Don’t miss our first ever, one-day boot camp on Differential Pair Design. Oct 4, 2010, Santa Clara, CA. Check www.beTheSignal.com for details.
So you think that by designing a differential pair transmission line as tightly coupled, you will reduce the channel to channel cross talk? In principle, this is exactly correct. But, by how much will tightly coupling a pair reduced cross talk? Is this really a strong motivation for tightly coupled differential pairs?
Of course, the most common answer to all signal integrity questions is “…it depends” and the only way to answer “it depends” questions is by putting in the numbers. When dealing with uniform differential pairs such as microstrips or striplines in circuit boards, the right tool to use to analyze channel to channel cross talk is a 2-D field solver.
When you put in the numbers, you find that for microstrip, in which far end cross talk can be a huge issue, using tightly coupled differential pairs gains you about 5 dB lower cross talk than using uncoupled pairs. Tightly coupled microstrips is a very good idea. Of course, if you are worried about far end cross talk, don’t use microstrip, route your traces in stripline.
In stripline, the near end cross talk is reduced by less than 1 dB by going to tightly coupled traces. While there are important reasons to consider tightly coupled differential pairs in uniform transmission lines, reducing channel to channel cross talk is not a strong driver.
Except when the interconnect paths are not uniform transmission lines. If the return path is screwed up, as in going through vias, or using flex or in a connector or an IC package, using tightly coupled differential pairs, will have a significant impact in reducing channel to channel cross talk.
Estimating the cross talk in these geometries is tough and often requires a 3D field solver. I recently worked with Bill Martins and Madhavan Swaminathan to explore channel to channel cross talk in via transitions using SPHINX, their tool from E-System Design.
The problem we analyzed is the transition of two widely spaced differential channels passing through a power and ground plane cavity and back to the surface again. The figure to the left shows the path of one channel. The second channel would be adjacent to it.
We looked at two configurations of vias: each of the four vias in the first transition on 25 mil centers and then on 100 mil centers. We would expect that the tighter the coupling between the two vias, in each channel, the lower the channel to channel cross talk.
The geometry was set up and the 8 port S-parameters were simulated up to 20 GHz using SPHINX. I then used Agilent’s PLTS to re-display the simulated, differential responses in the time domain.
The differential TDR response shows the location of the via transitions as high impedance peaks, compared to the uniform lines with no via transitions.
In addition, in the bottom traces, you can see the near end channel to channel cross talk of about 1% occurring at the via transitions, and reduced noise when the vias are closely spaced. When the two vias that make up each differential pair are spaced on 100 mil centers, they inject quite a bit of differential noise in the power and ground plane cavity and we see this as residual noise picked up between the two channels long after the edge has passed by.
Whenever you have an interconnect with a screwed up return path, it’s important to first engineer this region as short as possible, and second to use as tight a coupling in the differential pair as is practical. These two features will minimize the discontinuity of the region and enable the highest bandwidth possible.
To learn more about this topic and others related to differential pair design, check out my Differential Pair Boot Camp, presented in partnership with PCBDesign007, Oct 4, 2010 in Santa Clara. You can learn more about the signal integrity training programs we offer by visiting our web site: www.beTheSignal.com.