Fiber Weave Skew in Differential Pair Routing

A differential pair is supposed to arrive at the receiver with both traces carrying the same signal at the same time. On a real board the two traces often see slightly different dielectric material, because the laminate is not homogeneous. The fiber weave of woven glass cloth creates regions of higher and lower resin content, and a trace that happens to run along a bundle of glass sees a different effective permittivity than its partner.

Why Woven Glass Is Not Homogeneous

FR-4 and most other laminates are made from woven glass fabric impregnated with resin. The glass has a dielectric constant near 6 and the resin near 3.5, so the local average depends on how much of each material lies under the trace. Above a glass bundle the dielectric constant is high and the signal travels more slowly; between bundles, where resin dominates, the constant is lower and the signal travels faster.

The scale of the pattern is what makes it a problem for high speed design. A typical weave has a pitch of a few hundred microns, which is the same order as the width and spacing of a differential pair on a dense board. A pair routed horizontally across a weave sees the full variation, while one routed diagonally crosses bundles at an angle and averages the effect out.

How Skew Appears in a Differential Pair

Skew is the difference in propagation delay between the two traces. If one trace runs along a glass bundle and the other runs between two bundles over a distance of several centimetres, the delay difference can reach several picoseconds, and at 10 or 25 gigabits per second that is enough to convert part of the differential signal into a common mode component and to close the eye at the receiver.

The effect is measured in picoseconds per inch, and the number depends on the laminate and the routing direction. Published figures for standard weave materials are in the range of 5 to 15 picoseconds per inch for a pair routed parallel to the weave, and much lower for a pair that crosses the weave at an angle. This is why the routing direction is a design decision rather than a matter of convenience.

Routing Angles and Their Trade-offs

The standard remedy is to route high speed pairs at an angle to the weave, typically 10 to 15 degrees off the board axis. The traces then cross both glass-rich and resin-rich regions repeatedly, so the two members of a pair accumulate nearly the same average delay. The penalty is a less tidy layout, a small increase in board area, and a need to keep the angle consistent between the two traces of the pair.

Some designs go further and route the entire high speed interface at the same angle, which keeps the pair geometry consistent and simplifies length matching. Where the layout is dominated by a fine pitch ball grid array with an axis-aligned escape pattern, a global rotation is impractical, and the angle is applied only to the long parallel segments where the skew would accumulate.

Choosing a Laminate for Skew Control

Several laminate families reduce the problem at the material level. Spread glass fabrics flatten the bundles and distribute the resin more evenly, which reduces the local permittivity variation. Some products use a thinner glass cloth with more plies, and others replace the weave entirely with a non-woven or particle filled construction. Each approach reduces skew at a cost in price or in mechanical properties.

The choice should follow the data rate and the trace length. A 3 gigabit per second interface with short traces will not notice the effect, while a 25 gigabit per second channel with 20 centimetres of parallel routing will. Where the laminate cannot be changed, the routing angle and the pair geometry are the available levers, and simulation with an extraction that includes the weave is the only way to confirm that the budget is met.

Simulation and Verification

Modelling weave skew requires a field solver that can represent the periodic structure of the laminate. A simple homogeneous model will predict zero skew, which is exactly the answer that has caused the problem to be missed on many projects. Where a full model is not available, published skew data for the material and the routing angle can be used to bound the effect, and the budget can be allocated accordingly.

Measurement then confirms the design. A time domain reflectometer or a vector network analyser with a suitable fixture can measure the differential to common mode conversion of a test structure, which is the signature of skew. Test structures should be routed at the same angle and length as the production traces, otherwise the measurement says nothing about the real channel.

Practical Design Rules

Keep high speed differential pairs short where possible, and route the long segments at an angle to the weave rather than parallel to it. Keep both traces of a pair on the same layer, since a change of layer introduces its own skew and the two effects add. Avoid routing one trace of a pair along a glass bundle while the other runs between bundles, which is what happens when a pair is aligned with the axis of the cloth.

Match lengths after the fact only where the skew is predictable. Serpentine tuning adds delay to one trace but it also changes the coupling between the two, and it can convert differential energy into common mode on its own. Where the pair must be tuned, the compensating section should be short and the spacing should be maintained. The relevant high frequency routing and mixed signal considerations apply to the same trace geometry.

Where the Effect Is Small Enough to Ignore

Not every design needs a weave strategy. A pair carrying a slower interface, or one whose total length is a few centimetres, will accumulate only a fraction of a picosecond of skew, which is well inside the timing budget. The decision should be based on the bit period instead of on the data rate alone: skew that is a small fraction of the unit interval is not a problem.

The same reasoning applies to the frequency content of the signal. A pair that carries a low frequency clock or a slow serial link has a long unit interval, and the delay difference from the weave is irrelevant. Where a design mixes slow and fast interfaces on the same layer, the angle strategy can be applied only to the fast ones, which keeps the layout workable.

FAQ

Can skew be corrected with length matching? Not reliably, because weave skew is distributed along the trace and the compensating section introduces its own coupling change. Prevention through topology and material is the better route.

Does stripline suffer the same problem? Yes. The traces are surrounded by laminate that contains the same weave, so the effect is present, although the geometry and the distance to the reference planes change the magnitude.

Which laminates reduce weave skew? Spread glass and non-woven constructions reduce the local permittivity variation and therefore the skew. The supplier data and a test structure on the panel are the best evidence.