1
\$\begingroup\$

I have 5 differential pairs (4-lane MIPI-DSI) going from left (SOM header) to right (FPC connector). As you can see on the right, for each lane, the signals end on opposite ends of the FPC connector. This is -unfortunately- a given, because the panel I want to drive just has this pin assignment.

To make the lengths of each pair match, I have to have the signals leave the SOM header in opposite directions, to make up for the length difference they have on the other end of the signal. As you can see some signals also have pointless vias, just to make sure each signal has exactly the same amount of vias.

Now my question is if in this form, does it still make sense to adhere to the spacing rules of differential pair routing. i.e. I chose 6 mil width, 5 mil spacing for 100 Ohm diff impedance with a specific stackup. But there is no actual location in any of these pairs where the N and P data would actually be the same at any given point in time. Isn't it just better to leave them apart, and what would it do to impedance?

All signals are about 7.7mm long and length matched within 6 mil.

I can't really move the SOM or connector further apart.

enter image description here

\$\endgroup\$
2
  • 1
    \$\begingroup\$ What kind of signals are these? What are the frequencies and/or edge rates? What is driving the length matching requirement? \$\endgroup\$
    – The Photon
    Jun 5, 2020 at 16:29
  • \$\begingroup\$ It is 4 lane MIPI-DSI, that is used for driving mobile displays mostly. The bit clock is 800Mhz. It should be length matched because the pixel data is spread across 4 data lanes. 1 pair is a clock. \$\endgroup\$
    – Genoil
    Jun 5, 2020 at 16:40

1 Answer 1

3
\$\begingroup\$

First, with 800 MHz bit periods, you can afford at least 10 mm, probably 20 mm length differences between pairs. From the perspective of getting the signal decoded correctly, you could also allow ~10 mm differences within pairs, however even small intra-pair length differences cause common mode currents, which lead to radiated signals, so it's better to keep the two lines within each pair well-matched.

Second, it's preferable to route your pairs as actual pairs, not just as separate lines with the same length. For example, you could do something like this (sorry, I used red for the layer you have in blue and blue for the layer you have in salmon)

enter image description here

The main point is that for as much of the trace length as possible, the two lines of the pair are actually routed directly next each other. Symmetry in the break-out from the chip or connector pads ensures (with a tiny bit of tweaking) length matching.

\$\endgroup\$
1
  • \$\begingroup\$ Your last remark is exactly what I was looking for. I’ll give it another go. \$\endgroup\$
    – Genoil
    Jun 5, 2020 at 17:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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