Techniques for Routing Differential Signals Print E-mail
Written by Charles Pfeil   
Monday, 01 July 2013 17:59

Matching trace lengths can overcome impedance discontinuities.

Fundamentally, a differential signal route path enables two complementary signals (N and P) to be sent from a driver to a receiver. These signals are out of phase 180°, and the receiver expects a particular voltage difference between the two signals (within a tolerance), which can be used for a number of purposes, including a switching event, data transmission and to provide a clock signal. If the differential signal is too distorted or the timing is significantly out of phase at the receiver, the expected voltage difference will not be recognizable and the circuit will fail.

Differential signals are used because, when properly configured and routed, they are less susceptible to noise effects and less likely to generate EMI compared to single-ended nets. The environment is well-suited for high frequencies and low voltages.

Some would say that the oft-employed conservative approach to routing differential signals is more about managing paranoia than managing signal integrity. Maybe paranoia is too strong a word here, but certainly when engineers venture into the unknown, unfamiliar potential problems are often addressed by over-constraining. Hopefully, the engineer and PCB layout designer will work together to reach an understanding of which methods are necessary and which are superfluous.

Routing concerns. There are several orders of concerns, and each order has potential consequences. Here is a summary of those concerns. Table 1 lists the potential consequences.

[Ed.: To enlarge the figure, right-click on it, then click View Image, then left-click on the figure.]

1st Order: A first order concern is one that, for all frequencies, will likely distort the differential signal sufficiently to cause a failure of the circuit. These concerns should be addressed in all cases and all data rates.
2nd Order: One that, for all frequencies, may distort the differential signal sufficiently to cause a failure of the circuit. These concerns should be analyzed to determine if they are significant in the context of the particular circuit, regardless of the data rates.
3rd Order: One that, for frequencies above a certain level (i.e., >5GB/s), may distort the differential signal sufficiently to cause a failure of the circuit. Once the data rate gets high enough for these concerns, the effects can be significant and must be properly managed.
4th Order: One that has a very small impact on signal integrity for all reasonable frequencies, and is highly unlikely to cause or contribute to a failure of the circuit. Frankly, these concerns should be ignored.

Techniques for Matching Trace Lengths

Here is an overview of several techniques for matching trace lengths. For greater discussion, seek out one of the many books on the subject.

Symmetrical pad entry. When a differential pair is routed to pins or fanout vias, the traces should converge as soon as possible and be of equal length from the pads to the convergence point (within the specified tolerance). Ideally this provides phase matching and the best impedance management.

Most EDA tools support a constraint such as “Distance to Convergence” to help guide the layout designer. Figures 1 to 3 illustrate examples of symmetrical pad entries.

Nonsymmetrical pad entry. When pins or fanout vias are blocked or located so that a symmetrical pad entry is not possible, it may be appropriate to add some length to the shortest trace as soon as there is room. This length can be added with a sawtooth pattern or some uncoupled tuning. Again, if the length difference is within the tolerance derived from the skew budget, the length adjustment is not necessary.

Sawtooth. Commonly used to add length, it minimizes the impedance discontinuity compared to uncoupled tuning because the minor adjustments do a better job of maintaining the coupling (Figures 4 and 5). The width (W) and height (H) of the sawtooth may vary; however, a common setting is:

W = 3X the trace width       H = 2X the spacing (S)

Note that the sawtooth method increases the length in a rather small increment. As shown in Figure 5, only the diagonal segments add length, and each of these diagonal segments only adds about 30% more trace length than the equivalent horizontal line.

Uncoupled tuning definition. When uncoupled tuning is appropriate, the spacing (S) can vary. A common setting for the spacing is twice (Figure 6) the normal trace-trace clearance with the intent to avoid same-net coupling.

Phase matching. When lengths of differential pair traces get out of tolerance from the source, it may be necessary to add length to the shorter trace. This length is also commonly added by sawtooth or uncoupled tuning. The topology looks the same as the Non-Symmetrical Pad Entry methods, but should be located near the location where the signal gets out of phase. If the logic family protocol requires compensation at corners to enable phase matching within tolerance, one method is to add a sawtooth at the inside corner.

Arcs. If routing of the signal requires arcs, the tuning added to accomplish phase matching should also have arcs.

Ultimately, the differential signal routing needs to fulfill the signal integrity and performance requirements and still maintain fabrication quality, as well as meet the cost budget. Most companies develop design methods over the years that have measured success. When new technologies are introduced, it is important to evaluate them in the context of known good practices, ideas and test results from others in industry.

Charles Pfeil is engineering director at Mentor Graphics, Systems Design Division;  This e-mail address is being protected from spambots. You need JavaScript enabled to view it . He was recently inducted into the PCB Design Hall of Fame.

Last Updated on Monday, 01 July 2013 23:52




CB Login



English French German Italian Portuguese Russian Spanish

Printed Circuit Design & Fab Magazine on Facebook