Fiber, free space, or silicon? - Challenges in optical backplanes, Part 2
While work continues within the industry to prolong the life of metallic interconnects, optical technologies have evolved in the hopes of bringing cost in line with market demands. To meet those goals, most work surrounding optical interconnects has centered around reducing the cost per bit by either pushing single-wavelength bitrates to extremes, or by infusing multiple, lower bitrate optical transceivers into a single package. However, the cost of optical interconnects can also reap benefits just by being implemented in backplane applications, simply due to reach.
Because backplanes are typically short-reach applications, optical transmission only requires shorter fibers or polymer waveguides about 1 meter in length. These shorter transmission distances not only result in favorable link power budgets for optical interconnects (just a few dB of path loss between connectors is typical), but also imply lower receive sensitivities. Together, this enables the use of less sophisticated, less expensive components to implement an optical backplane, says Chuck Byers, Technical Leader and Platform Architect, Enterprise Networking Group, Cisco Systems, Inc. (www.cisco.com).
"One can tolerate a bit of loss in connectors, so alignment tolerances can be relaxed and connectors with higher density or lower cost can be used," Byers says. "Together, these factors allow for lower cost, more compact, and less power hungry optical components - all deemed important properties for backplane applications. The comparable loss through metallic backplanes and connector systems for SERDES signals at tens of Gbps is at least an order of magnitude higher than optical signals. This potentially leads to an opportunity to use lower cost optical components to achieve shorter reach, albeit still adequate for backplanes."
So, assuming optical interconnects continue to approach market parity, what technology is best suited for an optical backplane?
Free space, silicon photonics, or fiber?
To date, Byers identifies three classes of optical transport systems with the potential to replace copper as the main interface for backplane communications, each with a different approach to the challenges of optical data exchange:
- Free space optics provides high bit rates and low bit error rates, immunity to electromagnetic interference, and other benefits, using air as the only transport medium. However, free space technology is still immature, and can be extremely vulnerable to variations in temperature and vibration. As a result, free space optics is presently still somewhat of a wildcard for in-system communications.
- Silicon photonics represents an intersection of technologies that could yield the size, cost, and power savings to make optical interconnects viable on the backplane. Currently, a primary area of silicon photonics research focuses on improving the density of pluggable optical interface modules - which are already commonly used on front panels - to make them applicable in backplane settings.
- Fiber array interconnects using parallel optics are capable of interfacing dozens of 10-25 Gbps fiber streams while using only a fraction of the power, area, and backplane height available on a typical board. By routing a subset of fibers with blind mate connectors from each board position to all other board positions, a mesh interconnect can be created that eliminates the need for a centralized switching infrastructure.
Although none of these alternatives have produced bandwidths exponentially higher than what can be achieved with advanced SERDES technology in a reasonable, commercial-ready system, comparable data rates have been demonstrated on fiber optic backplanes. Combined with the fact that fiber optic interconnects can be implemented in dual star topologies for economic purposes, this gives them the best chance at eventual market adoption, Byers says.
"I believe fiber mesh interconnects are the surest bet," he suggests. "The big advantage of fiber-based backplanes to copper-based backplanes is the aggregate theoretical bandwidths of the hundreds of fibers they contain. Each single-mode fiber or polymer waveguide can theoretically carry over one Petabit per second (Pbps) of traffic split over thousands of wavelengths using Wavelength-Division Multiplexing (WDM). If you have on the order of a dozen boards interconnected in a mesh of these fibers, there is over 100 Pbps of theoretical capacity in a box - this is future proof. However, it is unlikely any reasonable board-level component will be able to load a single fiber to anything close to those theoretical data rates; we will have to push very hard to get more than a few hundred Gbps per fiber in the foreseeable future, with a maximum of a few dozen wavelengths operating at 25-100 Gbps each. This is still somewhat more future proof than a large mesh of parallel 25 Gbps metallic backplane connections. [It also supports] many more cycles of Moore's Law evolution, especially as the laws of physics are going to be kinder to scaling optical signal capacities than they will be for scaling metallic SERDES rates."