The Great Gigabit Backplane Shootout - Question
#11
For most system manufacturers, building an entirely new system that requires their customers to remove an existing chassis with all its line cards and purchase a new chassis and new line cards is not an option. Their engineering efforts are focused on increasing the bandwidth in an existing system. The backplane transceiver manufacturer does not have the luxury of dictating the system parameters. In fact, more often than not, the systems were designed for speeds vastly slower than the current requirements.
In the future when the economic environment is more conducive to the development of an entirely new chassis, the system vendors will need to employ the best design practices in conjunction with the SERDES manufacturers to ensure the highest system level performance possible and ensure future upgradeability.
Materials, manufacturing processes, and design architectures will all have to improve simultaneously in order to keep the complexity, cost, and power consumption of transceivers within acceptable limits for system manufacturers. If cost and power consumption are not contained, system manufacturers will not have a good reason to upgrade to higher speed SERDES technologies.
For system designers, improvements must come from all areas of the design, not just the silicon. For legacy applications, improvement of materials and connectors is not an option, and silicon offers the only degree of freedom. Depending on the system roadmap, these two paths may also be inter-dependent (i.e. new products designed for legacy must be able to migrate to new backplane.. at some cost). I don't think I can provide an answer that would encompass all of the permutations.
For new chassis development, improved materials and connectors command little or no price premium. Such improvements reduce the complexity of silicon. Reduced cost of 10Gbps silicon, combined with same-cost connectors & materials, results in more cost-effective systems. Watt-per-Gigabit, as well as price-per-gigabit will be lowered as a result.
Yes, for new designs. In older designs where you have an existing installed base, it may be more cost-effective to develop transceivers that are able to operate in the existing environment.
An immediate problem with changing material sets is that it eliminates the upgrade market. There are hundreds of thousands of modular switches and routers in the field today. The owners of these boxes will want to upgrade the bandwidth handling capabilities of this equipment over time without buying a completely new box ("forklift upgrade"). Current equipment makers desire an upgrade option to avoid the potential of losing the installation to a competitive solution. Considering the current economic environment this will be the biggest demand for backplane transceivers for the foreseeable future.
The question of new materials and connectors argument is really subtle misdirection. The optimal solution is one that doesn't increase power and can utilize existing materials and manufacturing process. This is specifically the objective of the 4 PAM full duplex EchoWave architecture. For new boxes a 10% investment in slightly more expensive materials such as Getek can provide 1-2 dB of SNR improvement to a system. Better connectors can slightly improve the level of crosstalk seen in a system but all of this adds cost. While there is little other choice for other solutions, if a full-duplex PAM-based transceiver is utilized with these improved materials it can operate at even lower power since fewer taps are required by the equalizer and echo canceller.
In general, silicon implementations are more cost-effective ways to overcome signaling problems as opposed to improved material and connectors. In particular, pre-emphasis has so little overhead that it is certainly the most cost-effective way to solve the signal integrity problem for backplanes and copper cable.
For new systems, incorporating a low loss material such as "Rogers" or optimizing the impedance match of the backplane connectors are the most cost-effective way to improve overall performance. However, most of the early market interest is in upgrading existing systems using the antiquated backplane so, again, SerDes technology must make up the difference.
A bit of both, it's very situation dependent in terms of # of channels, size of backplane, cost target of the system, etc.
Margin is margin. Our customers have often gone to selective layers of lower loss PCB materials to add margin. However no one wants to go to the most exotic materials. Meanwhile good design practice to reduce crosstalk and S12 discontinuities is way more important than anything else.