High-Frequency PCB Manufacturer: Essential Process Capabilities and Hardware Infrastructure

A capability table is easy to write and hard to verify. Every supplier claims fine lines, tight impedance and a vacuum press; almost none of the claims are comparable, because a number without a copper weight, a coupon geometry, a measurement frequency or a batch size is a slogan rather than a specification. This article treats the capability question the way a buyer should: seven questions that separate a reproducible window from a marketing figure, the hardware that cannot be substituted for any amount of skill, and the three documents that are worth more than the brochure. It is a screening script, not a sales list — the point is to know what a credible answer sounds like before the audit starts.

High-Frequency PCB Manufacturer at XCE PCB's high-frequency PCB factory

Three Kinds of Numbers in a Capability Table

Capability figures fall into three classes, and only one of them is a commitment. Physical limits are set by material and geometry — a 0.075 mm line on 2 oz copper is not a tolerance, it is an impossibility. Measured values are what a coupon reports on a particular lot with a particular method. Commitments are what the supplier will put in a purchase order and accept a claim against. A table that mixes the three is unreadable, and most tables mix them.

Number on the pageWhat it usually isQuestion that makes it comparable
Minimum line width 0.075 mmA physical limit at one copper weightAt which copper weight, and with which etch factor?
Impedance ±10 %A commitment, if the method is statedCoupon geometry, frequency band, points per layer?
Impedance ±5 %A selected-laminate service, priced separatelyWhat does it cost, and on which stack?
Layers 2–72A real range, but not an interchangeable oneAt which layer count does the aspect ratio bind?
Dielectric constant 2.2–10.6A range that spans several process familiesWhich grades, and can you bond them together?

Read that way, a capability table stops being a scoreboard and becomes a questionnaire. The seven questions below are the ones that decide whether a factory can build a specific board, in the order in which the answers eliminate candidates fastest.

Question 1: Which Material Families Do You Actually Run?

The first question is not “do you process Rogers” but which family, because the families are different processes inside the same plant. A supplier who runs RO4000 well and quotes a 5880 stack on the same line has answered a different question from the one asked.

FamilyBonding routeHole wallA credible answer mentions
Hydrocarbon / ceramic thermoset (4003C, 4350B, 4835)190–200 °C with 4450B bondplyStandard desmear, conventional PTHUL 94 status per grade, design vs process Dk
Ceramic-filled PTFE (3003, 3006, 3010)260–300 °C fusion, or 180–190 °C low-flow prepregSodium naphthalene or plasmaCold flow allowance, peel strength, no glass
Micro-glass PTFE (5880, 5880LZ, TLY-5)FEP film 260–300 °CSodium naphthalene mandatoryZ-CTE 217–237, aspect ratio 6:1, no laser
High-Dk ceramic PTFE (6010.2LM, RF-10)As PTFE, heavily filledAs PTFE, plus filler pull-outTool wear, Dk tolerance ±0.25

The follow-up question that reveals real experience is narrow and specific: how do you bond PTFE to copper, and what peel strength do you guarantee? A supplier who has done it will answer with a chemistry and a number — sodium naphthalene or plasma activation, a FEP film or low-flow prepreg, and a peel figure on the order of 5.5 N/mm for 1 oz ED foil on 5880. A supplier who has not will answer with the word “special treatment”.

Question 2: What Does Your ±10 % Mean?

Impedance tolerance is the most quoted and least comparable number in the industry. On a high-frequency board it is the sum of four independent terms, and only one of them is under the operator’s control.

TermTypical variationEffect on 50 ΩControlled by
Dielectric thickness±10 %±6 %Laminate and press
Etch±25 µm±2.8 % at 0.254 mm; ±5.5 % at 0.127 mmImaging and etch
Dk±0.05±0.6–0.7 %Material batch
Solder mask−1 to −3 Ω−2 to −6 %Design

A supplier who quotes ±10 % and means it will say three things without being asked: the coupon geometry, the frequency band at which it is measured, and how many points per layer are taken. Note that the coupon number is usually measured on bare copper, which reads 1–3 % higher than the finished board with mask on it — a detail worth knowing before comparing two reports.

For ±5 %, the honest description is a service, not a capability: it requires a selected laminate, a measured press, more coupon points and a scrap allowance, and it should be quoted as an uplift. Anyone offering it at the standard price on a 0.254 mm stack is either measuring something else or has not done the budget — thickness alone consumes ±6 % there. The same arithmetic underlies the observation in high-frequency versus standard PCB that the error budget changes owner when the material changes, and the acceptance side is set out in high-frequency PCB testing.

Question 3: 0.075 mm Lines at What Copper Weight?

Fine-line capability is meaningless without the copper weight, because the etch attacks sideways as well as down. The lateral attack per side is approximately the copper thickness divided by the etch factor, so the same artwork produces very different finished widths on 0.5 oz and on 2 oz.

CopperSide etch at EF 2.50.075 mm artwork → finishedVerdict
0.5 oz (17 µm)7 µm61 µmPossible, 19 % narrow
1 oz (35 µm)14 µm47 µmMarginal, 37 % narrow
2 oz (70 µm)28 µm19 µmNot producible

Two consequences for an audit. First, ask for the copper weight every time fine-line capability is claimed, and ask whether the figure is artwork width or finished width — the difference is a whole copper thickness. Second, expect LDI rather than film for anything below about 0.10 mm: a phototool moves 40 ppm with temperature and humidity, which is 0.024 mm across a 600 mm panel, a quarter of the feature itself. The etch side is covered in PCB etching.

Question 4: What Is Your Aspect Ratio Limit, and Why?

The aspect ratio question separates factories faster than any other, because the answer requires understanding rather than equipment. On PTFE the working limit is 6:1 finished depth to diameter, and the reason is a combination of plating uniformity and the Z-axis expansion of the dielectric.

A 1.6 mm board at 6:1 allows a 0.27 mm finished hole; at 10:1 it would allow 0.16 mm, and the plating solution would struggle to exchange inside it. Add the thermal term and the argument becomes quantitative: PTFE expands 82 µm across a 215 K excursion on a 1.6 mm board while the copper barrel expands about 6 µm, so the barrel has to absorb roughly 76 µm of differential strain. A thinner board reduces the strain in proportion; a deeper hole concentrates it at the corners and reduces the plating thickness where the stress is highest.

The credible answer therefore mentions the aspect ratio and the plating ductility in the same breath — elongation of at least 20 % for PTFE work — and states that laser drilling is not used on PTFE because the decomposition products above roughly 400 °C are hydrogen fluoride and carbonyl fluoride and the carbonised edge will not bond. A supplier who offers laser drilling on PTFE is describing the FR-4 part of a hybrid.

Question 5: How Many Press Cycles, in What Order?

Hybrid construction is where a plausible factory and a capable one diverge, because the answer cannot be improvised. The three bonding systems cannot be opened in the same cycle: FR-4 cures at 170–190 °C, RO4000 with 4450B at 190–200 °C, PTFE fuses at 260–300 °C.

So a stack containing PTFE and FR-4 is built in at least two cycles, PTFE first, FR-4 second, and the second cycle re-heats a finished assembly — adding a registration term and re-opening the thickness account. A good answer to this question names the bondply, the order, the temperature for each cycle and the resulting number of drill spans. A poor answer names a temperature and hopes. The pressure side has a trap of its own: PTFE does not resist load, so the highest pressure the press can deliver is not the best setting — a 4 % thinning turns 50.0 Ω into 48.8 Ω permanently. The construction detail is in multilayer lamination for high-frequency PCBs.

Question 6: Which Finish, and How Thick Is the Nickel?

Every finish question has a thickness answer, and on high-frequency work the thickness of the nickel is the one that matters most. At 10 GHz the skin depth is 0.65 µm in copper and 1.33 µm in nickel; an ENIG deposit is 3–6 µm of nickel under 0.05–0.10 µm of gold, so the current travels in nickel, whose surface resistance is about twice copper’s.

FinishThickness windowNickel in the path?Where it belongs
ENIGNi 3–6 µm, Au 0.05–0.10 µm (IPC-4552B)YesFine pitch, multiple reflow, wire bonding with caution
Immersion silver0.12–0.40 µm at ±4σ (IPC-4553A)NoMillimetre wave, low loss, press-fit
OSP0.2–0.5 µm (IPC-4555)NoLowest loss, single reflow, short shelf life
ENEPIGNi 3–6, Pd 0.05–0.15, Au 0.03–0.07 µm (IPC-4556)Yes, plus palladiumGold and aluminium bonding, no black pad risk

Two audit notes. Black pad is a phosphorus problem before it is a gold problem: keep the deposit in a 7–10 wt % window, cap the gold at 0.10 µm, and write the IPC-4552B corrosion rating into the purchase order. And above 20 GHz, the loss difference between ENIG and a nickel-free finish is 0.2–0.3 dB/in, which on a 100 mm feed line is a number worth arguing about. The full comparison is in PCB surface treatment processes.

Question 7: What Evidence Comes With the Lot?

The last question is the one that converts promises into a purchase order. Three documents do more work than a factory tour, because they are specific to the lot rather than to the plant.

A coupon measurement report with the coupon geometry, the frequency band, the number of points per layer and the instrument used. A microsection report at defined locations — hole wall, plating thickness, registration at a via chain — with a photograph and a scale, not a summary table. A material certificate per lot carrying the manufacturer, the grade, the batch number and the key parameters (Dk, Df, thickness, copper type), which is what makes a substitution visible. A supplier who can produce these three on request, for a lot that has already shipped, is a supplier whose other numbers can be believed.

The absence of them is not proof of a bad factory; it is proof that the buyer has no way to know. That distinction matters because the documents cost almost nothing to supply and everything to reconstruct afterwards.

Hardware That Cannot Be Substituted by Skill

Most equipment lists are decoration. Five items are not, because their absence produces a specific defect that no operator can work around.

EquipmentWhat it actually buysWhat happens without it
LDI (laser direct imaging)Per-panel distortion compensation, registration to about ±0.01 mmPhototool drift of 40 ppm = 0.024 mm on 600 mm; fine lines and layer registration both suffer
Vacuum lamination pressRemoval of air before the resin gelsVoids: 10 µm of air raises 50.0 Ω to 52.3 Ω locally
High-speed drilling (≥ 120 krpm)Clean entry and exit in soft PTFESmear, burr, nail-head; the defect appears as an intermittent open later
VNA with a coupon fixtureInsertion and return loss on the actual stackNo way to demonstrate loss; acceptance becomes an opinion
Microsection capabilityBarrel thickness, registration, void evidenceLatent barrel defects ship; scrap is discovered at fifteen times the cost

Note what is deliberately not on the list: the newest press, the largest factory, the cleanroom class. Those affect throughput and cost, not the ability to build a specific high-frequency stack. A buyer who screens on throughput is optimising the wrong variable.

Factory Tour: Four Things Worth Looking At

If a visit happens, four observations carry more information than the equipment hall.

The lay-up room. Is it separated, temperature and humidity logged, with prepreg in sealed bags and a bake log? Dust between layers is a delamination cause, and the room is where it either exists or does not. The bath analysis records. Are they daily, signed, with the phosphorus content of the nickel bath and the copper concentration of the plating bath? A bath controlled by calendar rather than by analysis is where black pad and thickness drift come from. The coupon log. Does it show values with dates and lot numbers, or only pass marks? A log with numbers can be argued with; a log with ticks cannot. The scrap Pareto. Is it current, and does it name a mechanism rather than a station? A factory that can name its top three scrap mechanisms knows its process; one that cannot is still guessing.

None of these four requires equipment or an interpreter, and all four predict the answer to the seven questions above more reliably than a presentation does.

What XCE PCB States, and What It Does Not Claim

For transparency, the numbers this site commits to are these: 2 to 72 layers, panels to 600×1200 mm, impedance ±10 % verified on coupons, samples in as fast as 24 hours, 5,000 m² of floor space, more than 80 engineers, operation since 2009, a monthly capacity around 10,000 m², and inspection to IPC-A-600G and IPC-6015. Materials are drawn from Rogers, Taconic, F4B, TP-2 and FR-4 families, from Dk 2.20 to 10.2.

What is not claimed: no certification badge is quoted on this page, because a certificate is an audit document rather than a selling point and should be verified against the issuing body rather than read on a website. ±5 % impedance is available as a selected-laminate service where the stack allows it, not as a standard figure. And the layer range is a capability at the plant level, not a licence to combine anything with anything — a 72-layer PTFE stack with a 6:1 aspect ratio and a 0.25 mm hole is a design conversation before it is a quotation.

That is the distinction this article is about. A capability is a window plus the evidence that the window was held; everything else in a table is a hope. The process map that the windows sit in is at high-frequency PCB manufacturing process.

Conclusion

Auditing a high-frequency PCB manufacturer comes down to seven questions and three documents. The questions are: which material families do you actually run; what does your ±10 % mean and how is it measured; 0.075 mm lines at what copper weight; what is your aspect ratio limit and why; how many press cycles and in what order; which finish with what nickel thickness; and what evidence arrives with the lot. The documents are the coupon report, the microsection report and the material certificate.

The trap in almost every capability table is a number without its qualifying condition: ±5 % without a stack, 0.075 mm without a copper weight, a Dk range that spans four process families as though they were interchangeable. Ask for the condition and most of the table dissolves; what remains is the part worth paying for.

XCE PCB will answer all seven against a specific stack, in writing, with coupon and microsection data. Send the drawing rather than the inquiry form — the answer is only meaningful for a particular board. The material side of the same decisions is in Rogers board manufacturing difficulties.

Frequently Asked Questions

What are the essential capabilities of a high-frequency PCB manufacturer?
Four, in order of how quickly they eliminate candidates: the ability to run the PTFE and thermoset families on separate bonding routes with the right hole-wall chemistry; a stated and measured impedance process with coupon geometry and frequency defined; fine-line capability quoted against a copper weight; and hybrid construction with the press order and cycle count declared. Hardware follows from those, not the other way round.
Is ±5 % impedance control better than ±10 %?
It is tighter, and on a 0.254 mm stack it is not a process setting — laminate thickness alone varies ±10 %, which is ±6 % on impedance. A ±5 % window therefore requires a selected laminate, a measured press, more coupon points and a scrap allowance, and should be quoted as an uplift. At standard pricing, a ±5 % claim usually means a different measurement, not a better process.
Why does copper weight matter for fine-line capability?
Because etch attacks sideways at roughly the copper thickness divided by the etch factor. At EF 2.5 the lateral attack is about 7 µm per side on 0.5 oz, 14 µm on 1 oz and 28 µm on 2 oz — so a 0.075 mm artwork becomes 61 µm, 47 µm or 19 µm respectively. Fine lines and thick copper are mutually exclusive, not a matter of skill.
Can a supplier laser drill PTFE?
No. Above about 400 °C PTFE decomposes and releases hydrogen fluoride and carbonyl fluoride, the edge carbonises and the carbonised layer will not plate. Laser drilling is used on the FR-4 portions of a hybrid; PTFE is drilled mechanically at 80–120 krpm with a 130–140° point angle, and the hole wall is then activated with sodium naphthalene or plasma.
What documents should come with a high-frequency PCB lot?
Three: a coupon measurement report stating geometry, frequency band and points per layer; a microsection report at defined locations with photographs and a scale; and a material certificate per lot carrying manufacturer, grade, batch and key parameters. These are lot-specific, cost little to produce, and are the only way a buyer can tell a substitution from a specification.
Does a bigger factory mean better high-frequency capability?
Not necessarily. Capacity affects cost and lead time, while capability is a set of windows: bonding temperature per family, aspect ratio, etch precision, impedance measurement. A small plant with a vacuum press, LDI, a VNA and a disciplined coupon log will build a 5880 stack better than a large plant that treats it as FR-4 with a different laminate.
RF & Microwave PCB Fabrication

Audit Us Against These Seven Questions

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

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