High-Frequency PCB vs FR4: Manufacturing Process Differences, Station by Station

The high-frequency PCB vs FR4 question is usually answered with adjectives: tighter tolerance, finer lines, better equipment. That list does not help anyone standing in front of a press. Walk the same twelve-station flow with both material families and the picture is far more specific — only five stations actually change the physics. Lamination, drilling, desmear and plating, solder-mask adhesion and surface finish each move to a different window, and each has a signature failure when the window is missed. The other seven stations run on the same machine with a different recipe card. This article walks the flow in order, gives the window at each station, and names the defect that appears when the window is exceeded.

High-Frequency PCB vs FR4 manufactured by XCE PCB for RF and microwave applications

Why the Standard Answer Fails

Most comparisons of high-frequency PCB vs FR4 manufacturing processes reduce to a claim that RF boards need better everything. In practice a modern FR-4 shop already owns laser direct imaging, automatic optical inspection and impedance coupons; nothing about the equipment is unusual. What differs is that six material properties push five stations outside the FR-4 recipe, and the remaining stations stay exactly where they were.

Those six properties are the loss factor, the permittivity tolerance, the absence of a glass transition, the through-thickness expansion, the modulus under pressure, and the surface energy. Only the first one never appears on the shop floor — dissipation factor is a design input, not something a press operator can adjust. The other five each map to a station.

Material propertyFR-4, typicalRF laminate, rangeStation it changes
Dk / Df4.2–4.6 / 0.015–0.025 at 10 GHz2.20–10.2 / 0.0009–0.0037None directly; tolerance shows up at imaging and etch
ConstructionWoven glass in epoxyPTFE with microglass, ceramic-filled PTFE, hydrocarbon/ceramic thermosetCutting, drilling, desmear
Thermal transitionsTg 130–180 °C, Td 300–325 °CPTFE has no usable Tg; melt 327 °C; Td above 500 °CLamination, reflow
Z-axis CTE50–70 below Tg, 200–300 above32 (4350B), 46 (4003C), 237 (5880) ppm/°CDrilling, plating, via reliability
Modulus / cold flowRigid, negligible creep at press temperaturePTFE creeps; 4 % thinning is routineLamination, drilling
Surface energy40–45 mN/m18–20 mN/m for PTFESolder mask, coverlay, bonding

Everything below follows from this table. If a process step does not appear in the right-hand column, it does not need a new machine — it needs a new recipe, and usually a new acceptance criterion.

Stations 1 and 2: Incoming Inspection and Cutting

Both flows start by cutting a panel to size, and here the first real difference appears: dimensional stability. Every laminate moves with temperature, and the amount it moves sets how the panel must be measured and how much material has to be left for the finish cut.

Over a 600 mm panel and a 20 °C swing in shop temperature, a TLY-5 panel moves 0.31 mm (26 ppm/°C), a 5880 panel 0.37 mm (31 ppm/°C), a 4350B panel 0.12 mm (10 ppm/°C) and a 4003C panel 0.13 mm. On an FR-4 panel of the same size the figure is around 0.19 mm. This is why RF shops measure at 20 ±2 °C and why the drill program is generated from a measurement, not from the drawing.

LaminateIn-plane CTEGrowth over 600 mm, 20 °CPractical consequence
FR-416 ppm/°C0.19 mmStandard tolerance stack
RO4350B10 ppm/°C0.12 mmBest dimensional match to copper artwork
RO4003C11 ppm/°C0.13 mmSame order as 4350B
RT/duroid 588031 ppm/°C0.37 mmMeasure and cut in the same climate
Taconic TLY-526 ppm/°C0.31 mmRough cut, relax, finish cut

Two cutting rules are specific to RF laminates. PTFE must not be laser cut: at around 400 °C it decomposes and releases hydrogen fluoride and carbonyl fluoride, leaving a carbonised edge that will not bond. Waterjet is only marginally better, because it drives water and abrasive into an edge that is about to be laminated. And ceramic-filled grades wear carbide quickly, so diamond-coated tooling is the norm. The working practice is a rough cut leaving 3–5 mm, a relaxation pause, then a finish cut — the same reasoning that governs the cutting step described in substrate cutting for high-frequency PCBs.

Stations 3 and 4: Inner-Layer Imaging and Etching

Here the difference is copper weight rather than imaging. RF laminates are commonly supplied with half-ounce copper because conductor loss in a thin line is dominated by surface roughness and by how much of the line survives the etch. Imaging on 0.5 oz is not harder than on 1 oz; what changes is that the finished width now carries a larger share of the drawing tolerance.

Etching removes copper laterally as well as vertically. The undercut per side is approximately the copper thickness divided by the etch factor, so at an etch factor of 1.5 a 1 oz layer loses about 23 µm per side while 0.5 oz loses about 11 µm. On a 50 Ω microstrip over 0.254 mm of 4350B the designed width is 0.59 mm; a 25 µm total error on that line is 2.8 % of the impedance, but the same 25 µm on a 0.127 mm dielectric, where the line is 0.24 mm wide, is 5.5 %.

ItemFR-4 multilayerRF laminateEffect on impedance
Copper weight, inner1 oz typical0.5 oz typicalLess undercut, but thinner conductor
Undercut per side~23 µm at 1 oz, EF 1.5~11 µm at 0.5 oz0.56 mm line: 1.6 % vs 0.8 %
0.24 mm line, 25 µm errorRare at this widthCommon on thin dielectric5.5 % of 50 Ω
Imaging methodContact or LDILDI below 0.075 mmRegistration, not width

The claim that RF boards routinely run 2/2 mil (0.05 mm) lines is worth correcting, because it drives bad quoting. At 0.5 oz a 0.05 mm line loses over 20 % of its width to undercut and its resistance rises by a third; 0.075 mm is the practical floor for controlled impedance, and 0.10 mm is the comfortable one. Fabricators who quote 0.05 mm are usually quoting a minimum feature, not a controlled-impedance feature. The arithmetic behind that separation is in inner-layer circuit fabrication.

Station 5: Lamination, Where the Two Flows Diverge

If one station has to be singled out, it is this one. FR-4 prepreg cures between 170 °C and 190 °C under 200–400 psi, and the resin flows, gels and crosslinks in a single cycle. Hydrocarbon/ceramic thermosets such as 4350B and 4003C run at 190–200 °C and are close enough that a mixed stack can share a press cycle. PTFE does not cure at all: it is bonded by a FEP film or a 3001 bondply that has to be taken to 290–315 °C, or by a low-flow prepreg at 180–190 °C. Those three windows cannot be satisfied at once.

SystemBonding mediumPress temperatureCan it share a cycle?
FR-4Epoxy prepreg170–190 °CYes, with RO4000 series
RO4000 series4450B or FR-4 prepreg190–200 °CYes, with FR-4
PTFE + FEPFEP film / 3001 bondply290–315 °CNo — separate cycle required
PTFE + low-flow prepregLow-flow prepreg180–190 °CYes, with FR-4, at the cost of a higher Dk bond layer

Two consequences follow. The first is thickness: dielectric height sets impedance, and a 10 % error in h moves a 50 Ω line to 46.8 or 53.0 Ω — the entire ±10 % specification spent on one variable. On PTFE the risk is asymmetric, because the material creeps under pressure: a 4 % thinning takes 50.0 Ω to 48.8 Ω and it does not come back when the pressure is released.

The second is registration. Prepreg shrinks 0.02–0.05 % on cure, which on a 600 mm panel is 0.12–0.30 mm — two to six times the ±0.05–0.075 mm registration budget that a multilayer stack is planned around. PTFE-based stacks avoid that specific term but introduce a bond-layer thickness term instead, which is why the press recipe and the stack-up are written together. Voids behave the same way in both families but matter more in RF: a 10 µm air layer under a line raises its impedance to 52.3 Ω, and 50 µm takes it to 60 Ω while the panel average still passes. The full arithmetic is in lamination in high-frequency PCB manufacturing.

Station 6: Drilling

Drilling is where the two materials behave most differently at the tool tip, and the reason is mechanical rather than thermal. FR-4 is glass-filled epoxy: hard, abrasive, and it produces a clean shear. PTFE is soft and creeps, so the drill pushes material instead of cutting it, and the exit side tends to burr rather than fracture. Ceramic-filled grades are the opposite problem: abrasive enough to wear a bit out in a few hundred holes.

ParameterFR-4PTFE-based RF laminateNote
Point angle120–130 °130–140 °Sharper point, less push
Helix angle30–35 °35–45 °Faster chip evacuation from soft material
Spindle speed60–100 k rpm80–120 k rpmPaired with lower feed per revolution
Stack height3–4 panels2–3 panelsLess push-out, less burr
Burr acceptance0.05 mm0.025 mmBurr survives plating and lifts the joint
Aspect ratio limitup to 10:16:1 practicalPlating throw, not drilling

The aspect-ratio limit is not a drilling limit at all; it is the point at which the plating solution can no longer renew itself inside the barrel, and PTFE needs more margin because the barrel has to be activated before it can be plated. A drill that produces a 0.30 mm finished hole needs a 0.25 mm drill, since plating adds roughly 25 µm per wall — and on a 1.6 mm board that is 6.4:1, already at the edge. Drill parameters and their acceptance criteria are covered in detail in drilling for high-frequency PCBs.

Station 7: Desmear, Activation and Plating

FR-4 desmear is a permanganate process: the alkaline solution attacks the epoxy smear left by the drill and leaves a micro-roughened barrel for the palladium catalyst to cling to. PTFE is immune to it. There is no epoxy to attack and no smear in the FR-4 sense; instead the surface has to be chemically modified so that something can bond to it at all. That is done with sodium naphthalene or with plasma, and it is not optional — skipping it produces a barrel that plates but does not adhere.

The plating itself is where the through-thickness expansion becomes visible. Over a reflow excursion of 215 K on a 1.6 mm board, a 5880 stack expands about 82 µm through the thickness, a TLY-5 stack about 75 µm, a 4350B stack 11 µm and an FR-4 stack 17–24 µm below its glass transition. The copper barrel has to stretch by that amount every cycle, which is why RF boards specify plating with at least 20 % elongation and why the aspect ratio is capped.

StepFR-4 windowPTFE windowWhat fails if missed
DesmearPermanganate, 70–80 °CSodium naphthalene or plasmaBarrel pull-away, no adhesion
CatalystPd/Sn, standardSame, after activationSkip plating, voids
Electroless copper45–50 °C, 2–3 µm/hSameThin seed, then voids in the electrolytic step
Surface current density25–30 ASFRaise to reach 42 µm at 5:1Barrel thinner than surface, cracked in thermal cycling

The last row is the one that gets overlooked. At an aspect ratio of 5:1 the barrel receives measurably less copper than the surface, so a board specified at 25 µm nominal may have 18 µm inside the hole. On FR-4 with a Z-CTE of 50 ppm that is survivable; on a 237 ppm PTFE stack it is the mechanism behind every barrel crack found after temperature cycling. Plating control for this family is covered in plating in high-frequency PCB manufacturing.

Station 8: Solder Mask

Solder mask is treated as cosmetic in most process discussions and it is not. Two numbers make the case: the mask is an electrical layer with a permittivity of 3.5–4.2 (white ink reaches about 6.5), and covering a tuned 50 Ω line with it drops the impedance by 1–3 Ω. On a 0.24 mm line where the whole tolerance is 5 Ω, that is a decision, not a finish.

On PTFE the problem is adhesion before it is electrical. The surface energy of PTFE is 18–20 mN/m against 40–45 for epoxy, which is below the wetting threshold of ordinary liquid photoimageable mask. The industry answer is a plasma or sodium-naphthalene pre-treatment plus a mask chemistry formulated for low-energy surfaces, and even then the acceptance test is tape pull plus a 260 °C float, not a visual check. Where the RF structure cannot tolerate mask at all, the mask is opened over the line and the impedance is designed without it.

One process detail follows from this: on an RF board the mask opening is part of the electrical drawing. If the opening is redrawn at the fab to meet a minimum dam width, the impedance of that line changes and the coupon no longer represents the product. The mask step and its acceptance criteria are set out in solder mask application for high-frequency PCBs.

Station 9: Surface Finish, Where the Common Advice Is Backwards

Most articles on this subject say that RF boards use ENIG or hard gold for better high-frequency performance, and that FR-4 uses HASL, immersion silver or OSP because they are cheap. The first half is defensible, the second half is wrong in a way that costs money. At 10 GHz the skin depth in copper is 0.65 µm, in nickel 1.33 µm and in gold 0.78 µm. An ENIG deposit is 0.05–0.10 µm of gold over 3–6 µm of nickel, so the gold is thinner than its own skin depth and the current travels almost entirely in the nickel, whose surface resistance is 2.04 times that of copper. Measured as insertion loss on a 50 Ω line, that is roughly 0.1–0.15 dB/in at 10 GHz and 0.2–0.3 dB/in at 28 GHz.

FinishBarrier layerLoss impact at 10 GHzWhere it belongs
OSPNone, organic film 0.2–0.5 µmBaseline, lowestRF lines that are not bonded or wire-bonded
Immersion silverNone, 0.12–0.40 µm AgWithin a few percent of baselineRF plus press-fit and test points
ENIG3–6 µm Ni+0.1–0.15 dB/inFine-pitch assembly, shelf life, bonding base
ENEPIGNi plus 0.05–0.15 µm PdSimilar to ENIGGold and aluminium wire bonding
Hard goldNi plus 0.8–1.5 µm AuHighest, but only on fingersEdge connectors, selective only
HASL, lead-freeNone, 1–2 µm SnAgCuThermal shock, not lossThrough-hole boards; not on PTFE

Two further points matter on an RF board. HASL at 250–260 °C is a thermal shock to a PTFE stack whose through-thickness expansion is 237 ppm/°C — the same 80 µm excursion computed earlier — and the air knife can thin the dielectric under a pad, moving the impedance permanently. And immersion silver is not chosen because silver conducts slightly better than copper, which it barely does; it is chosen because it adds no nickel. The finish comparison against the IPC windows is in surface treatment processes for PCB manufacturing, and the process-level version in surface treatment in high-frequency PCB production.

Stations 10 to 12: Routing, Test and Packing

Routing looks identical and is not quite. Laser and waterjet are excluded for the same reason as in cutting, so both families are routed mechanically; the difference is tool wear on ceramic-filled grades and the burr criterion on PTFE, where a soft edge tears rather than chips. Fineblanking is not an option on PTFE either, because the material cold-flows instead of shearing, which is why the tooling note in the drawing usually reads drill-only for locating holes.

Test is where the two flows genuinely differ in what can be proven. FR-4 is verified by electrical test, AOI and a dimensional check. An RF board adds a coupon measurement: a TDR for impedance and a vector network analyser for insertion and return loss, on a coupon that is built in the same panel and etched in the same step as the product. Without that coupon there is no evidence of impedance at all — the ±10 % figure on a purchase order is a specification, not a measurement.

The same applies to the tolerance claim. RF boards are routinely specified at ±10 %, the same as FR-4, and a fabricator quoting ±2 % is describing a coupon result on a benign test pattern rather than a production capability. ±5 % is achievable with coupon TDR feedback, ±3 % is a premium service at low frequency, and below that the measurement uncertainty of the coupon itself becomes the limit.

Packing is the one place where the usual assumption reverses. PTFE absorbs about 0.02 % moisture, among the lowest of any laminate, so the bake-before-assembly rule that exists for FR-4 is not the driver. What matters on an RF board is the finish: OSP has a shelf life measured in months and is sensitive to handling, while ENIG survives storage far longer. Baking an OSP board before assembly destroys the film.

The Five Numbers That Decide Whether a Board Is Buildable

Every discussion of process differences eventually arrives at a quotation, and a quotation is only as good as the five constraints behind it. These are the ones that make an RF stack difficult, in the order they usually bite.

ConstraintPractical limitWhy it bitesWhat to change first
Aspect ratio6:1 on PTFE, 10:1 on FR-4Plating throw inside the barrelReduce thickness or increase drill
Dielectric height tolerance10 % standard, 5 % premiumSets 6 % of the impedance budgetMove to a thicker core, or to a design that tolerates a wider line
Line width vs copper weight0.075 mm at 0.5 ozUndercut eats a fifth of the width0.5 oz everywhere the impedance is controlled
Registration budget0.05–0.075 mmPrepreg shrinkage is 0.12–0.30 mm on 600 mmSymmetrical stack, per-cycle compensation
Z-axis strain per cycle4.4 % on 5880Barrel must stretch by 80 µmDuctile plating, fewer reflows, thinner board

The order matters. A design that needs 0.075 mm lines, a 1.6 mm finished thickness and a 0.25 mm hole is asking for 6.4:1 on PTFE, which is over the limit no matter how good the drill is. Moving to a 1.0 mm finished board takes the same design to 4:1 and removes the problem entirely — cheaper than any process upgrade.

Where the Yield and the Money Go

Comparisons usually stop at material price, which is the smaller half of the difference. The larger half is that an RF board loses yield at different stations, and the losses are concentrated rather than uniform.

StationTypical RF scrap mechanismDetected atRecoverable
LaminationVoid, bond-layer thickness drift, registration shiftCoupon TDR, C-SAM, cross-sectionNo
DrillingBurr, barrel roughness, hole positionAOI after plating, X-rayNo
PTHBarrel pull-away, thin plating at 5:1Thermal cycling, cross-sectionNo
Solder maskDelamination after reflow, mask on RF lineTape test, 260 °C floatSometimes, by stripping
FinishNickel thickness out of window, silver tarnishXRF, visualStrip and re-plate, rarely

Compared with an FR-4 board of the same layer count, an RF board at 4350B or 4003C typically lands at 3–5 times the bare-board price, a 5880 stack at 8–15 times, and a hybrid at a weighted figure that depends on how much of the area is RF. Most of that multiple is material, but the part that surprises buyers is the cycle time: a PTFE stack needs its own press cycle, its own drill recipe and often a re-plate after a mask rework, so a 10-day FR-4 turnaround becomes 15–20 days without anyone being inefficient.

What to Put on the Fabrication Drawing

The process differences only get executed if they are written down. Ten items cover almost every case:

1. Laminate part number and thickness per layer, with the bond layer named separately from the core — a missing 0.10 mm bond layer can take a 50 Ω line to 61 Ω. 2. Copper weight per layer, and where it is 0.5 oz. 3. Impedance targets with the coupon design, not just the value. 4. The mask: chemistry, opening rule over RF lines, and whether mask is allowed on controlled-impedance lines at all. 5. Finish by area, because a board can carry ENIG on the assembly side and hard gold on the fingers. 6. Maximum reflow cycles, since every cycle is a strain increment. 7. Drill chart with finished and drilled diameters, plus the aspect ratio. 8. Acceptance standard — IPC-A-600G Class 2 or 3 for the visual, IPC-6015 for the metal-core case if it applies. 9. Panel size and array, so the press cycle is planned around a real panel rather than a guess. 10. A statement of what must be measured and reported, because a coupon nobody measures proves nothing.

XCE PCB runs both flows in one plant: 2 to 72 layers, panels to 600×1200 mm, impedance held to ±10 %, samples in as fast as 24 hours, inspected to IPC-A-600G and IPC-6015. Mixed stacks are built in separate cycles rather than forced into one, which is the only way the press windows above can be respected — the design-side version of that discipline is in mixing Rogers and FR-4 in one board.

Conclusion

The difference between high-frequency and FR-4 manufacturing is narrower and more specific than the usual description. Seven of twelve stations run the same machine and the same operators with a different recipe. Five stations change physics: lamination moves to a separate press window because PTFE is bonded by a 290–315 °C FEP film, drilling tightens because the material creeps instead of shearing, desmear becomes a sodium-naphthalene or plasma step because permanganate has nothing to attack, solder mask needs a low-energy surface treatment to adhere at all, and the finish choice is driven by whether it adds a 3–6 µm nickel layer that would carry the current at 10 GHz.

Read against the numbers, the two families are not a quality ranking. A 4350B stack shares a press cycle with FR-4 and differs mainly in dielectric height control. A 5880 stack does not, and adds a 237 ppm through-thickness expansion that governs the plating specification. That is the whole of the process distinction, and it is checkable: five constraints, five windows, and a coupon that either measures what was drawn or does not.

Frequently Asked Questions

Can high-frequency laminates and FR-4 be laminated in the same press cycle?
The hydrocarbon/ceramic family can, and often is. RO4350B and RO4003C are pressed at 190–200 °C, which overlaps the FR-4 window of 170–190 °C, so a mixed stack uses one cycle. PTFE cannot: it is bonded by an FEP film or 3001 bondply at 290–315 °C, or by a low-flow prepreg at 180–190 °C. The first route needs a separate cycle; the second keeps the shared cycle but adds a higher-Dk bond layer that must be in the impedance model.
Is ±2 % impedance control realistic on a high-frequency board?
Not as a production specification. ±10 % is standard for both FR-4 and RF laminates, ±5 % is achievable with coupon TDR feedback, and ±3 % is a premium service that also requires the coupon itself to be verified. Quotes of ±2 % usually refer to a measurement on a benign test pattern. The physical limit is not the etcher but the laminate thickness tolerance, which alone accounts for about 6 % of the impedance budget at ±10 %.
Why is ENIG not the best finish for a 10 GHz board?
Because of the nickel. Skin depth at 10 GHz is 0.65 µm in copper and 1.33 µm in nickel, while an ENIG deposit is 0.05–0.10 µm of gold over 3–6 µm of nickel. The gold is thinner than its own skin depth, so the current runs in the nickel, whose surface resistance is 2.04 times copper. That costs roughly 0.1–0.15 dB/in at 10 GHz and 0.2–0.3 dB/in at 28 GHz. OSP or immersion silver avoid the barrier layer entirely; ENIG is chosen for flatness, shelf life and bonding, not for loss.
Do PTFE boards need baking before assembly?
Not for moisture. PTFE absorbs about 0.02 % water, among the lowest of any laminate, so the bake that protects FR-4 from popcorning is not the driver. Two other rules do apply: do not bake an OSP-finished board, because it destroys the protective film, and do hold and measure panels at 20 ±2 °C, because in-plane expansion over a 600 mm panel is 0.31–0.37 mm at PTFE CTE.
What aspect ratio can a PTFE board support?
About 6:1 in practice, against 10:1 for FR-4. The limit is plating throw, not the drill: PTFE needs activation before plating, and the solution renews poorly inside a deep barrel. A 0.30 mm finished hole needs a 0.25 mm drill once 25 µm per wall is allowed, so a 1.6 mm board is already at 6.4:1. Reducing finished thickness to 1.0 mm takes the same design to 4:1.
Does solder mask really change the impedance?
Yes, by 1–3 Ω on a 50 Ω line. Mask permittivity is 3.5–4.2, and white ink reaches about 6.5, so covering a tuned line with it lowers the impedance measurably. On a 0.24 mm line where the tolerance band is ±5 Ω, that is a design decision rather than a cosmetic one, which is why the mask opening over RF lines belongs on the electrical drawing and not on a fabrication note.
RF & Microwave PCB Fabrication

Need a Fabricator Who Runs Both Flows in One Plant?

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

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