Talking with Dave from Aircell about this, since they are moving the PCIe traces farther apart from each other in order to reduce loss due to capacitive coupling between the positive and negative signals in the pair.
___ h ___ h h _____________
In the drawing above, the two traces are the same distance (h) apart as they are from the ground plane below. The EM field from one of the pairs in the differential signal couples to the other about as equally as its EM field couples to the ground plane below. The effective impedance of each of the signals in the pair is therefore influenced by the distance between the pair as well as the distance between the pair and ground. This is great for external signal (noise) rejection, but this extra capacitive coupling creates additional attenuation of the overal signal and reduces link margin. You effectively have a microstrip calculation as well as an edge-coupled signal calculation to perform.
___ 3h ___ h h __________________________
As the distance between the traces increases the interaction between them becomes less and less important. At some point (not sure what) they don’t interact with each other at all anymore and the driving force to calculating their impedance is only the width of the trace and the distance to the plane underneath; each signal in the pair is a classic microstrip. There is less loss in the transmission line since the signal in each of differential pairs is no longer fighting with its counterpart and you have only the capacitive coupling to ground and whatever series resistance the trace creates to worry about. As you move the traces in the differential pair farther apart, however, you lose the benefit the close coupling creates: external signals (noise) are now less likely to couple equally to both halves of the pair and your receiver can no longer subtract the common noise from the signal.
Characteristic Impedance of a Differential Pair¶
When you hear a driver or receiver call for “100 ohm differential impedance” it is looking for the total impedance to be 100 ohms. Each signal in the pair is characterized and then summed to create the differential impedance. (say 50 ohms to ground each). This is where it gets tricky when you have to take into account edge coupling effects when you start mashing the pair close to each other.
Basically you want your signals in a differential pair close together; external signals will couple equally to both of the signals in the pair and the receiver will be able to ignore this common mode noise, which is probably the biggest point of using differential signalling in the first place. The closer the signals are to each other, however, the more they will fight each other due to capacitive coupling and this leads to loss of link margin. Also, the effective impedance of the differential pair gets more difficult to calculate because the EM field of one signal couples to the signal in the other pair, and vice-versa.