Published Feb 28, 2017 by admin
While polymer-based data communications has found viable niches in home networking and in the automotive industry, there has been no success in bring polymers in the data center. Polymer datacom start ups have risen and fallen. One recalls, especially, Telephotonics, which had very ambitious plans to create a full range of polymer-based optical components for the telecommunications industry. Large specialty chemical firms – including Dow, DuPont and Dow Corning — have promised polymer optical components but said promises have not produced much.
CIR believes that this is about to change and polymer photonics is finally emerging as a viable solution for data centers. Whereas the old polymer photonics was mostly focused on plastic optical fiber (POF), the new approach is all about polymer waveguides. And while the old “polymer photonics” was technology in search of an application, today’s polymer photonics lowers the cost of handling “big data” in routers, switches and data centers which makes for an actual business case.
Backplanes will soon be lit
The point here is that the tidal wave of “big data” that is washing over data centers will create a need for powerful interconnects to connect up the PCBs inside boxes. First this trend will be seen in backplanes then in server motherboards. The obvious strategy here is to revision backplanes and motherboards as optical rather than electrical systems.
Designs for optical boards can be taken directly from the world of supercomputing where optical backplanes have been around for several decades. But the commercial aspects of optical boards for the data center are very different to the supercomputing world. The literature suggests that optical boards would be more expensive than the the equivalent electrical boards by as much as an order of magnitude.
However much the need is for optical boards in the data center, no OEM is going to consider pushing up the price of its routers and switches (except for the largest boxes) to accommodate the cost of optical boards, since few data center managers would have the budget for such boxes.
The need to reduce the cost of optical boards is partly being addresses by industry efforts to create standards for optical boards. But part of the cost reduction strategy must certainly address how boards can be better built – and polymer waveguides is part of this story.
IBM and the Others
IBM has evangelized for polymer waveguides for quite a number of years now in the form of scattered announcements and presentations. The starting point for understanding how polymer waveguides can create opportunities is an examination of what IBM has done in this space so far.
IBM favors polymer waveguides as the technology of choice for optical boards because of:
(1) the ease with which they can be fabricated and integrated onto a board
(2) the low cost of fabricating the waveguides themselves and
(3) their high density.
These all speak to lower costs in future. Why IBM is enthusiastic about polymer waveguide-based optical boards is presented in Exhibit I.
What we don’t know yet is how long before we see this kind of polymer optics in data center boxes. Both Cisco and Juniper have adopted optical backplanes for very large boxes in the past (not using polymer photonics). Ericsson and Oracle (via Sun Microsystems) continue to do so. So there is some precedent for deployment in the data center. If IBM is correct about the virtues of polymer waveguides it would be no surprise to see some partnerships emerge to bring polymer waveguide backplanes to the data centers sooner rather than later.
However, for optical backplanes and motherboards to really make an impact in the data center, CIR believes that some outstanding technical issues must still be dealt with. Thermal issues are still problem as is alignment. These will both need to be dealt with before optical boards become common in datacom boxes.
So the evolution of optical boards for the data center may be two-phased. Some early deployment over the next few years, with productization of optical boards occurring in a three to six-year time frame.
But there are plenty of uncertainties. One, of course, is that something other than polymer waveguides may win out in the optical board stakes. Other possibilities are glass fiber, POF, one of the many other kinds of waveguides, and active boards with embedded VCSELs.
Assuming that polymer waveguides make the final cut, however, there are a number of firms that one should watch to see how the optical backplane business is emerging. Chances are that IBM will be at the center of it all and that Juniper and Cisco will have a foot in the door. We suspect that some polymer firm will also be part of the supplier mix.
The new Dow Chemical-DuPont-Dow Corning conglomerate would seem to be the place to look in this regard, since all three have done R&D in the areas of polymer photonics. Sumitomo could be another player. There is room for start-ups too, we expect. But we note that polymer data com start-ups have popped up occasionally over the past twenty years, but do not seem to have been long-lived.