Back

@gsuberland Basically any time you have a connector, passive, BGA, whatever pad that's physically larger than the natural track width for your target impedance, the excess capacitance is going to make your impedance drop at the launch.

The usual fix is to void the plane under the connector to reduce the capacitance until it's matched again. For my typical many-layer high speed digital stackups with 75-100 μm dielectrics, I find a void the size of the pad is usually a good starting point. So I start with a sim of no cutout as a baseline, then I add a pad-sized cutout which is usually pretty close, then sometimes I tweak up or down if it's still not matched enough for my liking.

@gsuberland If you aren't simulating, my general experience is that a pad-sized cutout on a typical stackup will do more good than harm (i.e. it might not go far enough, or it might make the impedance a little too high, but the overall return loss will almost certainly be better than with no cutout at all) assuming that the pad is significantly wider than the track.

If you have thick dielectric like on e.g. OSHPark 4 layer, on a lot of smaller connectors you don't even need a void because the pad is only a little bit larger than (or matched with) your 50/100Ω track so it's already near perfect.

The thinner your dielectric gets, the narrower your controlled impedance tracks get and the greater the magnitude of the mismatch for a fixed-sized connector pad.

@gsuberland It also depends a lot on the specifics of the connector. Something like a Rosenberger 32K243-40ML5 will have much less inherent mismatch at the launch than a Samtec SMA-J-P-H-ST-EM1 just because the pad is so much smaller (in fact, with the Rosenberger on something like OSHPark you might have the opposite problem, you can't make the track big enough to maintain 50 ohm impedance in the confines of the connector)