Xinchenger High-Frequency PCB Manufacturing Capabilities: Seven Numbers and the Conditions Behind Them

A capability statement is only useful if it comes with the condition under which the number holds. “2 to 72 layers” means nothing until you ask what dielectric thickness 72 layers implies, and “impedance to ±10 %” means nothing until you ask on which coupon, at which frequency, with which thickness tolerance. This page sets out the seven numbers that define what the Xinchenger plant in Shenzhen can build, and states the boundary of each one, because the boundary is the part a buyer can audit and a competitor would rather not print.

High-Frequency PCB vs FR4 2 manufactured by XCE PCB for RF and microwave applications

Why a Capability Claim Needs a Boundary Condition

Procurement teams read the same twelve adjectives on every high-frequency fabricator’s website: advanced equipment, stringent quality standards, professional technical team. None of them can be inspected, so none of them survives the first engineering review. What survives is a number with a condition attached, and the habit that separates a usable supplier page from a brochure is that every claim is followed by the sentence “this holds up to here, and beyond here it does not.”

The reason the boundary matters is physical rather than commercial. A laminate has a fixed dielectric constant, a press has a fixed number of cycles it can survive, a drill has a minimum diameter for a given board thickness, and an impedance specification is the sum of four independent tolerances. Each capability number on this page is the output of one of those constraints. Written without the constraint, the number is a marketing figure; written with it, it becomes something a customer can put in a purchase order and hold the plant to.

The seven capabilities below are the ones that decide whether a high-frequency job is quotable at all: layer count, panel size, impedance tolerance, material coverage, mixed lamination, surface finish, and sample turnaround. Each section gives the number, the arithmetic behind it, and the document that proves it when an auditor asks. The process stations that sit between these numbers are covered in the high-frequency PCB manufacturing process, and the plant-level page for capacity and equipment is at process capability.

The Seven Numbers and Their Conditions

The table is the short version of the rest of this article. Read the right-hand column first: it is the column that gets deleted when the same table is copied from a competitor’s site.

CapabilityThe numberThe condition under which it holds
Layer count2 to 72 layers72 layers requires a symmetric build at 3 mm or more, because 71 dielectric layers in a 1.6 mm board would average 22 µm each
Panel size600×1200 mmLimited by the standard laminate sheet, not by the press; utilisation is set by the array, not by the plant
Impedance±10 % on coupon-qualified linesNeeds a coupon per layer per target, a stated thickness window and an etch compensation per panel
Material coveragePTFE, ceramic-filled PTFE, hydrocarbon/ceramic thermoset, FR-4Not every family is stocked in every thickness; PTFE above 6:1 aspect ratio is refused rather than attempted
Mixed laminationPTFE + FR-4, RO4000 + FR-4Three cure windows mean two or three press cycles, which is quoted as such
Surface finishENIG, ENEPIG, immersion silver, immersion tin, OSP, hard gold, HASLAbove 20 GHz a nickel-bearing finish is offered with a loss warning, not as an equal option
Sample turnaround24 hours at the fastestCovers 2 to 4 layers, one material family, no mixed lamination and no coupon measurement

Two plant facts sit behind that table and are not capability claims at all, but they are the ones a buyer usually checks first: 5,000 m² of floor space with a monthly output of about 10,000 m², roughly eighty engineers, a part-number history of more than ten thousand boards, and continuous operation in high-frequency work since 2009. Those are scale numbers. They tell you whether the plant can hold a schedule; they do not tell you whether it can hold an impedance, which is why the rest of this article is about the second question.

Layer Count: 72 Layers and 0.2 mm Vias Cannot Coexist

Layer count is quoted as a maximum because the maximum sounds impressive, but it is a trade against three other variables: finished thickness, hole diameter, and dielectric thickness per layer. The arithmetic is not complicated and it is worth doing once.

A 72-layer board has 71 dielectric gaps between copper layers. Spread over a conventional 1.6 mm finished thickness, that is 22.5 µm per dielectric layer — below the 60–80 µm that a build-up film or a prepreg sheet can reliably hold after pressing, and far below what a glass-reinforced core can be bought in. So a 72-layer board is not a thin board. At a realistic 75 µm per dielectric, 71 layers come to about 5.3 mm before copper is counted.

That thickness then collides with the hole. A 5.3 mm board drilled with 0.3 mm holes has an aspect ratio of 17.7:1, which is beyond what any plating bath can throw into reliably and more than double the 8:1 that epoxy systems are usually held to. To stay inside a 10:1 aspect ratio on a 5.3 mm board, the finished hole has to be 0.53 mm. On PTFE the ceiling is lower still: about 6:1, so a 5.3 mm PTFE stack needs 0.9 mm holes. This is the sense in which the layer-count range is a real range and not a single promise: high layer counts are for large boards with large vias, and high-density boards with 0.2 mm microvias are for thin builds.

BuildLayersDielectric per layerFinished thicknessMinimum reliable finished hole
Fine-line HDI core4–875–100 µm0.4–0.8 mm0.10–0.15 mm microvia, 1:1 aspect ratio
Standard multilayer8–16100–200 µm1.0–1.6 mm0.25 mm at 6:1
High-layer count20–4075–125 µm2.0–3.2 mm0.40–0.50 mm
Maximum layer count48–7275 µm minimum4.5–6.0 mm0.60–0.90 mm; PTFE effectively excluded

The practical consequence for a buyer is that the question to ask is not “how many layers can you do” but “at what thickness, with what hole, in which material”. A fabricator that answers the first question only has not answered. The build-up side of the same trade is worked through in high-frequency PCB stack-up design, where the number of reference planes, not the number of signal layers, is the variable that actually sets the count.

Panel Size: 600 × 1200 mm Is a Laminate Format, Not a Press Limit

The 600×1200 mm figure is quoted so often that it is worth being precise about what it is. It is the standard sheet format in which Rogers and Taconic high-frequency laminates are sold, and it is therefore the largest single piece of material that can enter the plant without a splice. A press that is larger than the sheet adds nothing, because the sheet is the constraint.

What the format buys is antenna work. A 600 mm dimension covers a 5G massive-MIMO subarray without tiling, and a 1200 mm dimension covers the long feed networks that would otherwise need a board-to-board launch, which at 28 GHz costs more in connectors and in reflection than the larger panel costs in material. What the format costs is utilisation, and that is decided at the drawing stage, not on the shop floor.

Array on a 600×1200 sheetUtilisationBoards per sheetTrade
100×100 mm, 5×11 with railsabout 76 %55Balanced: border for tooling holes plus two coupon sites
120×160 mm, 4×7 with railsabout 74 %28Fewer handling steps, larger loss per scrapped panel
80×80 mm, 7×14 with railsabout 82 %98Best material yield, most routing and most handling
Single 250×500 mm antennaabout 52 %2Only defensible when a connector launch is the worse option

The utilisation figure also carries a tolerance cost that rarely appears in a quotation. A panel needs 5–8 mm of border for tooling holes and LDI fiducials, and it needs the impedance coupon in a place where the copper density matches the interior. Putting the coupon in the open border is the classic error: the border etches at a different rate from the array, so the coupon measures a line that does not exist anywhere on the board, and the resulting impedance report is precise and wrong.

Impedance ±10 %: The Error Budget Behind the Number

The impedance specification is the one capability number that is almost always quoted without its basis, so it is the one worth unpacking. On a 50 Ω microstrip over 0.254 mm of RO4003C, four contributors dominate, and the largest of them is not the one designers spend their time on.

ContributorTypical windowEffect on 50 ΩWho controls it
Dielectric thickness±10 % (standard laminate specification)±6 %Laminate supplier and press cycle
Etched width±25 µm±2.8 % on a 0.59 mm line; ±5.5 % on 0.254 mmEtch compensation in CAM
Dk tolerance±0.05 on 3.38±0.7 %Laminate supplier
Solder mask over the trace20–30 µm−1 to −3 ΩDesign, through the mask opening
A bondply layer left out of the model0.10 mm+22 % (50 → 61.1 Ω)Whoever wrote the stack

Read down the right column and the reason a ±10 % promise is honest while a ±5 % promise is conditional becomes clear. Thickness alone consumes ±6 % at the laminate’s own stated tolerance. Getting to ±5 % overall requires buying thickness at ±5 % or better, which is a selected grade rather than a catalogue item, plus per-panel etch compensation and a coupon on every layer. That is a quotable job, but it is not a catalogue capability, and any plant that offers ±5 % as a standard figure is either measuring a coupon that does not represent the board or has not done the sum.

The measurement half of the claim is equally specific. A time-domain reflectometer with a 35 ps edge resolves features of about 3.5 mm, which is adequate for a coupon and not adequate for a connector launch; a vector network analyser in the frequency domain gives the loss that a link budget needs. The coupon definition and the acceptance frequency belong on the drawing. The comparison between the two methods is set out in impedance control in high-frequency PCB manufacturing.

Material Coverage: Four Families, Four Different Processes

“We process Rogers and Taconic” is a sentence that hides four process families which behave differently at drilling, at desmear and at lamination. The distinction that matters commercially is not the brand, it is whether the laminate contains glass and whether it is thermoplastic.

FamilyExamplesDk / Df at 10 GHzZ-axis CTEWhat it changes on the floor
PTFE with microglassRT/duroid 5880, Taconic TLY-52.20 / 0.0009237 and 217 ppm/°CSodium-naphthalenide or plasma desmear only; 6:1 aspect ratio ceiling; cold flow under pressure
Ceramic-filled PTFE, no glassRO3003, RO3010, RO30063.00 / 0.0010; 10.2 / 0.002225 and 16 ppm/°CSoft, dimensionally stable, abrasive to drills; no glass weave to hide registration errors
Hydrocarbon / ceramic thermosetRO4003C, RO4350B, RO48353.38 / 0.0027; 3.48 / 0.003746 and 32 ppm/°CFR-4 process compatible; cures at 190–200 °C; handles like epoxy
Epoxy / FR-4Standard and high-Tg FR-44.3–4.6 / 0.016–0.02550–70 below TgTen to twenty-five times the dielectric loss; used as the digital or mechanical layers in a hybrid

Two of these rows are routinely confused in supplier literature, and the confusion has consequences at the quotation stage. RO3000 series laminates are ceramic-filled PTFE with no glass — they are not hydrocarbon materials, and they are not processed like RO4000. RO4000 series laminates are hydrocarbon and ceramic in an epoxy-like thermoset, which is why they press at 190–200 °C with FR-4-compatible prepreg while PTFE needs a 260–300 °C FEP cycle. A quotation that treats the two as one family will be wrong about the number of press cycles, and therefore about both price and registration.

The loss consequence of the family choice is easy to state and hard to internalise. At 10 GHz, dielectric attenuation follows αd ≈ 2.6 · fGHz · √Dk · Df dB/in, which gives about 1.08 dB/in for FR-4, 0.18 for 4350B, 0.13 for 4003C, 0.045 for 3003 and 0.035 for 5880 and TLY-5. Between FR-4 and 5880 that is a factor of thirty on the dielectric term alone — and the reason a plant keeps all four families on the floor rather than standardising on the best one.

HDI: Where the Aspect Ratio, Not the Laser, Sets the Limit

Microvia capability is usually quoted as a diameter, which is the least informative way to quote it. What limits a microvia is the ratio of its depth to its diameter: a laser-drilled blind via in a build-up film works up to about 1:1, and preferably 0.75:1, which means the dielectric thickness has to be no greater than the hole diameter. A plant that can drill 0.1 mm holes can therefore only do so through 0.075–0.1 mm of dielectric — one build-up layer, not three.

The structural types follow from that arithmetic rather than from ambition. A 1+N+1 build has one build-up layer per side and can carry 0.1 mm microvias; a 2+N+2 build has two, and the second layer is where the registration budget starts to bite because the press operations accumulate; an any-layer build stacks vias directly and pays for it in yield. The registration arithmetic is unforgiving: each lamination adds roughly ±25 µm of positional error, so a 0.1 mm via landing on a 0.25 mm pad leaves only 75 µm of annular ring to absorb three press cycles. Staggered vias reach 96–98 % yield against 91–94 % for stacked vias in the same shop, which is why a competent fabricator will propose stagger where the design allows it.

On PTFE the constraint moves again. Laser drilling relies on the resin absorbing the beam, and PTFE does not absorb CO2 energy the way epoxy does; the practical route is mechanical drilling with plasma or sodium-naphthalenide clean-up. A 0.254 mm PTFE core asked for 0.1 mm microvias is a request for a 2.5:1 hole in a material that does not laser, and the honest answer is a different construction rather than a tighter process window. The build-up side is covered in considerations in the production process of HDI PCB.

Mixed Lamination: Three Cure Windows, Therefore Two or Three Presses

The most common capability question from a mixed-dielectric design is whether FR-4 and high-frequency material can be pressed together. The answer is a table, not a yes or no, because three systems cure in three windows and the only pair that shares a cycle is the one whose windows overlap.

SystemBond temperaturePressure / vacuumCo-curable with FR-4?
FR-4 epoxy170–190 °C200–350 psi, vacuum below 50 mbar—
RO4000 hydrocarbon / ceramic190–200 °C200–400 psiYes, with the matching prepreg, in one cycle
PTFE with FEP bonding film260–300 °C200–500 psi, slow rampNo — separate cycle, or a low-flow prepreg at 180–190 °C

The cost arithmetic is direct. A four-layer FR-4 and 4350B hybrid is one press. A four-layer FR-4 and 5880 hybrid is at least two presses, plus a bonding film, plus a second imaging and drill sequence if the PTFE layer is internal. That is the reason the instruction “just use PTFE where it matters”, issued without a layer allocation, comes back either much more expensive or with a yield number that nobody wants to sign.

Two process numbers belong to the same decision. Prepreg shrinks 0.02–0.05 % on cure, which is 0.12–0.30 mm across a 600 mm panel and larger than most layer-to-layer registration budgets, so it has to be compensated per layer. And PTFE cold-flows: a 0.254 mm core that loses 4 % of its thickness under pressure takes a 50 Ω line to 48.8 Ω, permanently, because the material is thermoplastic. Both numbers are why lamination is treated as a design decision in lamination in high-frequency PCB rather than as a shop-floor step.

Surface Finish: The Last Tenth of a Decibel Is a Metal Choice

Surface finish is bought for solderability and shelf life, but on a high-frequency board it is an electrical layer, and the number that decides the comparison is the skin depth: at 10 GHz it is 0.65 µm in copper and 1.33 µm in nickel. Any nickel thicker than that carries the current at about four times the resistivity of copper.

FinishThickness windowEffect at 10 GHzWhere it is the right choice
OSP0.2–0.5 µm organicLowest insertion loss; no metal added to the traceAssembly within months; loss matters more than shelf life
Immersion silver0.12–0.40 µmClose to OSP, no nickel barrierLong storage with controlled atmosphere
ENIGNi 3–6 µm, Au 0.05–0.10 µmSurface resistance about 2 × copper; comfortable to roughly 10 GHzFine pitch, long shelf life, aluminium wire bonding
ENEPIGNi 3–6, Pd 0.05–0.15, Au 0.03–0.07 µmSame nickel penalty as ENIGGold wire bonding plus repeated reflow
Hard gold on fingers0.5–1.5 µm over nickelNickel and cobalt under the gold dominateEdge connectors only, never a transmission line

The rule the plant applies is that above 20 GHz a nickel-bearing finish is offered with a warning rather than as an equal option: a 20 mm launch through ENIG at 20 GHz adds a fraction of a decibel that no laminate upgrade will recover, because the loss sits in the finish. The eight common finishes and their IPC windows are compared in PCB surface treatment processes.

Sample in 24 Hours: Exactly What That Covers

Fast sampling is a genuine capability and also the most over-read one, so the scope is worth stating in the same terms a purchase order would use. Twenty-four hours is achievable because a two-layer or four-layer single-material board can be imaged, etched, pressed, drilled and finished inside two shifts. It is not achievable for anything that needs a second press, a bonding film, a coupon measurement or an HDI build-up.

What is requestedRealistic turnaroundWhat moves the date
2–4 layer FR-4 or RO4000, no impedance coupon24 hoursMaterial in stock; single press cycle
2–4 layer with impedance coupon and TDR report2–3 working daysCoupon imaging, press, then coupon measurement before shipment
PTFE 2–4 layer (5880, TLY-5)3–5 working daysSodium-naphthalenide line scheduling; PTFE needs its own desmear route
Mixed FR-4 and PTFE5–8 working daysTwo or three press cycles plus an extra drill and imaging sequence
HDI with stacked microvias7–10 working daysOne press per build-up layer; X-ray and cross-section on the first article

A sample that arrives quickly and cannot be measured has bought nothing but a day, which is why the coupon belongs in the quotation even when the customer has not asked for it. The point of a prototype is to answer a question about loss or impedance; a board without a coupon answers only the question of whether the artwork opens.

Inspection: Five Gates, and What Each One Cannot See

Acceptance on a high-frequency board rests on IPC-A-600G for the workmanship visible in cross-section and on the surface, and IPC-6015 for the performance requirements of the finished board. Those are acceptance standards, not certificates, and the useful way to present them is as a sequence of gates where each gate has a blind spot.

GateWhat it provesWhat it cannot see
AOI per layerOpens, shorts, nicks, registration against the digital imageAnything inside the dielectric, and any impedance error that is uniform
Electrical testNet continuity and isolation at the test voltageA barrel that is 90 % plated passes and fails later
X-rayDrill registration, fill voids, layer alignmentDielectric thickness, resin content, copper-to-resin adhesion
Coupon TDR / VNAImpedance and loss on a representative lineAnything that differs between coupon and array, including copper density
Cross-sectionPlating thickness, annular ring, dielectric thickness, voidingThe rest of the panel: it samples a few points, not a distribution

Because each gate is partial, the first-article package is where the claim is actually tested. A first article for a high-frequency multilayer should contain a cross-section at the worst-case aspect ratio, coupon impedance against the stated budget, at least three thermal cycles at 288 °C or the reflow equivalent, and a dimensional check on layer-to-layer registration. A part that passes the electrical tests alone is a part that has been lucky once.

Where the Plant Says No

The boundary conditions that matter most are the ones that produce a refusal, because a refusal is the only capability statement that cannot be faked. Five of them are worth publishing.

PTFE above about 6:1 aspect ratio. A 3.2 mm PTFE stack cannot hold a 0.3 mm hole: the plating cannot throw into a non-wetting barrel that deep, and the Z expansion of 237 ppm/°C puts roughly 165 µm of movement on the barrel at reflow. The answer is a thinner PTFE section with FR-4 build-up, not a tighter plating process.

Resin plugging on a 0.203 mm board. The paste has to displace something. On a 0.203 mm board the dielectric compresses, PTFE cold-flows, and the cure bake shrinks the panel in the plane, which moves every hole already drilled. Below about 0.25 mm the honest options are plated-over vias without fill, or a thicker construction.

Laser cutting or water-jet cutting of PTFE laminate. PTFE decomposes above roughly 400 °C, releasing hydrogen fluoride and carbonyl fluoride, and the cut edge carbonises and will not bond. Routed or sawed edges, with a rough cut, a stress-relief pause and a finish cut, are the only route.

ENIG on a 28 GHz launch, sold as equivalent. It can be built; it will cost insertion loss that the customer will attribute to the laminate. The plant’s position is to quote a nickel-free finish and explain why.

72 layers with 0.2 mm vias in one build. The geometry does not close: 71 dielectric layers in a thin board, or a thick board with an aspect ratio no bath can plate. Any quotation that accepts both numbers at once is a quotation that will be renegotiated at first article.

The Audit Checklist: Twelve Documents to Ask For

Everything above converts into a document request. This is the list worth sending before a purchase order, because each item maps to one of the capability claims and none of them can be produced retroactively without effort.

#DocumentWhat it verifies
1Laminate batch certificate with measured Dk and DfThat the material is the grade quoted, not a substitute with a similar Dk
2Stack-up drawing including bondply and thickness toleranceThat the bondply is in the electrical model, not only in the mechanical drawing
3Coupon drawing per layer and per impedance targetThat the measurement is defined before tooling, not after
4TDR and VNA reports with sample locationsThat the centre and the edge of the panel were both measured
5Press cycle records: temperature, pressure, vacuum, rampWhich cure window was used, and how many cycles the build needed
6Drill records: diameter, spindle speed, stack height, hole position CPKWhether the drill programme was tuned to the laminate
7Plating thickness report, before and after thermal shockBarrel copper, surface copper, and ductility evidence
8Cross-section report at the worst-case aspect ratioAnnular ring, dielectric thickness, voiding, resin recession
9Thermal shock or IST resultThat the barrel survives the Z-axis movement the laminate imposes
10XRF thickness data for the surface finishNickel and gold thickness against the IPC window
11Ionic contamination resultThat the residue left after finish and cleaning is below the acceptance limit
12First-article report with the deviation listWhat was not to drawing, and why — the most informative document of the twelve

Item twelve is the one that separates two plants with identical equipment. Every first article has deviations; the difference is whether they are documented and dispositioned, or discovered by the customer in incoming inspection. A fabricator that hands over a first-article report with an empty deviation list has either an unusually mature process or an unusually relaxed definition of deviation, and the drill and plating records tell you which.

Conclusion

The seven numbers on this page are not unusual in the high-frequency industry; what is unusual is publishing the condition under which each one holds. Two to 72 layers is a range whose top end forces large vias and thick boards. Six hundred by twelve hundred millimetres is a laminate format that trades utilisation for the ability to build an antenna without a connector. Plus or minus ten percent impedance is a promise whose arithmetic is dominated by a thickness tolerance the fabricator buys rather than creates.

For a buyer, the practical use of a page like this is to convert each claim into a question. At what thickness does the layer count hold? Which coupon, at which frequency, with which thickness window? How many press cycles, and in which material? What is refused, and why? A plant that answers those four questions in writing is a plant whose capability table means something, because the answers constrain the table rather than decorate it.

XCE PCB builds high-frequency boards on the terms above: 2 to 72 layers, 600×1200 mm panels, impedance held to ±10 % on a defined coupon, PTFE, ceramic-filled PTFE and hydrocarbon laminates on the floor at the same time, samples in as fast as 24 hours for simple builds, inspection to IPC-A-600G and IPC-6015. The station-by-station windows behind those claims are in considerations in the production process of high-frequency PCBs, and the material side in PCB copper foils, which decides how much of the loss budget the copper takes before the laminate is chosen.

Frequently Asked Questions

Does a 72-layer capability mean a 72-layer board can have 0.2 mm vias?
No. Seventy-one dielectric layers in a 1.6 mm board would average 22 µm each, well below what prepreg or build-up film holds after pressing, so a 72-layer board is at least about 5.3 mm thick at a realistic 75 µm per layer. Staying inside a 10:1 aspect ratio on that thickness requires 0.53 mm finished holes, and on PTFE, where the ceiling is about 6:1, roughly 0.9 mm. High layer count and microvias are opposite ends of the same trade.
Why is impedance quoted as ±10 % rather than ±5 %?
Because dielectric thickness at the laminate’s own ±10 % specification already produces about ±6 % impedance on a microstrip, before etch error and mask are counted. A ±5 % overall window requires a selected thickness grade, per-panel etch compensation and a coupon on every layer, so it is quoted per job rather than offered as a catalogue capability.
Can RO3000 series and RO4000 series be processed the same way?
No. RO3000 series are ceramic-filled PTFE with no glass, supplied as PTFE and bonded with FEP film at 260–300 °C. RO4000 series are hydrocarbon and ceramic in an epoxy-like thermoset, pressed at 190–200 °C with FR-4-compatible prepreg. Treating them as one family gets the number of press cycles wrong, which changes both price and registration.
What does a 24-hour sample actually include?
Two to four layers in a single material family, no mixed lamination, and no impedance coupon measurement — typically FR-4 or RO4000 with material in stock. Adding a coupon and a TDR report moves it to two or three working days; PTFE moves it to three to five; a mixed FR-4 and PTFE build to five to eight because of the extra press cycles.
Why would a fabricator refuse a 3.2 mm PTFE board with 0.3 mm holes?
Two reasons, both physical. The aspect ratio is about 10.7:1 against a practical PTFE ceiling near 6:1, because plating cannot throw into a non-wetting barrel that deep. And Z-axis expansion of 217–237 ppm/°C puts roughly 165 µm of movement on the barrel across a reflow excursion, which a thin plating cannot survive. The fix is a thinner PTFE section with FR-4 build-up.
Which documents prove an impedance claim?
Three together: the coupon drawing per layer and per target, the stack-up drawing that includes the bondply in the electrical model, and the TDR or VNA report with the sample locations shown. A single impedance figure without the coupon geometry and the thickness window it was measured over cannot be audited.
RF & Microwave PCB Fabrication

Need the Capability Sheet for Your Stack?

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

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