Processing Considerations for High-Frequency PCBs

Processing considerations for high frequency PCBs are usually written as a list of good intentions — pick a good material, control the etching, mind the lamination — and a list of good intentions cannot be inspected. The useful version of the same subject is arithmetic. A 50 Ω specification of ±10 % is a budget of ±5 Ω, and every process step from substrate cutting to surface finish spends part of it: dielectric thickness ±10 % is worth ±6 % of impedance, ±25 µm of etch is worth another ±2.8 %, and a nickel layer under the gold quietly moves the current path in the wrong direction at 10 GHz. This article allocates that budget across nine steps, gives the window each step has to hold, and corrects four claims from the conventional list, including the one about laser drilling high-frequency laminate.

Processing Considerations for High-Frequency PCBs

Reframing the List as an Impedance Budget

The original version of this article named seven considerations and supported none of them with a number, which is why it reads as advice rather than as engineering. The same seven subjects become useful once they are attached to the quantity they actually move. On a 0.254 mm laminate of Dk 3.38, a 50 Ω microstrip is 0.59 mm wide, and the ±10 % window on the drawing is ±5 Ω of room for everything that happens afterwards.

StepWhat it movesTypical magnitudeEffect on 50 ΩRandom or systematic
Substrate cuttingDatum position, panel distortion0.3 mm over 600 mm at 20 °C swingIndirect, through registrationSystematic per lot
Inner-layer etchingTrace width25 µm per edge2.8 % (±1.4 Ω)Random, varies across the panel
LaminationDielectric thickness10 % of nominal6.0 % (±3.0 Ω)Random, lot and press dependent
Drilling and platingHole wall, pad annulus, plane clearance25 µm of copper per wallNeutral on a microstrip, critical on a striplineSystematic
DesmearResin residue, wall adhesion2–5 µm of smearNone until the via failsSystematic
Solder maskEffective Dk over the trace20 µm of Dk 3.5–4.2 ink1–3 Ω lowerSystematic
Surface finishWhere the current flows3–6 µm of nickel0.1–0.15 dB/in at 10 GHzSystematic
Test couponThe number you reportMask-free coupon reads 1–3 % higher1–3 % offsetSystematic

Add the three random terms in root-sum-square — 6.0 %, 2.8 % and 0.75 % for Dk tolerance — and the result is 6.7 %, or ±3.3 Ω. That looks comfortable against ±5 Ω until the systematic terms are added on top: the mask takes 1–3 Ω in one direction, and a coupon measured without mask reads 1–3 % high. A worst-case stack of just three terms already reaches 9.5 %, and the specification is gone.

That is the first processing consideration worth stating plainly: ±10 % is not a property of the material, it is the residue of a budget that has to be watched. A fabricator that holds ±5 % on one stack and only ±10 % on another is not doing anything differently — the second stack has a thinner dielectric, so the same ±25 µm of etch costs twice as much impedance.

Consideration One: Substrate Cutting Sets the Datum

Cutting looks like the least technical step and is the one that quietly erases registration margin. PTFE-based laminates move with temperature far more than FR-4, and the movement is in the panel, not in the machine. Over a 600 mm panel and a 20 °C swing, a laminate with 31 ppm/°C in X expands 0.37 mm; at 26 ppm it is 0.31 mm; a hydrocarbon-ceramic at 11 ppm moves 0.13 mm.

Laminate familyCTE X / Y (ppm/°C)Growth over 600 mm at 20 °CCutting consequence
PTFE, micro-glass (5880)31 / 480.37 / 0.58 mmRough cut 3–5 mm oversize and let the panel relax; no laser (decomposition above 400 °C releases HF); water jet only with edge trim
PTFE, woven glass (TLY-5)26 / 150.31 / 0.18 mmSame, plus weave-direction marking, or the drill programme drifts one way only
Hydrocarbon-ceramic (4350B)10 / 120.12 / 0.14 mmBehaves close to FR-4; standard routing and stacker
FR-4 reference14 / 160.17 / 0.19 mmThe baseline that PTFE is being compared against

Two rules follow. The first is that the cut must leave material to be removed after the panel has relaxed, so the finishing cut happens once, at the end, on a stable sheet — this is the point of the 3–5 mm allowance rather than a tidy margin. The second is that PTFE must not be cut with a laser: the decomposition temperature is far below the ablation threshold a CO2 source works at, and the products include hydrogen fluoride, which is a hazard and a contamination problem for every board in the cell. The full set of requirements is in substrate cutting in high-frequency PCB manufacturing.

Consideration Two: Etching Buys 2.8% of the Budget

Etching is the step most often described as “precision” and least often quantified. It removes copper isotropically, so it cuts sideways as it cuts down, and the undercut is roughly the copper thickness divided by the etch factor. On 1 oz foil at an etch factor of 1.5 that is about 23 µm per edge; on 2 oz it is 28 µm per edge, which is why heavy copper carries a 0.15 mm minimum width rather than a preferred one.

Translated into impedance, 25 µm per edge on a 0.59 mm line is a 4.2 % width change, worth about 2.8 % of impedance. Move the same tolerance onto a 0.127 mm dielectric, where the 50 Ω line is only 0.29 mm wide, and the identical 25 µm is worth 5.5 %. Nothing in the etch line changed; only the line width did.

The process variable that matters most is not the etcher but the distribution. Spray etch rate depends on fresh chemistry reaching the surface, so a dense array of fine lines depletes the etchant locally and etches narrower than an open area, while the panel edge etches wider. The same drawn width can finish 10 % apart across one panel. Two controls answer it: etch compensation applied in CAM at no less than twice the copper thickness, and coupons placed in a copper-balanced area rather than the open border. The step-by-step version is in inner-layer circuit fabrication.

Consideration Three: Lamination Owns the Largest Single Term

Dielectric thickness is the dominant term because impedance depends on the ratio of width to thickness, while width is fixed. A ±10 % thickness tolerance is worth ±6 % of impedance — more than twice the etch term, and it is the term most often left out of a tolerance discussion because the laminate datasheet quotes a nominal, not a distribution.

PTFE adds a failure mode that no other family has. The material cold-flows under pressure, so the dielectric finishes thinner than the stack assumed: a 4 % thinning takes a 50 Ω line to 48.8 Ω before any other error is counted. Bondply is the other trap — if a 0.10 mm bondply is left out of the impedance model because it was treated as “just glue”, the same geometry calculates 61.1 Ω, a 22 % error introduced by a line item on the stack drawing.

 FR-4Hydrocarbon-ceramic (4000 series)PTFE + FEP
Press window170–190 °C190–200 °C260–300 °C
Can be co-cured?With 4000 series, in one cycleWith FR-4, in one cycleNo — separate cycle, or a bondply transition
Cold flowNegligibleNegligibleUp to 4 % of dielectric thickness
Prepreg shrink0.02–0.05 %0.02–0.05 %0.02–0.05 %
Shrink over 600 mm0.12–0.30 mm0.12–0.30 mm0.12–0.30 mm, plus flow

The three press windows are the reason a mixed stack is a scheduling decision, not just a materials decision: the three families cannot be cured in one cycle, so a hybrid build is either two cycles with a bondply transition or a redesign. The mechanics are in lamination in high-frequency PCB manufacturing.

Consideration Four: Drilling and Plating Move the Reference Plane

A finished hole is not the drill. Electroless copper plus pattern plating adds about 25 µm per wall, so the finished diameter is the drill minus 50 µm, and the drill programme has to be written backwards from the finished requirement. That arithmetic is routine; what is not routine is the mechanical consequence of the material.

PTFE expands in Z far more than FR-4 and has almost no glass to restrain it. Across a 1.6 mm board and a 215 °C excursion to reflow, a 5880 build moves 82 µm — four to seven times the movement of a thermoset. The copper barrel has to stretch with it, which is why ductility matters more than tensile strength in the plating spec, and why the aspect ratio for PTFE is held at 6:1 or better even though the same shop runs 10:1 on FR-4.

LaminateZ-CTE (ppm/°C)Expansion, 1.6 mm, 215 °CBarrel requirement
5880 (PTFE, micro-glass)23782 µmElongation ≥20 %, aspect ratio 6:1, sodium-naphthalene or plasma desmear
TLY-5 (PTFE, woven glass)21775 µmSame
4350B (hydrocarbon-ceramic)3211 µmStandard FR-4 barrel, still 25 µm per wall
FR-4 above Tg50–7017–24 µmThe reference everyone is calibrated against

Plating also sets the plane clearance that nobody draws. A 25 µm wall on a via that passes through a ground plane closes the antipad by 25 µm per side, and on a 0.2 mm via in a 0.6 mm antipad that is an 8 % change in the clearance, which is a measurable change in via impedance and a small change in stub behaviour. The same arithmetic applies to drill positional tolerance: it consumes annulus, not impedance. Details are in drilling and plating.

Consideration Five: Desmear Is the Step That Cannot Be Generic

Every plated through hole needs the resin smear from drilling removed before electroless copper, and on FR-4 the industry answer is a permanganate swell-and-etch. That chemistry does not attack PTFE. A permanganate line run on a PTFE panel will produce a hole that passes visual inspection and fails thermally, because there is no mechanical key between the barrel and the dielectric.

The two processes that do work are sodium-naphthalene etching, which chemically modifies the PTFE surface so it can be plated, and plasma, which uses an oxygen or oxygen-CF4 discharge. Both are slower, both are batch processes, and both have to be qualified on the actual laminate rather than on a coupon from a different supplier. This is a processing consideration with no substitute: it is not a parameter to be optimised, it is a route to be selected.

The same reasoning disposes of the claim in the original article that laser drilling is used on high-frequency materials to minimise the heat-affected zone. Laser drilling is the tool for HDI build-up dielectric — 60–100 µm of FR-4-based or polyimide film — where the energy is absorbed in a thin layer and the underlying copper stops the via. PTFE cores are mechanically drilled, because the risk is not a heat-affected zone but thermal decomposition, and because there is no glass mat to hold a clean wall. A CO2 source on 5880 leaves carbonised residue and a hole that cannot be reliably plated.

Consideration Six: Solder Mask Is an Electrical Layer

Solder mask is specified as a mechanical and chemical layer and behaves as a dielectric. Its Dk is 3.5–4.2 depending on the ink, white ink reaches about 6.5 because of the titanium dioxide loading, and 20 µm of it over a microstrip raises the effective Dk of the air above the trace. The measured result is 1–3 Ω lower impedance than the bare coupon, which is why a coupon measured without mask reads 1–3 % high and why the two numbers should never be compared directly.

On a 0.254 mm stack the effect is small enough to absorb inside a ±10 % window. On a 0.102 mm stack it is not, and neither is it on a 0.203 mm board where the mask thickness is a meaningful fraction of the dielectric. The practical controls are to keep mask off impedance-critical lines where the design allows it, to specify thickness rather than “standard”, and to qualify the finished board rather than the coupon. The process side is covered in solder mask application.

Consideration Seven: Surface Finish Puts Nickel in the Current Path

At 10 GHz the skin depth in copper is 0.65 µm and in nickel it is 1.33 µm. ENIG deposits 3–6 µm of nickel under 0.05–0.1 µm of gold, so the current at microwave frequencies is travelling in nickel, not in gold, and nickel has about seven times the resistivity of copper. The surface resistance of the conductor rises by roughly a factor of two compared with bare copper.

FinishLayer under the surfaceAdded loss at 10 GHzWhen it is the right choice
OSPNone — organic film on copperLowest availableHighest-frequency, shortest shelf life accepted
Immersion silver0.1–0.3 µm silver, no nickel+0.02–0.05 dB/inThe usual compromise for RF
ENIG3–6 µm nickel, 0.05–0.1 µm gold+0.1–0.15 dB/in (0.2–0.3 at 28 GHz)Fine pitch, long storage, wire bonding after hard gold
Immersion tin0.8–1.0 µm tin+0.05–0.08 dB/inPress-fit and solder-only assemblies

The decision is therefore not cosmetic. A 10 GHz link with 0.38 dB/in of total loss on 4350B is spending 0.1–0.15 dB/in of it inside the nickel, which is a third of the budget for a layer that exists to protect the copper. Where the run is long and the loss budget tight, OSP or immersion silver is the processing consideration, and the corrosion trade-off has to be accepted explicitly rather than by default. See surface treatment.

Consideration Eight: Heat Leaves Through the Dielectric, Not the Copper

The intuition that thicker copper improves heat dissipation is correct for a heat spreader and wrong for a high-frequency board, because the bottleneck is not the copper but the 0.2–0.8 mm of laminate underneath it. PTFE has a thermal conductivity of 0.20 W/m·K, roughly a five-hundredth of copper, so a 0.254 mm dielectric has the same thermal resistance as 500 mm of copper of the same area. Adding copper weight changes the spreading on the surface; it does not open the vertical path.

Materialk (W/m·K)0.254 mm dielectric equals this much copper
58800.20508 mm
30030.50203 mm
4350B0.69147 mm
4003C0.71143 mm
6035HTC1.4471 mm

The only real vertical path is metal, and it has to be put there on purpose. A filled 0.2 mm via through a 1.6 mm board has a thermal resistance of about 127 K/W; twenty-five of them on a 0.5 mm pitch give about 5 K/W, which is the difference between a 10 °C rise and a 40 °C rise at the same power. That is a design and drilling decision, not a copper-weight decision, and it is why thermally critical RF boards are specified by via array rather than by foil thickness.

Consideration Nine: The Coupon Proves the Process, Not the Board

Impedance is verified on a coupon, and a coupon is a witness, not a sample. It is etched in the same panel, laminated in the same press cycle and plated in the same line, which is what makes it worth measuring. It is also mask-free, placed at the panel edge, and connected through a launch that the product does not have. Those three differences account for the 1–3 % offset between coupon and product, and for the fact that a ±3 % coupon result does not deliver a ±3 % board.

TDR locates a discontinuity and gives an impedance profile along the trace; VNA gives insertion and return loss over frequency and, with the right fixture, the material’s Dk and Df. Neither proves the via stub, the connector launch or the mask loading. The practical acceptance chain is coupon first, then a first-article measurement on the product where the design allows a test access, then a cross-section for the geometry. The method set is in testing in high-frequency PCB production, and the tolerance question is the same one that separates a ±10 % shop from a ±3 % shop.

Three Material Families, Three Process Windows

Most processing mistakes on high-frequency boards come from applying an FR-4 habit to a laminate that does not share it. The three families differ on every step that matters, and the differences are not refinements — they are route changes.

StepPTFE (5880, TLY-5)Hydrocarbon-ceramic (4350B, 4003C)FR-4
CuttingRough cut and relax; no laser; edge trim after water jetStandard routingStandard routing
Inner-layer etchSame chemistry, tighter compensation because the lines are widerStandardStandard
Lamination260–300 °C, cold flow up to 4 %190–200 °C, co-curable with FR-4170–190 °C
DrillingAspect ratio 6:1, sharp tools, no smear left behindStandard, 10:1 achievableStandard
DesmearSodium naphthalene or plasma onlyPermanganatePermanganate
MaskLow-temperature cure, adhesion promoterStandardStandard
FinishOSP or silver preferred; ENIG acceptableAnyAny

The comparison that puts the whole subject in perspective is between these families and a standard board, which is a different article entirely — high-frequency PCB versus standard PCB — but the processing consequence is one line: a shop that runs both has to keep the routes separate from the moment the panel is cut, and a shop that runs only FR-4 cannot fake the second half of this table.

What to Settle Before the First Drill

The list below is what a fabrication review should close before the panel is cut. Each item is a processing consideration because each one, if left open, is resolved by whichever machine happens to run first.

ItemWho decidesWhy it cannot be left to the shop floor
Material family per layerDesignerIt sets press cycle, desmear route and drill parameters at once
Impedance target per layer, with coupon geometryDesignerWithout it, width and impedance are two independent numbers that will drift apart
Dielectric thickness with tolerance, including bondplyDesigner and fabricatorThe largest term in the budget; a missing bondply is a 22 % error
Finished hole sizes and aspect ratioFabricatorDrill is written backwards from the finished diameter, not forwards
Desmear route per materialFabricatorA permanganate line on PTFE passes inspection and fails thermal cycling
Mask thickness and keep-out over RF linesBoth1–3 Ω of systematic offset
Finish by frequency bandBothNickel in the current path is a loss decision, not a cosmetic one
Thermal path: via array or noneDesignerCopper weight does not substitute for vertical metal
Coupon type, position and acceptance limitFabricatorA coupon at the panel edge is not the product
What “±10 %” means — coupon or productBothThe two numbers differ by 1–3 % and the argument is usually about which was meant

Ten items, one page. A programme that closes them before the first drill spends its tolerance where it chose to, and a programme that does not spends it wherever the process happens to land.

Conclusion

The seven considerations in the original version of this article were the right subjects with no arithmetic attached. With the arithmetic attached, the picture is not seven independent good practices but one budget: dielectric thickness at ±6 % is the largest single term, etch at ±2.8 % is second, Dk tolerance and mask contribute the systematic offsets, and the sum of the worst cases exceeds the ±10 % that the drawing promises long before the board reaches the finisher.

Three conclusions follow. Dielectric thickness, not trace width, is where a high-frequency programme should put its tolerance effort, because PTFE cold flow and a forgotten bondply each cost more than the whole etch line. Desmear and drilling are route decisions that cannot be optimised into existence on the wrong material. And the finish is an electrical choice: at 10 GHz, 3–6 µm of nickel carries the current, not the 0.1 µm of gold on top of it.

XCE PCB runs these routes separately rather than approximately: 2 to 72 layers, panels to 600×1200 mm, impedance held to ±10 %, PTFE drilled at 6:1 with sodium-naphthalene or plasma desmear, samples in as fast as 24 hours, inspected to IPC-A-600G and IPC-6015. The step-by-step version of the whole sequence is in the high-frequency PCB manufacturing process.

Frequently Asked Questions

What are the main processing considerations for high frequency PCBs?
Nine steps move the result: substrate cutting (PTFE grows 0.31–0.37 mm over 600 mm at a 20 °C swing), inner-layer etching (±25 µm per edge is 2.8 % of impedance on a 0.254 mm stack), lamination (±10 % dielectric thickness is ±6 % of impedance), drilling and plating (25 µm of copper per wall, 82 µm of Z expansion on 5880 through reflow), desmear (sodium naphthalene or plasma for PTFE, never permanganate), solder mask (1–3 Ω lower impedance), surface finish (nickel carries the current at 10 GHz), thermal path (through the dielectric, not the copper), and coupon testing.
Is ±10 % impedance tolerance realistic on a PTFE board?
It is realistic if the terms are tracked rather than assumed. The three random terms — dielectric thickness ±10 %, etch ±25 µm and Dk ±0.05 — combine to about 6.7 % root-sum-square, which fits inside ±10 %. Add the systematic terms and it does not: solder mask pulls impedance 1–3 Ω low, PTFE cold flow of 4 % takes 50 Ω to 48.8 Ω, and a bondply omitted from the model gives 61.1 Ω instead of 50 Ω. On a 0.127 mm stack, ±25 µm of etch alone is worth 5.5 %, and ±10 % becomes hard to hold without tighter thickness control.
Can laser drilling be used on PTFE high-frequency laminate?
Not on the PTFE core, which is why the common statement needs correcting. Laser drilling is the standard tool for HDI build-up dielectric — 60–100 µm of FR-4-based or polyimide film over copper — where the via stops on the underlying pad. On 5880 or TLY-5 the material decomposes rather than ablates cleanly, the residue carbonises, and the hole cannot be reliably plated; the 400 °C decomposition also releases hydrogen fluoride. PTFE cores are mechanically drilled at an aspect ratio of 6:1 or better and then desmeared by sodium naphthalene or plasma.
Why is thicker copper not the answer to heat on an RF board?
Because the heat has to cross the laminate before it reaches any copper. PTFE conducts 0.20 W/m·K, so 0.254 mm of it has the same thermal resistance as 500 mm of copper of the same area; on 4003C at 0.71 W/m·K the equivalent is 143 mm. More copper weight improves spreading on the surface and leaves the vertical bottleneck unchanged. The vertical path has to be metal: a filled 0.2 mm via through 1.6 mm is about 127 K/W, and twenty-five of them on a 0.5 mm pitch give roughly 5 K/W.
Which surface finish should a 10 GHz board use?
OSP if the loss budget is tight and shelf life can be managed, immersion silver as the usual compromise, ENIG where fine pitch or long storage dominates. At 10 GHz the skin depth in copper is 0.65 µm and in nickel 1.33 µm, so 3–6 µm of ENIG nickel carries the current rather than the 0.1 µm of gold above it, raising surface resistance roughly twofold and adding 0.1–0.15 dB/in at 10 GHz, or 0.2–0.3 dB/in at 28 GHz.
Does solder mask change the impedance of a high-frequency trace?
Yes, by 1–3 Ω in the low direction, because the ink has a Dk of 3.5–4.2 and displaces air above the trace — white ink reaches about 6.5 because of its titanium dioxide loading. The practical consequence is that a mask-free coupon reads 1–3 % higher than the finished board, so the two numbers should never be compared directly. On a 0.254 mm stack the effect fits inside a ±10 % window; on a 0.102 mm stack it does not.
RF & Microwave PCB Fabrication

Need the Impedance Budget Held Across All Nine Steps?

Send us your Gerber files and requirements for a manufacturability review and a tailored quotation.

2–72 layers ±10% impedance tolerance Manufacturing since 2009