
How to Design for HDI: Microvia Stacking and Any-Layer
What each BGA pitch band demands in via type, line width and stack-up, from 1.0 mm down to 0.35 mm.
Read article
Batch-to-batch impedance drift is the root-sum-square of four tolerances, and the two largest are not the ones designers worry about. The arithmetic below is why ±10% is an industry convention rather than a standard, and why ±5% is a process change rather than a tighter promise.
No IPC clause sets a default impedance tolerance. ±10% is a commercial convention: the number most fabricators quote and most designs accept, because it is what a standard process produces without extra control. A drawing that attributes it to a standard is citing something that does not exist.
On a 50 Ω microstrip the budget runs ±1.4% on Dk, ±10% on dielectric thickness, ±8% on etched width and ±10% on finished copper. Combined by root-sum-square that is about ±7.1%; if every term lands at its worst case at once, ±12%. That gap between 7.1 and 12 is the reason a coupon average can pass while an individual board fails.
Impedance is set by the dielectric constant, the dielectric thickness, the trace width and the copper thickness. Each has its own tolerance, each maps to impedance through its own sensitivity, and the sensitivities are not equal — which is why the same ±10% on two different parameters produces very different drift.
For a microstrip near w/h of 2, a 1% change in dielectric thickness moves impedance about +0.5%, a 1% change in Dk moves it about −0.5%, a 1% change in width moves it about −0.6%, and a 1% change in copper thickness moves it about −0.15%. Those four numbers are what turn a tolerance table into an impedance budget.
| Source | Typical tolerance | Sensitivity | Impedance effect |
|---|---|---|---|
| Laminate Dk, lot to lot | ±0.05 on 3.5 = ±1.4% | −0.5 | ±0.7% |
| Dielectric thickness | ±10% | +0.5 | ±5.0% |
| Etched trace width | ±8% | −0.6 | ∓4.8% |
| Finished copper thickness | ±10% | −0.15 | ∓1.5% |
| Root-sum-square | — | — | ±7.1% |
| Worst case, all terms aligned | — | — | ±12.0% |
The sensitivities are linear approximations around a 50 Ω microstrip at w/h of 2 and hold within a few percent over the ranges shown. The tolerance column is an industry window for a standard process, not a capability claim.
±10% on dielectric thickness is ±5% on impedance — half the entire window before any other term is counted. The thickness is set by the prepreg build and the press, and the press is where most of the variation lives: resin content, flow and cycle all move it.
A stack built from two plies has more variation than one built from a single ply of equivalent thickness, because it carries the variation of both. A 0.2 mm dielectric made from two 0.1 mm plies is not the same process as a single 0.2 mm ply, even when the nominal is identical.
What tightens it: one ply per dielectric where the stack allows, a press recipe developed for that stack rather than inherited, and a coupon that measures thickness as well as impedance. The coupon is the only feedback that closes the loop, which is why an impedance order without a thickness measurement on the coupon is only half verified.
RF-grade laminates are specified at ±0.02 to ±0.05 on Dk, which on a 3.5 nominal is ±0.6 to ±1.4%. Through the −0.5 sensitivity that is ±0.3 to ±0.7% of impedance — the smallest of the four terms, and rarely the cause of a failed board.
It still matters, because it is systematic within a lot. Every board from one lot shares the same Dk, so a batch drifts together rather than scattering: a group offset that is visible on a coupon average and invisible on any single board. The control is to ask for Dk on the certificate of analysis, one value per lot, and to simulate at the ± limits rather than at nominal.
Plating adds copper to the trace, which lowers impedance by widening the effective conductor, and it also reduces the dielectric gap slightly because the copper grows into the space the prepreg filled. The second effect is small and usually ignored, which is acceptable until the copper is heavy.
A 1 oz base foil finishes at 50 to 58 µm on an outer layer, so a model built on 35 µm is 15 µm thin. That is a 40% error in the copper thickness and about a 6% error in the impedance term — enough to matter, and it is the single most common mistake on outer layers.
The finished number belongs on the drawing; the plating route belongs in the process. Panel plating and pattern plating are alternative routes rather than steps in series, so a stack-up table that adds a base figure and a pattern figure double counts.
To reach ±5% by root-sum-square the four terms have to come down together. Halving the thickness window to ±5%, holding etch to ±4% and copper to ±5% gives a root-sum-square of about 3.8% — inside ±5% with margin. That is our published standard, and it is a process setting rather than a different promise.
A tighter stack means a single-ply dielectric where the build allows, or a controlled build written down; a coupon per lot rather than per shipment; and a higher scrap rate on panels that measure out. That is where the cost sits, and it is why ±5% is quoted against a stack-up rather than offered as a checkbox.
| Tier | Thickness window | Etch window | Coupon | Root-sum-square |
|---|---|---|---|---|
| Industry default, ±10% | ±10% | ±8% | Per shipment | ±7.1% |
| HSX standard, ±5% | ±5% | ±4% | Per lot, two positions | ±3.8% |
| HSX on request, ±3% | ±4% | ±3% | Per panel | ±3.2% |
Computed from the same sensitivities and tolerances as the budget table, with Dk and copper held at their standard figures. Each tier describes a level of process control, so the windows are what has to be held on the press and at etch, not a label applied after the fact.
A coupon built on the same panel, measured by TDR, proves the process at that panel position on that lot. It gives a number, a mean and a spread — and the spread is the part that predicts batch behaviour. A report with a single number is evidence of one measurement, not of control.
A coupon at the panel centre sees a different plating thickness and a different etch rate from a board at the edge. A mean sitting at 50 Ω can hide a 3% spread across the panel, so the range belongs in the report alongside the mean.
| Field | Example | What it establishes |
|---|---|---|
| Mean | 50.4 Ω | Process centring |
| Minimum and maximum | 48.9 to 51.6 Ω | Spread across the coupon |
| Panel position | Centre and edge | Plating and etch gradient |
| Layer measured | L1 and L4 | Outer against inner |
| Conditions | 23 °C, 35 ps rise time | Reproducibility |
Ask for these five fields before treating a coupon as evidence. A report carrying only a mean cannot separate a centred process from a wide one.
No. It is an order convention — a commercial default that most fabricators quote and most designs accept. No IPC clause sets it, and a drawing that attributes it to a standard is citing something that does not exist.
Yes. The coupon sits at one panel position, and plating thickness and etch rate vary across the panel. A coupon mean at 50 Ω can hide a 3% spread between centre and edge, which is why the range belongs in the report next to the mean.
Whatever the purchase document says. A drift that stays inside the ordered window is normal variation and does not warrant an investigation; a drift that leaves it does, and the first thing to check is which of the four terms moved rather than the average.
It depends on how the order defines the measurement condition. If the coupon is measured at 23 °C and the product runs at 70 °C, the Dk at temperature is a separate allowance and has to be stated separately. Mixing the two into one number produces a tolerance with no verification path.
From holding the thickness and etch windows to about half the industry default and measuring two coupon positions per lot. It is a process setting on a specific stack-up, so it is quoted against the stack-up you send rather than promised for every design.

What each BGA pitch band demands in via type, line width and stack-up, from 1.0 mm down to 0.35 mm.
Read article

Two resin systems through one press cycle, a 190 °C compromise profile, and why desmear behaves differently on each plastic.
Read article

Six finishes on flatness, shelf life and cost, and the two questions that remove four of them immediately.
Read article
Send the stack-up with the target and the tolerance named per layer. Quote in 24 hours, coupon plan included. Every build runs to IPC Class 2 standard, Class 3 on request.