Backplanes hit a wall at 56G

-January 30, 2015

Something has to change to enable the next generation of fast computer and communications boards, but just what it will be is the subject of hot debate. A senior engineer from Rambus Inc put a spotlight on the issues in a paper at DesignCon.

The problem hits at the 56 Gbit/second speeds needed to drive systems that cost-effectively deliver 400 Gbit/second Ethernet. Big data centers and carrier networks are hungry for the fast interfaces now in development. But engineers are finding they just can’t drive signals at 56G across boards with two connectors that have traces typically up to 40 inches long.

“That’s out of the picture for now, something has to change,” said Wendem Beyene (below), a senior principal engineer at Rambus in a paper nominated for a DesignCon award.

Beyene showed results of research sending 56G signals between chips and modules and even between boards at distances up to 20 inches. The work between chips and modules used conventional NRZ signaling and four-tap FFE equalization.

As the lengths reached 20 inches, Rambus researchers had to either adopt PAM-4 signaling or five-tap DFE equalization, both requiring more complex and expensive chips. All the boards used relatively new Megtron-6 HLVP board materials with foil finish back drilled vias and the latest low-crosstalk connectors available.

Beyene said he holds out hope for a new generation of low-loss board materials that could emerge for the first time at a May event that often draws top suppliers from Japan.

There are different opinions. We worked hard to get Megtron-6 into the production flow, so some people are more interested in the option of adding re-timers. We are still trying to get information about the price differences at low and high volumes.

Companies are understandably shy about shifting materials. The 28 Gbit/s boards made for today’s systems were some of the first to use Megtron-6. The prior 10 Gbit/s generation was among the first to go off mainstream FR4 boards, adopting Nelco 4000.

“My sense is we may need new materials again, but many people don’t see it that way,” said Beyene in discussion after his talk.

Another engineer suggested assumptions about crosstalk are a wild card in the debate over NRZ and PAM-4 signaling. NRZ has been the work horse signaling technology for many years. Experts have been saying for at least a year that NRZ will not be able to stretch to use in 56G backplanes.

Rambus found boards driving 56G signals 20 inches across one connector needed either PAM-4 or five-tap DFEs.

This article was originally posted on EE Times.

Also see:

For more from DesignCon -- the premier event for the chip, board, and system design community -- click here.

Loading comments...

Write a Comment

To comment please Log In