January 27, 2012

Systems Vendors Consider Next-Gen Interconnects

At the recent Optical Internetworking Forum workshop in Santa Clara, Calif., a variety of optical systems vendors discussed the challenges and opportunities in building next generation optical interconnects.

Developing common standards for connecting boxes from different vendors is important to drive costs down and to provide a viable alternative to copper. Said Boris Golubovic, director & fellow of the Optical Products Group at TE Connectivity, “The proliferation of variants won’t help use because it will not lead to economies of scale.”

Replacing the existing copper cables in data centers promises to increase data rates and to reduce overall power consumption. However, optical interconnects face stiff competition from copper, which has been growing in capacity and shrinking in cost ahead of optical.

One of the biggest challenges lies in figuring out what kind of techniques will be used to increase the bandwidth on the fiber. Vendors are looking at a variety of techniques to increase the performance of each cable, including tuning the modulation scheme, coarse wavelength division multiplexing and coherent communications.

These types of techniques are practical in long-range communications because they add relatively little overall cost compared to the installation of the cable. Golubovic noted the next generation of submarine cables will carry 10 terabit/s. Carriers are willing to use expensive modulation techniques because there are only a handful of fibers — the most expensive part.

However, in the data center, the cost of short-reach and indoor optical lines generally is driven by the cost of endpoints rather than the number of fibers, he continued. Consequently, vendors need to think about adding more performance without using exotic and expensive multiplexing techniques.

Other Options

Another approach is to incorporate more fibers into each cable, but this presents alignment challenges when the fiber count grows to more than 72 optical fibers.

Bob Brunner, research scientist at Ericsson Research, expects more work on extending the concept of a chassis from a few meters to a hundred meters to accommodate new data-center designs. He predicted a movement toward multichip modules, in which the photonics is built into the chip directly.

However, achieving these longer distances could be a challenge for the lower-cost multi-mode fiber now preferred in data centers. This may create a need to use single-mode fiber, which is more costly and challenging to install.

Brunner expects to see the first optically enabled chips in 2015, and it could take another three years before these are rolled into commercial telecommunications systems.

Vendors are exploring several options to lower the cost of components, some of which could introduce problems in systems design down the road. For example, one idea is to incorporate forward error correction into the system, which would allow the use of cheaper lasers.

However, Mike Dudek, senior manager/Signal Integrity at QLogic Corporation, said this approach could introduce latency into the system. While a few nanoseconds of latency is not a concern for many types of communication, it could be a deal-killer for intensive read/write database applications. He urged the industry to start thinking about a latency budget when it designs new systems so that seemingly minor causes of latency do not create major problems down the road.

— George Lawton