Fiber capacity improvements have been driven by carriers deploying more fibers per bundle and by the use of dense wavelength-division multiplexing to increase the number of wavelengths carried in a single fiber. The rate of information transferred per wavelength has risen from 2.5 to 10 and, now, 40 Gbits/second.

In the central office, carriers terminate the majority of optical signals with inefficient Sonet equipment. Wavelength switches have been introduced to address that problem, but there is more capacity in a single fiber pair than there is in the largest of them. What's needed is a step-function change in the scaling of the wavelength switch to equalize central-office switching capacity with fiber-transport capacity. The two scaling approaches are photonic switches, which use reflective techniques to switch the path of the wavelength, and electronic switches, which convert the wavelength to electrical then switch the bit stream before turning the signal back to optical.

The primary argument for using photonic switches is lower cost by eliminating transponders that perform the optical-to-electrical-to-optical conversions. But most carriers considering photonic switches still use a transponder for performance monitoring, switching signals at a finer granularity than the wavelength and simplifying protection design.

Therefore, the cost of the transponders must be calculated separately from the switching fabric. When transponder cost is excluded, photonic switching is at least two orders of magnitude more expensive than electronic switching.

Today's electronic switches are just 512 ports, while photonic switches have been announced with 1,000 ports, scaling to 4,000. But the perception that electronic switches cannot scale is no longer correct. This is because electronic crossbar chips have arrived that support 128 x 128 ports each, with 2.5-Gbit/s serial interfaces. Interconnecting these chips in a rich mesh, such as a Clos architecture, allows electronic switches that can practically scale to 2 million ports.

Implementing such a switch requires two other key technologies: 2.5-Gbit/second serial streams between chips across backplanes, and a dense, low-cost in-terconnect between shelves. This is possible using vertical-cavity surface-emitting laser array transmitters, driving 12-way parallel fiber ribbons.

Bringing together crossbar chips, serial streams and VCSEL technology will enable very scalable wavelength switches at prices below what is possible with photonic technology. And since electronic technologies are well understood, it can be done with much less risk.

Ian Wright is senior vice president for optical engineering at Altamar Networks (Mountain View, Calif.).