Sunking PCB

Impedance Control: Where the Deviation Actually Comes From

Design Guide2026-10-16Sunking Circuits Technical Team
Impedance Control: Where the Deviation Actually Comes From

The conclusion first: an impedance tolerance is not a number produced by one process step. It is the sum of several independent deviation chains. Trace width, dielectric thickness, copper thickness and the spread of the dielectric material properties are each fixed at a different point in the manufacturing sequence. The four chains are not synchronised: they can point the same way and add up, or run against each other and partly cancel out. Anyone who reflexively blames an out-of-tolerance reading on etching will keep adjusting a parameter that is not the dominant cause, while the number in the report stays put.

1. Why a tolerance band alone is useless for diagnosis

An impedance requirement in the data package states a target value and an allowed band. Both are requirements on the end result. Neither says anything about the path by which that result is produced. Yet that path is the only information you can act on once a reading falls outside the band.

Technically, the impedance of a trace is set by the conductor geometry, its distance to the reference plane, and the dielectric properties of the material in between. In manufacturing each of these is a separate process outcome with its own spread:

  • Trace width is formed during imaging and etching.
  • The distance to the reference plane is formed during lamination, together with resin flow.
  • Copper thickness is formed during plating and depends on distribution across the panel.
  • The dielectric properties arrive with the material, before it enters the line at all.

Four quantities, four process stages, four owners. So the first question after an out-of-tolerance report is not "how far off is it" but "which of the four chains moved". How tolerance classes are formed and selected in the first place is covered in impedance control tolerance classes; this article assumes the class is already agreed and asks only where the deviation came from.

2. Chain 1 — trace width: governed by etching

What sits in the dataset as a line with a width leaves the line as a conductor with a cross-sectional profile. Etching does not only remove copper from above; it also attacks the sidewalls. The result is usually not a rectangle but a shape that narrows towards the top. What matters for impedance is the effective cross-section, not the number in CAD.

Two properties make this chain distinctive:

Its magnitude follows the starting copper weight. The more copper has to be removed, the longer the etchant acts on the sidewalls, and the more the final width departs from the drawn width. That is why the width chain is structurally the more visible one on heavier copper. How etch compensation is applied on heavy copper is covered in heavy copper etch compensation review.

It is the only one of the four chains that is systematically pre-compensated. Because the sidewall attack has a known direction, the starting width can be held back in the production dataset. This is common practice rather than special treatment — but it means the fab changes the width in the data. So anyone interpreting a deviation has to know first whether compensation was already included. Without that, a width measurement supports no conclusion about process stability.

What the designer can do here: the width chain offers the most design latitude of the four. The same absolute sidewall loss has a smaller relative effect on a wider trace than on a narrow one. That is a directional statement, not a formula; what width is actually available in a given build is decided by the stackup.

3. Chain 2 — dielectric thickness: governed by lamination

The distance between conductor and reference plane is a target dimension in the stackup table. It is actually created during lamination. Between the two sits resin flow: under heat and pressure, resin is displaced out of the prepreg, fills the gaps between copper features, and in doing so co-determines how thick the remaining insulation layer ends up.

This produces a property that distinguishes it from the width chain: it depends on the copper distribution of the neighbouring layers. An area carrying a lot of copper leaves the resin less room than an open area. The resulting thickness can therefore vary across one and the same panel without anything in the press cycle having gone wrong. For impedance this means the same trace can measure differently at two places on the same board.

The second property is material availability. A stackup is not assembled from arbitrary thicknesses but from the core and prepreg grades that can actually be sourced. Specifying a target that would only be reachable through an unusual combination moves the problem into procurement — see core and prepreg availability constraints in stackup.

What the designer can do here: the thickness chain can barely be narrowed in design, but it can be redistributed. A build in which the impedance-critical layer works against a thick, evenly populated reference plane puts this chain in a calmer region than a build where the same layer faces a densely patterned neighbour. That is a decision about the stackup, not about the trace.

4. Chain 3 — copper thickness: governed by plating

The copper thickness of the finished conductor combines the base material portion and the electroplated portion. The plated portion is distribution-dependent: deposition across a panel is not perfectly uniform but follows the current distribution, which in turn depends on geometry and copper coverage.

Three points matter for attribution:

  • It affects outer layers differently from inner layers. Inner layers essentially carry the base copper; outer layers also receive the plated build-up. An impedance structure on an outer layer therefore has one more deviation source than the same structure inside.
  • It is position-dependent, like the thickness chain — but with a different physical cause and a different pattern across the panel.
  • It acts on the width chain at the same time. More plated copper means more material to remove during imaging, hence more sidewall loss. The two chains are not fully independent at this point, and this is exactly where single-parameter troubleshooting goes astray.

What the designer can do here: almost nothing directly. Whether an impedance-critical trace runs on an inner or an outer layer is the most effective lever — and that too is a stackup decision, not a routing decision.

5. Chain 4 — material spread: before the line

The dielectric properties of the base material are not process outcomes. They come with the material and vary within what the supplier states. This chain therefore has a peculiarity: it cannot be narrowed in manufacturing, only selected.

Two points are decisive for diagnosis:

  • Dielectric properties are frequency-dependent. A material figure applies to the conditions under which it was determined. If the trace operates in a different range than the one the figure covers, a gap opens between calculation and measurement that belongs to no process stage at all. Selection criteria are covered in base material selection criteria and for high frequency in high frequency PCB material selection.
  • In mixed builds this chain is not uniform across the board. When different materials are combined in one stackup, each layer brings its own spread.

What the designer can do here: the material chain is the only one where the decision lies entirely before manufacturing. If the required tolerance is tighter than the chosen material allows under the intended operating condition, the problem cannot be handed to the fab to solve.

6. Which chain can be narrowed, which can only be redistributed

The four chains respond differently to intervention. This distinction is the practical core of the topic:

Deviation chainGoverning stageNarrowable in designRedistributable via stackup
Trace widthImaging and etchingyes, through chosen widthindirectly
Dielectric thicknessLamination and resin flownoyes, substantially
Copper thicknessPlatingnoyes, via inner/outer layer choice
Material spreadMaterial selection before the linenono, selection only

That yields a sequence for working an out-of-tolerance case that runs against the usual reflex:

  1. Check whether the structure sits in a robust region at all. A target impedance reachable only through a narrow trace against a thin insulation layer concentrates three chains at their most sensitive point. No process improvement will make such a build calm.
  2. Check whether the critical layer is placed correctly. Inner or outer decides whether the plating chain participates; what neighbouring copper coverage it faces decides how wide the thickness chain is.
  3. Check whether the material permits the required tolerance under the intended operating condition.
  4. Only then look at the process side, and there first establish whether the width chain compensation was applied correctly.

The first three steps are design decisions. Only the fourth is a manufacturing question. The common route — report the deviation, adjust etching, measure again — starts at step four and leaves the three more effective points untouched.

Where the build interacts with layer registration is covered in multilayer lamination registration control. Which impedance structures SKPCB carries as a product category is described on impedance control PCB; general manufacturing scope is under capabilities and the material side under materials.

7. What a measurement report must state to support attribution

A report containing only target value, measured value and a verdict allows no attribution to any of the four chains. It says something deviates, not what it deviates from. For a report to be usable in root-cause work, common practice requires at minimum:

  • Which structure was measured. A result from a test structure transfers to a functional trace only if the structure shares its layer, its reference plane situation and its neighbouring copper coverage. How test structures are laid out and verified is covered in impedance test coupon verification.
  • Where on the panel it was measured. Because the thickness and copper chains are position-dependent, a measurement without a location cannot be interpreted repeatably.
  • Which layer is involved, and whether it is an inner or an outer layer.
  • Whether width compensation was present in the manufactured dataset.
  • Which material and which stackup it was based on, particularly in mixed builds.
  • Under what conditions it was measured — otherwise calculation and measurement are not comparable at all.

If any of these is missing, the analysis ends at a hypothesis. That is the real reason impedance topics often circle through several rounds: not because the cause is hard to find, but because the report does not contain the information that would bound it.

8. Which inputs must come from design

For an impedance requirement to be attributable on the manufacturing side, the following must come from design rather than be guessed at in the fab:

  • Which nets are impedance-controlled — named, not inferred from geometry.
  • Which structure is meant and which reference plane it works against.
  • Which tolerance class applies and whether it applies equally to all named nets.
  • Under which operating condition the requirement holds, because the material chain depends on it.
  • What takes precedence when target impedance, trace width and build height cannot all be satisfied at once. Manufacturing cannot make that call; it is a design priority.

The last point is the one most often missing in practice, and the one that generates the most back-and-forth. Further notes on the quality side are under quality.

FAQ

Is etching the most common cause of impedance deviation?

It is the most visible one, because it can be measured directly on the conductor. Whether it dominates in a given case depends on the build: heavy copper with narrow traces raises the weight of the width chain, while thin insulation against densely patterned neighbours raises the thickness chain. There is no universal ranking, which is precisely why attribution needs the declarations in section 7.

Why do two boards from the same lot measure differently?

Because the thickness and copper chains are position-dependent. Resin flow follows surrounding copper coverage, plating deposition follows current distribution across the panel. Both vary within a single panel without any process stage having run outside its window.

Can a tighter tolerance simply be ordered?

A tolerance is not a setting that can be tightened independently of the build. It is the outcome of the four chains. If three of them are already held in a sensitive region by the stackup, the requirement cannot be met by adding effort in the fab — the build has to change.

What changes with HDI builds?

In sequential builds the same layer passes through more than one lamination and plating cycle. That does not automatically mean larger deviation, but it does mean the thickness and copper chains contribute more than once. Attribution must therefore establish in which cycle the affected layer was formed.

Can dielectric thickness be pre-compensated the way trace width is?

Not in the same way. Width compensation works because sidewall loss has a known direction. Thickness deviation depends on local copper coverage and has no uniform sign across the panel. It is addressed through the stackup choice, not through a held-back value.

Who decides which chain to investigate first?

The order follows from the build, not from a routine. Setting it requires the design inputs from section 8 — without them, any investigation starts with the chain that is easiest to measure, which is not the same as the most effective one.

Need Professional PCB Manufacturing?

Founded in 2009, we provide one-stop PCB solutions from prototype to mass production.