperature as well as with water.
Turning our attention to co-packaging, it is certainly the new buzzword. The evangelicals of SiPh will say it has the only solution to pull it off, and part of the reason it is a valid contention is that in using both Indium Phosphide and Gallium Arsenide, it would not be truly co-packaging in the way it is being envisioned. Yet, of course, it is still hard to take SiPh seriously in this application, as it is again, still struggling with the fundamental issue of loss. In fact, the performance may be so poor that it will require some form of hybrid packaging with the III-Vs in order to make it work.
There is the idea that co-packaged optics will replace COBO because the latter is a two-piece connector solution. So, to the extent that one believes that onboard optics is a possibility, arguably the mistake of COBO was in buying into a relatively larger module to accommodate all sorts of reaches.
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.