Serpentine Routing: Length Matching Done Correctly

When two signals leave a source at the same instant and must arrive at the same instant, their traces have to be the same electrical length. Real boards rarely allow that naturally, so the designer adds length to the shorter path. The usual way is a serpentine: a run of copper that doubles back on itself to consume distance without consuming area. Done well, a serpentine is invisible in the measurement. Done badly, it introduces crosstalk, changes the delay it was meant to fix and creates impedance discontinuities that show up as reflections.

Why Matching Is Needed at All

Signals travel along a trace at a speed set by the dielectric constant of the material, roughly 150 to 180 millimetres per nanosecond for a typical board material. A difference in length therefore becomes a difference in arrival time, and the amount of skew matters relative to the setup and hold window of the receiver. Slow parallel buses tolerate a few hundred picoseconds of skew; a fast serial link may tolerate a fraction of that.

That difference in arrival time is the delay skew, and the requirement is normally expressed as a length tolerance for a group of nets, and the group is defined by the signals that must arrive together. For a synchronous bus, the data lines must match the clock within a stated window. For a differential pair, the two halves of the pair must match each other within a much tighter tolerance, because any skew within the pair converts part of the differential signal into a common-mode component.

Where the Length Goes

A serpentine should be placed where the mismatch occurs. If one trace of a group is short because its path is direct and another is long because it had to detour around an obstacle, the extra length belongs near the point where the detour happened, not at the far end of the board. Putting all the tuning in one place concentrates the impedance discontinuities and makes the structure more likely to couple into itself.

For differential pairs, the two traces should be adjusted so that the pair remains symmetric. Where the pair must be tuned, the tuning should be applied to both traces so that their average length grows together, and the spacing between the two should remain constant through the tuned section. Separating the halves of a pair to insert a serpentine in one of them creates an unmatched segment whose impedance is not the differential impedance of the pair.

Amplitude, Spacing and Self-Coupling

The rule that distinguishes a good serpentine from a bad one is the relationship between the tuning amplitude and the spacing between its adjacent sections, and getting that ratio wrong is the most common mistake in length matching. The amplitude is the distance the trace deviates from the general direction of travel, measured perpendicular to that direction. The spacing is the centre-to-centre distance between two parallel sections of the same serpentine.

Where the spacing is small relative to the amplitude, the sections couple to each other. The coupling adds capacitance and, more importantly, changes the effective delay of the structure, so a serpentine built with tight spacing produces less delay than its physical length suggests. The rule of thumb that follows is to keep the spacing at least four times the trace width, and to prefer larger amplitude with wider spacing over a compact structure with dense turns.

The coupling between adjacent sections is the same mechanism that causes crosstalk between unrelated nets, and the spacing criteria are described in more detail in this article on the 3W rule for crosstalk control.

Corners and Impedance

A serpentine made of right angles introduces a local change in trace width at every corner, which appears as a small discontinuity. The accepted practice is to use 45 degree segments and, where the serpentine is large, to use rounded turns. The difference is small for a single corner but a serpentine may contain dozens of them, and the discontinuities accumulate along the line.

The total number of turns should also be considered from the manufacturing side. Long serpentines with many tight turns are harder to etch uniformly, and the copper in the turns is narrower than the copper in the straight sections after etching. The result is a trace whose impedance varies along its length, which is exactly what the designer was trying to avoid. The general routing rules that apply to angle and width consistency are set out in this discussion of escape routing and pad fanout.

Serpentines and the Return Path

A serpentine changes the shape of the signal path, not the shape of the return. The return current underneath still follows the trace, so the return path also doubles back, and the loop that the signal and return form is larger than it would be for a straight line. Where the serpentine runs over a continuous plane this is tolerable. Where it runs over a split plane or close to the board edge, the return has to divert and the resulting loop radiates.

This is the reason serpentines should not be placed near the edge of a board or across a plane boundary. If the tuning must occur in such a region, the plane underneath should be made continuous first, or the tuning should be moved. The same considerations govern ordinary high-speed routing, as described in this article on high frequency traces and data buses.

Verifying the Tuning

Length matching is verified in the layout tool, but the number the tool reports is the physical length. What matters is the electrical length, which depends on the effective dielectric constant along the route and on how much of the trace runs on an outer layer versus an inner one. A group of nets whose lengths match on the outer layer can be mismatched after one of them dips to an inner layer for a short distance, because the propagation velocity differs.

The practical approach is to keep the tuning groups on the same layer and the same dielectric environment, and to treat any layer transition as a change that must be accounted for. Where transitions are unavoidable, the group should be checked with an extraction tool rather than with the length report alone.

FAQ

How much amplitude should a serpentine have? Enough to achieve the required length with spacing at least four times the trace width between adjacent sections. Using lower amplitude and denser turns makes the sections couple to each other, which reduces the effective delay and turns the tuning into a source of crosstalk.

Should both traces of a differential pair be tuned? Yes, and the pair should remain symmetric through the tuned region. Adding a serpentine to only one trace changes the spacing between the halves and creates a segment whose impedance is not the controlled differential impedance of the pair.

Does a serpentine slow the signal down as much as its length suggests? Not always. If the sections couple to each other, the structure behaves partly as a distributed capacitor and the added delay is less than the physical length implies. This is why the measurement of a tuned channel sometimes disagrees with the length report.