2.2 ohms isn't enough to do anything, for better or worse, really. I'd use 0 ohm jumpers, just in case the above (FB recommendation) proves useful?

Oh, broadside coupled I should say.

Outer layer ground pour isn't very useful, for the most part. The edge coupling from traces/pads is small, maybe 20% even for very thin traces and small ground spacing.

Which means a few things:

- Microstrip traces can be relatively close together without much danger of coupling between them (assuming nice large signals and wide thresholds, like LVCMOS). Convenient for dense parallel buses.

- Edge coupled differential pairs aren't really so differential (e.g. LVDS, PCIe, etc.) and the normal or common mode is more significant*.

- Also why same-layer edge-coupled differential has a relatively high impedance, even for very wide traces and minimum gap.

Indeed, suppose we let "trace width" go to infinity: now it takes infinite time for waves near the gap to reach the edges, so what happens at the edges, doesn't matter. Next suppose we short-circuit the edges -- not that a short circuit around an infinite perimeter is all that meaningful, anyways. Finally, suppose we bring the edges back in from infinity, this time shorted together -- we need to define some geometry for that short, as it has a finite size now; and also waves can propagate to, and reflect off, it, so we'll have some lower cutoff frequency. We've transformed differential stripline into slotline waveguide. And, yeah, the characteristic impedance for this geometry is still relatively high -- as can be seen from this sequence of transformations, which has little effect on the impedance of the propagating mode.

*Not to say there's no advantage to differential routing -- the point is to have both traces encounter the same disturbances, at the same distance along each route. So the induced noise is largely common mode, easily rejected by the receiver.

*Also, in terms of significance, it's not all that common anyway, to terminate the common mode -- it tends to work out that transmitters have a close enough impedance. Usually an LVDS transmitter is something like, a source follower paired with a current sink (and then a pair of those to realize a full differential H-bridge like thing), and the source follower is what sets the CM voltage and source resistance. The output isn't quite so constant-current as they might lead you to believe, which is fortunate because a true current source wouldn't terminate signals at all, and quite high voltages could build up in the common mode, eventually exceeding the receiver or transmitter CM voltage range, corrupting the differential signal.

Anyway, it can be helpful to fill in surface ground, when very low impedances are necessary (e.g., switchers), or extra shielding is necessary (e.g. internal traces, helping out around vias, etc.), or just where that tiny edge (10-20% improvement) is actually worthwhile (precision RF something?).

Bringing the grounds up closer to the signals, rather than being just behind them, also grabs that little bit more electric field -- so traces couple less to ambient fields, reducing radiation and susceptibility by that amount. Again, not a big improvement, but if you're in a situation where every little bit counts, it's something.

If you happen to have buried vias in a design, of course pouring copper above and below them, fully shields them -- so you still have the hole in the middle planes, but with stitching vias adjacent, and pours over top and bottom, you can patch over that very well.

Tim