4
\$\begingroup\$

I have been told that trombones are much better to use than sawtooths, even if sawtooths may be prefered in some certians scenarios.

Here I have added trombones to the differential pairs. If I had been using sawtooths, then the impedance would take damage becuase this extra "path" spares the impedance.

Let's summarize the pros and cones by using tromebones and sawtooths:

Sawtooths

  • Pros: The sawtooth can delay the signal at each corner, which makes the signal follow the same symmetrical speed
  • Cons: The sawtooth violates the signal integrity (impedance)

Trombone

  • Pros: The trombone spares the signal integrity (impedance)
  • Cons: The trombone make the signal not symmetrical in speed

Whether by using sawtooth or tromebone, the signals will arrive at the same time.

Question:

Which one is preferred when it comes to differential pairs?

This picture shows Giggabit Ethernet traces to an RJ45 connector. The impedance must be 100 ohms.

enter image description here

This picture shows HDMI traces to a HDMI connector. The impedance must be 100 ohms.

enter image description here

This picture shows USB OTG connection. The impedance must be 90 ohms.

enter image description here

This picture shows DDR connections. The impedance must be 100 ohms.

enter image description here

\$\endgroup\$
13
  • \$\begingroup\$ The reason of the use of differential pairs is to reduce emissions and susceptibility. \$\endgroup\$
    – Andy aka
    Commented Jun 2 at 13:37
  • \$\begingroup\$ @Andyaka Sure, the diff pair is more robust to noise. But the signals need to arrive at the same time. That can be achieved by using sawtoots or tromebones. But which one is prefered? \$\endgroup\$
    – euraad
    Commented Jun 2 at 13:39
  • 4
    \$\begingroup\$ Can you give an example of sawtooths? The problem is that when you are told something is better, it may be specific to some certain case, or a huge generalization. Sometimes, the correct answer is neither, because it may be better to not mess the impedance match with serpentine tracks, but to allow for mismatch in intra-pair length if it's within tolerance. Looks like your trombones bulge a lot so the impedance discontinuity may be more than you want. Also mention what signals those are and what are your specs for it - I have to assume Ethernet, but is that 100M or Gigabit or what? \$\endgroup\$
    – Justme
    Commented Jun 2 at 13:58
  • 2
    \$\begingroup\$ @euraad Yes, differential pairs should always have the same length. However, if you are within the allowed tolerance even without doing it, then it's better not to do it and cause impedance discontinuities by adding squiggly bits of wiring. \$\endgroup\$
    – Justme
    Commented Jun 2 at 15:14
  • 1
    \$\begingroup\$ Which of your examples are you calling "trombones" and which are you calling "sawtooths"? I have a good idea what you mean by a "trombone", but I haven't heard of a "sawtooth" design before, so it's hard for me to understand your question. Also when you say "spares the impedance", do you mean it preserves the impedance? \$\endgroup\$
    – The Photon
    Commented Jun 2 at 15:33

2 Answers 2

Reset to default
6
\$\begingroup\$

In your first three examples, the differential pairs do not appear to be particularly tightly coupled. The trace separation is at least 4x the trace width. This means that the impedance discontinuity due to the trombone is likely to be quite small. If you want to know how much, re-calculate your differential pair impedance with the same trace widths and height above ground, but with the trace separation increased to some very large number (10 mm, or whatever). Very likely for the geometries shown, it will still be 100 ohms +/- a few ohms.

The 4th example might show traces closely enough coupled for the trombone to adversely affect the characteristic impedance. In that case, one possible solution is to widen the track slightly in the trombone sections to increase the capacitance between track and ground, compensating for the reduced capacitance track-to-track. This is a bit fussy to do in CAD, but it is possible.

Working at 1 Gbps, the examples you showed would be adequate as-is for many applications. At 25 Gbps, you'd want to be using curves rather than 45-degree bends to form your trombones, and you'd want to use EM simulation (HFSS, for example) to verify your design (for reflection, transmission, and mode conversion) before manufacturing in any case, and you could use the results of that simulation to improve the design as needed.

\$\endgroup\$
8
  • \$\begingroup\$ So you're telling me that my routing for these specific trombones won't disturbe the computed impedance so much? My tolerance for the impedance is about 10% i guess. \$\endgroup\$
    – euraad
    Commented Jun 2 at 16:03
  • \$\begingroup\$ @euraad, probably not. If you want to know for sure, do the calculation. \$\endgroup\$
    – The Photon
    Commented Jun 2 at 16:04
  • \$\begingroup\$ Thank you. Can I just compute the impedance with a different space? For example, the space is 0.24mm for the DDR diff-pair and with that trombone, its 0.581 mm. \$\endgroup\$
    – euraad
    Commented Jun 2 at 16:05
  • \$\begingroup\$ @euraad, just compare the geometry you have, to the same geometry but with the track-to-track spacing increased to a large number. You could use 10x the track width or the actual largest spacing in your trombone structure or whatever. Or calculate the single-ended impedance for a single isolated track and double it. You'll get very close to the same result with any of these methods. \$\endgroup\$
    – The Photon
    Commented Jun 2 at 16:08
  • 1
    \$\begingroup\$ Step 1: Find the characteristic impedance of your differential traces if you increase the trace separation by "a lot". Step 2: Compare it to the characteristic impedance you had with the original geometry. Step 3: Ask yourself, "is this difference big enough to worry about?". There is no averaging to be done here. \$\endgroup\$
    – The Photon
    Commented Jun 2 at 16:28
4
\$\begingroup\$

Which one is better depends on the application and. They are both sort of bad anyway. If you want to maximize the total signal integrity for best performance (for whatever metric that is), you don't only minimize length difference or impedance mismatch - sometimes a perfectly valid option to maximize signal integrity and minimize impedance mismatch and reflections is to allow intra-pair length mismatch.

Both techniques cause an impedance mismatch, the other causes a short mismatch due to a single bulge, and the other causes a long mismatch due to longer length of sawtooths needed. In the end the sawtooths are just many small bulges in series.

So if you have calculated an impedance for a given trace with and separation, any deviations to that, whether or not sawtooths or trombone bulges, means that you are deviating from the impedance. This can of course be compensated by calculating or simulating the effect and using better parameters during the sawtooth or bulges.

The giveaway is, why not any other technique like smooth tracks to begin with with specified circular turns with specified turn diameters for the arcs to compensate for length differences? It would use more PCB area and might not be able to compensate the intra-pair difference right where it happens so it would take longer for the signal to travel with mismatch until the delay is equal.

\$\endgroup\$

Not the answer you're looking for? Browse other questions tagged or ask your own question.