Differential Pair Via Design Guide

Every differential pair has to change layers somewhere, and the transition is where most of the signal integrity problems in a high speed link are created. A differential pair via is not two independent vias; it is a coupled structure whose symmetry decides how much of the differential signal is converted into a common mode one. The rules that apply to a single via are the beginning of the story rather than the end of it.

What Makes a Via Pair Different

A single via is a short section of transmission line with a capacitance to the planes and an inductance along the barrel. A pair of vias adds a mutual capacitance and a mutual inductance between them.

The impedance of the transition depends on all four terms, so a pair of vias that are far apart and a pair that are close together behave differently even though each via is identical.

The coupling is the reason a differential via pair can be designed at all, and it is the reason the array cannot be laid out by drawing two vias and hoping for the best.

The same principles appear in the via clearance work, where the interaction between a via and its neighbours is treated as a design variable.

Symmetry and Mode Conversion

A differential signal is carried by two conductors with equal and opposite voltages. If the two paths through the transition are identical, the symmetry is preserved and no common mode signal is created.

Any difference between them, in length, in capacitance or in the position of a ground via, breaks the symmetry and converts part of the differential signal into a common mode one. The conversion is measured as mode conversion or as skew.

The conversion is easiest to see on a time domain measurement, where the two lines of the pair arrive at slightly different times and the difference shows as a widening of the eye. A spectrum analyser with a current probe shows the same effect as a peak at the data rate and at its harmonics.

Common mode current radiates, and it also reduces the eye height at the receiver. A pair with a poorly designed transition can pass a loss test and still fail an emission test.

The practical rule is to keep the two vias symmetrically placed with respect to the reference and to keep the ground vias symmetric as well, since an asymmetric ground via is as damaging as an asymmetric signal via.

Ground Vias and the Return Path

The return current of a differential pair flows in the reference plane beneath it. When the pair changes layers, the reference plane changes too, and the return current has to move from one plane to the other.

A ground via next to the signal vias provides that path. Without it the return current finds a long route through the plane and around the edge of the board, which increases the loop area and the inductance.

The ground via should be placed as close to the signal vias as the drill rules allow, and it should be symmetric with respect to the pair. Two ground vias, one on each side, give the best result and are common practice on a fast link.

The ground vias also need to be connected to the planes that the pair references on both layers, otherwise the path is interrupted at the via. A ground via that stops at the wrong plane is worse than none.

Stub and Layer Transitions

A via that passes through the whole board carries a stub below the layer where the trace leaves it. The stub is an open ended line that resonates at a frequency set by its length.

On a differential pair the stub affects both lines, and if the two stubs are the same length the effect is a reflection rather than a mode conversion. If they differ, which happens when the pair leaves on different layers, the effect includes skew.

Back drilling removes the stub and is the standard solution on a thick board. A blind via or a via in pad removes it by construction, at a higher cost and with a more complex process.

The options and their costs are compared in the via construction guide, and the choice is usually made from the loss budget rather than from the routing density.

Breakout Geometry

The breakout is the region where the traces separate from the pads of the component and converge on the vias. It is the point where the pair is most likely to lose its symmetry.

The two traces should have the same length and the same spacing into the vias, and any tuning of the pair should be done away from the breakout rather than inside it. A serpentine placed next to a via creates a discontinuity in the wrong place.

The via pad size should be the smallest that the drill rules allow, because a large pad adds capacitance and lowers the impedance of the transition. A pad that is several times the drill diameter is a common cause of a reflection.

The antipad, which is the clearance in the plane around the via, also sets the capacitance. A large antipad reduces the capacitance and raises the impedance, so the pad and the antipad have to be designed together.

Measurement and Simulation

A field solver is the usual tool, and it is used to compare two or three candidate geometries rather than to design from scratch. The model needs the stackup, the pad and antipad sizes, the drill and the material.

The measurement is made on a coupon with a through transition and a known pair of lengths. The result is a plot of insertion loss, return loss and mode conversion against frequency.

The mode conversion is the curve that is most often ignored and most often responsible for a problem, because a link can meet its loss budget and still radiate.

The coupon and the probe practice are described in the routing guide, and the coupon should be measured with the same launch as the product.

Practical Rules

Keep the pair symmetric through the transition, including the ground vias. Symmetry is the single most important rule and it costs nothing.

Add a ground via next to the signal vias, on both sides where the space allows, and connect it to every plane that the pair references.

Keep the via stub short, and back drill the transitions that are not. A stub that resonates inside the band is a reflection that no equaliser can remove.

Record the geometry with the board documentation, because the next revision will reuse it and a note is cheaper than a simulation.

FAQ

How far apart should the two vias of a pair be? Close enough that they stay coupled and far enough that the drill rules are met. The spacing is a design variable and should be simulated rather than guessed.

Does a ground via have to be on both sides? It should be where the space allows, because a symmetric pair of ground vias keeps the transition balanced. One ground via is better than none.

Why does mode conversion matter? Because the common mode component radiates and reduces the eye height. A link can pass a loss test and still fail an emission test because of it.