Fiber Weave Effect: Why Differential Pairs Drift

A differential pair is supposed to be two traces that see identical conditions, so that any disturbance affects both of them equally and cancels at the receiver. On a real board that assumption is not quite true, because the laminate underneath is not a uniform material. It is a woven glass cloth saturated with resin, and the two traces may be sitting over different parts of that weave.

This article explains what the fiber weave effect is, why it disturbs high speed links, and which remedies actually reduce it.

The effect is not new, but it became important when serial links moved into the range where a fraction of a picosecond of skew is a meaningful part of the unit interval.

The Weave Underneath The Trace

Prepreg and core are made from a glass cloth, woven from bundles of glass fibre in two directions, impregnated with resin and cured. The result is a material with a periodic structure: bundles of glass fibre, which have one dielectric constant, separated by windows of resin, which have a different and usually lower one. The period of that structure is set by the weave style and is not microscopic; it is comparable to the width of a high speed trace.

The consequence is that the effective dielectric constant under a trace depends on where the trace happens to sit. A trace running directly along a bundle of glass sees more glass than resin; a trace running between bundles sees the opposite. Over a long route the average evens out, but over the length of a differential pair the two traces can experience different averages.

Why The Pair Suffers

If one trace of a pair sits over more glass than its partner, its propagation delay is slightly different. That difference in delay is skew, and skew within a pair means the two signals no longer arrive together. Part of the differential signal is converted into a common mode component, which the receiver rejects and which radiates from the cable or the board, and the eye at the far end closes a little more.

The disturbance also shows up as an increase in insertion loss that varies with position. Data from simulations of a poorly chosen weave shows mode conversion varying strongly as the pair is shifted sideways across the cloth, with some positions much worse than others. Move the same pair to a different location on the same panel and the result changes, which is why the effect is so hard to reproduce in the lab.

The Conventional Remedies

The first remedy is to choose a cloth with a tighter, more uniform weave. Spread or flat glass styles reduce the difference in dielectric constant across the surface, so the average seen by each trace of a pair is closer to that seen by the other. They cost more, and they do not eliminate the effect entirely, but they reduce its magnitude considerably and they are the usual baseline for high speed work.

The second remedy is to angle the routing. Rotating the traces by a small angle relative to the weave means each trace crosses the same sequence of bundles and resin windows over its length, so the two averages converge. Angles of around ten to fifteen degrees are used in practice, and the same logic is behind rotating the panel itself during fabrication so that the weave sits at an angle to the routing. Conventions about routing geometry exist for reasons like this, and they are worth understanding rather than copying.

Matching The Pitch To The Weave

There is a third approach that attacks the cause directly. If the pitch of the differential pair, meaning the centre to centre distance between the two traces, is made equal to the period of the glass weave, then any sideways displacement leaves the pair in the same relative position with respect to the cloth. Both traces see the same mix of glass and resin, so no matter where the pair sits, the environment is shared.

The idea is attractive because it makes the result independent of placement, which removes the variability that makes the effect so difficult to manage. It requires knowing the weave period of the specific laminate being used, and it constrains the geometry of the pair, so it is only practical where the impedance target and the weave period happen to be compatible. Simulation data comparing a pair designed this way against one that ignores the weave shows a large difference in mode conversion, with the matched design staying well below the level that matters while the other varies by position.

What To Do In Practice

Three things are worth doing on any design where the effect is a concern. Ask the fabricator which glass styles are available, and specify a flat weave for the layers carrying the fastest differential pairs. Where the geometry allows, angle the affected routes so that both traces cross the weave in the same way. And keep the pair symmetric in every other respect, because weave induced skew is added to any skew the layout itself introduces.

Where the pair has to be length matched, remember that the tuning to match two nets is a separate exercise from matching the two halves of one pair, and that a serpentine added to fix a length error can itself create a place where the pair is asymmetric. gopcb works with customers on laminate selection and can advise on weave styles that suit a given differential impedance.

Measuring Whether It Matters

The direct measurement of the effect is mode conversion, and it is made with a four port network analyser on a differential structure. The quantity of interest is the transmission from the differential mode at the input to the common mode at the output, and a well designed pair keeps it far below the level that would trouble a receiver. Time domain equipment can be used instead to measure the skew between the two conductors directly.

Whatever is used, the coupon has to represent the product. A test structure that does not sit on the same laminate, with the same weave orientation and the same geometry as the pairs in the design, will not show the effect the board actually has. Agreeing what will be measured, and on what structure, is part of the stackup and CAM review rather than something to settle after the first build.

Where To Look Next

The weave is one of several effects that only appear once the design is treated as a transmission line. Its practical importance grows with data rate, with the length of the pair and with the number of pairs on the panel, because each of those increases the chance that a pair sits in an unfavourable position. Checking the laminate and the routing of the fastest pairs early is much cheaper than discovering the problem on a finished assembly, when the only remedies left are board respins.

FAQ

Does the fiber weave effect affect single ended traces? It can, as a small variation in impedance, but the serious problem is the skew and mode conversion it creates between the two halves of a differential pair.

Is a flat weave enough on its own? It reduces the effect substantially and is the usual first step. Where the margin is tight, angling the routing is worth adding.

Why does the result change between boards? Because the placement of the routing relative to the weave changes from panel to panel, so each board presents the traces with a slightly different average environment.