SFP+ transceivers emerge as key 10GbE trend
Great strides have been made over the past few years to bring 10-Gigabit Ethernet optical interfaces to market, and the technology has progressed remarkably in a few short years from lab demonstrations to widespread commercial shipments. At the same time, network bandwidth needs continue to increase, particularly in core data center and backbone connections, as desktop and server connections upgrade to Fast Ethernet and Gigabit Ethernet speeds.
However, the 10-Gbit/sec market ramp has been slower than forecasted to date, with cost cited as a major reason. In previous transitions between Ethernet generations (e.g., Fast Ethernet to Gigabit Ethernet), volumes have crossed over when the cost of the new generation is approximately 4× the cost of the previous generation-so, 10× the speed at 4× the cost. Right now 10-Gbit/sec costs sit at around 10× to 15× the cost of Gigabit Ethernet, which slows the transition to serial 10 Gbits/sec as users explore alternatives such as link aggregation. Copper-based approaches to 10 Gbits/sec have now been standardized and are moving to market as well, but will be limited to relatively low port densities by power dissipation and crosstalk issues in the near term.
Within this context, a new optical transceiver form factor has emerged. The SFP+ high-density, low-cost optical transceiver should help enable the 10-Gigabit Ethernet transition in enterprise networks by meeting a variety of customer concerns better than previous modules.
SFP+ is a variant of the SFP optical transceiver (also sometimes called a “mini-GBIC”). The SFP has been shipping in volume for years in Gigabit Ethernet and 1-, 2-, and 4-Gigabit Fibre Channel applications. The SFP+ module enhances the mechanical form factor of the SFP to add improved signal integrity and EMI shielding appropriate to higher data rates, and defines new electrical interface specifications.
The ANSI T11 standards group has been working along with signatories of the SFP+ multisource agreement (MSA) to standardize an SFP+ module for 8-Gigabit Fibre Channel as an upgrade from existing 4-Gigabit Fibre Channel SFP modules in SAN applications. The SFP+ MSA group is now also working on defining technical specifications for 10-Gigabit Ethernet SFP+ modules.
Several different optical transceiver form factors now compete for 10-Gbit/sec applications, each with its own advantages and disadvantages (see Figure 1). The XENPAK and X2 module form factors are shipping in volume today and are the most mature 10-Gbit/sec approaches. They support all required reaches, and their four-lane XAUI electrical interface eliminates the need for very high-speed design on the host card. The XFP form factor enables higher densities, supports most reaches, and typically costs less than the XENPAK and X2.
The SFP+ module should become the highest density and lowest cost option for 10-Gigabit Ethernet optical links, with the tradeoff of more high-speed design work up front for the host-card designer.
SFP+ optical transceivers will enable the highest port counts per card of any of the 10-Gbit/sec optical modules. The table gives a comparison of the various modules, along with the maximum number of ports possible per line card for each form factor, a number constrained by both size and thermal management considerations. The SFP+ will offer densities comparable to Gigabit Ethernet optical modules and to copper Gigabit Ethernet cards with RJ-45 connectors.
SFP+ optical transceivers also have the potential to offer significantly lower cost than the existing modules. The cost reductions will come from eliminating redundant silicon and significantly simplifying the design and test of the module to make them more similar to low-cost Gigabit Ethernet optical transceivers.
Figure 2 shows simplified line-card architectures using each of the optical transceivers; note that the overall SFP+ module architecture is identical to Gigabit Ethernet. In addition to a more streamlined host-card architecture, the higher port counts that SFP+ enables mean lower costs per port, as line-card costs are amortized over a larger number of ports.
SFP+ modules are sampling now; however, technical items remain to be resolved before SFP+ devices move into production. The SFP+ MSA group has been hosting industry efforts to complete a standard specification for SFP+, and module, system, and IC vendors have been participating in a variety of technical calls, presentations, and face-to-face meetings to agree on high-speed, electrical, mechanical, and digital diagnostics specifications.
One of the key topics of discussion is the high-speed electrical interface definition. The optical link specifications and distances for 10-Gigabit Ethernet are already set by the relevant IEEE 802.3 standards, so current discussions involve defining electrical specs to ensure compliance at the optical output and input of the module. A good deal of simulation and experimental work has been done to characterize signal degradations from various trace lengths and geometries, board materials and layout, connector types, etc., and IC and module vendors have presented materials on the capabilities of the current state of the art in component technology.
There has been some discussion in the SFP+ MSA forum about whether conventional “limiting” modules (which incorporate a limiting post-amplifier inside the module to put out a digitized “1” or “0” output) can meet the required link distances, or whether a “linear” module (which ideally puts out an exact, nondigitized copy of the input signal) is needed in conjunction with a receive equalizer or electronic dispersion compensation (EDC) IC. For LRM applications that require extended reach on multimode fiber, the EDC required by the IEEE standard will likely be located on the host card to meet the power targets for the optical module. This architecture will require a linear module to pass on an equalizable signal to the EDC engine. For short-reach (SR) multimode and 10-km singlemode (LR) links, there may be some advantage in using a limiting module to achieve compliance, as this approach would imply no need to integrate EDC on the host card to support LRM.
Numerous industry participants are working vigorously on the high-speed electrical specifications, as well as the open mechanical, firmware, and other issues. The target for closing the substantive open issues in the SFP+ MSA specification is the end of 2006, with official spec closure likely happening in early 2007.
The definition of rescoped optical specifications targeting cost-sensitive data center applications could reduce optical module costs even further. This “data center reach” module would trade off slightly reduced optical specs and link lengths (e.g., 30 m on OM2 fiber and 100 m on OM3 fiber, long enough for most data center applications) in return for the markedly reduced costs (up to 50% lower) that would result from simplified design, higher yields, and reduced manufacturing testing costs. The result would be a nonstandard optical interface, but the idea has been relatively well received by industry participants.
The SFP+ is not a magic bullet for 10-Gigabit Ethernet conversion; innovation in optics, silicon, and module design must continue to move optical transceivers down the cost curve. What the architecture choice can do is help 10-Gbit/sec get on the right curve-in this case, the same types of cost and volume curves as Gigabit Ethernet. SFP+ will enable high-volume 10-Gbit/sec production and offer the lowest cost, highest density optical module architecture available; it is expected to be the “end game” optical transceiver for 10-Gbit/sec.
The industry has already shown significant interest in SFP+ and its value propositions, with multiple vendors planning initial releases in 2007. System vendors are expected to move to SFP+ over the next 1 to 2 years to realize the commercial promise of 10-Gbit/sec, as industry observers start looking ahead to even higher connectivity speeds in the future.
Scott Schube is strategic marketing manager for enterprise products for Intel’s Optical Platform Division (www.intel.com), which makes high-speed optical transceivers for the datacom and telecom markets.