Differential Pair Matching And Phase Tuning

A differential pair carries information in the difference between two conductors, and the receiver recovers it by comparing them. Anything that delays one member of the pair relative to the other reduces that difference before the comparison is made, and the loss is not recoverable at the far end. Matching the two lines is therefore part of the signal path rather than a cosmetic layout preference.

This article covers what has to be matched, why intra-pair skew matters more than total length, how phase tuning is actually performed, and the layout habits that quietly undo the matching work.

What Is Being Matched

Two quantities are being kept equal along the pair: the electrical length, which sets the arrival time of the two signals, and the impedance environment, which sets the shape of the waveform on each line. Matching the length without matching the environment produces a pair whose two signals arrive together but with different shapes, and the common mode component that results is what radiates and what stresses the receiver’s input range.

Electrical length depends on the physical length and on the effective dielectric constant, and the second of these is not uniform across a board. A trace routed over a region where the reference plane is closer will be slower per unit length than one routed over a thicker dielectric, so equal physical lengths are not automatically equal electrical lengths.

Intra-Pair Skew And Why It Matters

Skew within a pair converts part of the differential signal into common mode. Common mode current has no return path on the other conductor, so it flows on the ground plane, radiates and produces emissions that a well matched pair would not produce. The conversion also consumes part of the timing budget, because the eye closes as the two edges separate.

The tolerance is usually stated as a fraction of the unit interval. A common specification is that intra-pair skew stays below five percent of the bit period, which for a ten gigabit link corresponds to about five picoseconds and therefore a few tens of thousandths of an inch of length difference on a typical laminate. That is a small number, and it explains why the matching has to be maintained through every transition.

Where Skew Actually Appears

Most skew is introduced at the ends of the route rather than in the middle. A package pinout where the two signals leave on different sides, a connector whose pin lengths differ, a via transition that one member of the pair takes and the other does not: all create an electrical length difference that no amount of serpentine tuning in the middle can fully compensate.

That is why the pair should be tuned after the transitions have been fixed, and why the tuning should be placed near the source of the mismatch when possible. Tuning in the middle places the compensation far from the discontinuity, and although the total length becomes equal, the two signals still arrive with different histories at the receiver.

How Phase Tuning Is Done

Tuning adds length to the shorter line by replacing part of it with a meander. The classic mistake is to make the meander tight: adjacent segments couple to each other, the impedance of the meandered section falls, and the added length does not behave as a delay line. Keeping the spacing between meander segments at several trace widths, and making the amplitude of the excursion small, preserves the impedance.

For a differential pair the tuning is done on both lines as a unit rather than on one line alone, because separating them also changes the intra-pair spacing and therefore the differential impedance. Where one line must be lengthened, the preferred method is a symmetric meander that adds length to both while introducing the required delay difference through a compensating shape.

Intra-Pair Spacing And Its Consequences

Reducing the spacing between the two traces raises the coupling and lowers the differential impedance, which is useful for controlling impedance in a thin stackup and harmful where the pair runs beside other traces. Tight spacing also increases the effect of any asymmetry, because the coupling is stronger and a small length difference matters more.

The standard compromise is to make the intra-pair spacing narrow enough to meet the impedance target and the spacing to neighbouring pairs at least three to five times the trace width. Where a pair must pass close to another, the separation needed to control crosstalk usually dominates the decision, and the impedance target has to be met by adjusting the geometry instead of the gap.

Layout Habits That Break The Match

Number one is the asymmetric via. When a pair changes layers, both members should change at the same place, with the same via style and the same return path. A pair that transitions through two vias at different distances from the ground stitching has an impedance and a delay difference at that point that shows up directly as common mode.

Number two is the right angle bend. Routing one member of a pair around the outside of a corner makes it physically longer than the other by the width of the corner. Using two forty five degree segments, and keeping the pair’s geometry symmetric, removes the problem; the behaviour of right angle routing on differential traces is a well documented consequence rather than a subtlety.

Tuning And The Rest Of The Bus

Matching within a pair is one requirement; matching between pairs on a bus is another. A bus needs its pairs to arrive within a defined window of each other as well as each pair being internally matched, and the two constraints compete for the same routing space. Reserving the tuning areas during placement, before the routing starts, is what prevents a matched design from becoming an unmatched one during the final cleanup.

When the tuning is complete, the report should show intra-pair skew and pair-to-pair skew separately, because a single number hides which constraint is at risk. The length matching techniques used to satisfy both are the same, but the measurement and the limits are not.

Process Control and Verification

Reviewing the design before the data is released is far cheaper than correcting it after the panel is in the tank, because every step downstream inherits the decision made at the front end. Where a value sits close to a process limit, the drawing should say so, since the shop can then open the process window rather than working to a nominal figure that carries no tolerance.

The measurements that matter are the repeatable ones: conductor width and spacing, annular ring, finished hole size, plating thickness and surface finish are all verifiable on a coupon that travels with the panel. Documenting the assumption is part of the design work, and a short note on the drawing prevents a question that would otherwise arrive a day later and cost a day of schedule.

FAQ

Does intra-pair skew matter if the link has equalisation? Yes. Equalisation corrects for loss and for channel delay, but skew inside a pair occurs before the two signals are compared, and part of the resulting common mode is simply lost from the differential signal.

Should the two traces be measured to their centres or their edges? To a consistent reference, and the same one for both. Design tools normally report centre-line length, which is adequate provided the trace widths are equal; unequal widths make the physical length comparison misleading.

Is a delay-matched pair enough for a long link? Only if the impedance profile is also matched. Length equality with differing impedance along the two lines produces reflection differences that appear as skew at the far end even though the nominal lengths agree.