High-Frequency PCB Testing Checklist: Ten Tests, Ten Pass Numbers

A high frequency PCB testing checklist is useful only when each line carries a number, an instrument and a standard — otherwise it is a list of good intentions. The usual version of this article named eight categories of test and gave a window for none of them, and three of the eight were not board-level tests at all: gain and harmonic distortion belong to an active device, and “thermal resistance testing” was a description of thermal cycling. This checklist keeps ten tests that can actually be run on a bare board, states the pass window for each, and says what each one still cannot prove.

High-Frequency PCB Testing Checklist manufactured by XCE PCB for RF and microwave applications

A Test List Is Not a Test Plan

Four questions separate a test plan from a test list. What instrument produces the number? Which standard fixes the window? What is the window, in a unit that can be compared? And what does this test fail to show? A checklist that answers only the first is a schedule, not a specification.

The four questions also sort the tests by owner. Impedance, insertion loss and crosstalk are properties of the stack-up and the artwork, so they are measured on a coupon that is built beside the product. Thermal shock, peel strength, plating thickness and cleanliness are properties of the process, so they are measured on samples cut from the same panel. Warpage and hole-wall cross-sections are properties of the finished article. A test that is run on the wrong object — a coupon number reported as a product number — is the most common way a passing report hides a failing board.

The sequence matters as much as the content. Non-destructive tests run first and in full; destructive tests run last, on a sample, and are the only ones that can be run once. Putting a cross-section in front of a TDR measurement means the coupon is gone before the electrical number exists.

The Master Checklist: Ten Tests and What Each One Proves

The table below is the whole checklist on one page. The sections that follow give the arithmetic behind the windows and the reason for each blind spot.

TestInstrumentGoverning standardPass windowWhat it cannot show
Characteristic impedanceTDR, 35 ps rise, on a couponIPC-6012 §3.8; coupon per IPC-2141A50 Ω ±10 % standard, ±5 % on requestAnything closer together than about 3 mm
Differential impedanceDifferential TDRIPC-6012 §3.890 or 100 Ω ±10 %Mode conversion inside a connector
Insertion lossVNA, S21, 2×-thru de-embeddingCoupon loss per laminate data sheetWithin 10 % of the αd predicted from DfRadiation and connector loss
Return lossVNA, S11Design budgetBetter than −10 dB across the bandLoss downstream of the reflection
Crosstalk4-port VNA, NEXT and FEXTDesign budgetBelow −30 dB differential at 5 GHzNon-linear effects in the driver
Thermal shock / cyclingAir-to-air chamberIPC-TM-650 2.6.7.2No delamination, resistance change ≤10 %Field life; it is an acceleration, not a prediction
Z-axis CTETMA on the laminateIPC-TM-650 2.4.24Within 10 % of the data sheet valueLocal stress at a single via
Peel strengthPull tester, 1 in stripIPC-TM-650 2.4.8Material-specific; 1.05 N/mm for RO4003C classAdhesion after three reflows
Coating thicknessXRF, confirmed by micro-sectionIPC-4552B (ENIG), IPC-4553 (Ag)Ni 3.0–6.0 µm, Au 0.05–0.10 µmPorosity and pinholes
Ionic contaminationExtraction conductivityIPC-TM-650 2.3.25≤1.56 µg/cm² NaCl equivalentLocalised residue under a component

Two tests that belong on the same page are missing from the table because they are destructive and sampled: warpage and hole-wall cross-section. They are covered in section 11, together with the sampling rate that keeps them affordable.

Impedance: Where the ±10 % Actually Goes

The ±10 % window is not a measurement tolerance, it is a budget, and it is consumed by four independent contributions before the TDR is switched on. On a 0.254 mm dielectric with Dk 3.38, the 50 Ω line is 0.59 mm wide, and the budget divides as follows.

ContributionMagnitudeEffect on 50 ΩWho controls it
Dielectric thickness±10 % on 0.254 mm±6 %Press cycle, prepreg lot
Etched width±25 µm±2.8 %Etch compensation in CAM
Solder mask2–6 % of impedance−1 to −3 ΩMask type and thickness
Dk lot spread±0.05 on 3.38±0.7 %Laminate supplier
Combined (RSS)—±7.0 %Leaves 3 % for coupling and coupon edge

Thickness dominates, and it is the term nobody negotiates because it is set by the press, not by the artwork. That is why a fabricator can hold ±5 % on one stack and only ±10 % on another without doing anything differently: the difference is whether the dielectric is 0.5 mm or 0.1 mm. On a 0.127 mm core the same ±25 µm etch moves impedance by 5.5 %, and the thickness term alone eats most of the budget.

The instrument adds its own limitation. A 35 ps rise time travelling at 0.18 mm/ps on a Dk 2.67 line resolves features separated by roughly 3 mm, so a coupon 50 mm long with two transitions at its ends reports one averaged impedance rather than a profile. Anything that changes impedance over a shorter distance — a via antipad, a connector launch, a neck-down — is invisible to the coupon and is only seen by a TDR on the finished board, where the launch itself is the largest discontinuity.

See impedance control in high-frequency PCB manufacturing for the process side of the same budget, and stack-up design for how the thickness is chosen in the first place.

Insertion Loss: S21 on a Coupon, Never “Gain”

Gain is a property of an amplifier. A bare board has no gain; it has loss, and the measurable quantity is S21, expressed in dB per inch at a stated frequency. The original version of this article listed a “gain test” and a “harmonic distortion test” among board-level RF tests, and neither can be performed on an unpopulated laminate. Harmonic distortion requires a non-linear device; a passive coupon is linear by construction and will show nothing.

The dielectric part of the loss is predictable from the data sheet, which is what makes the measurement a check rather than a mystery. For a 50 Ω line the attenuation is close to αd ≈ 2.6 · fGHz · √Dk · Df in dB per inch, and at 10 GHz the numbers separate the materials by an order of magnitude.

LaminateDkDf at 10 GHzDielectric loss, dB/inOn a 100 mm coupon
FR-44.30.0201.084.3 dB
RO4350B3.480.00370.180.72 dB
RO4003C3.380.00270.130.52 dB
RO30033.000.00100.0450.18 dB
RT/duroid 58802.200.00090.0350.14 dB

What the coupon does not capture is copper. At 10 GHz the current occupies the outer 0.65 µm of the foil, so surface roughness and the finish stack sit in the current path: an ENIG finish puts 3–6 µm of nickel under 0.05–0.1 µm of gold, and nickel at that frequency has about 2.04 times the surface resistance of copper. The measured S21 of an ENIG board is therefore legitimately worse than the data-sheet prediction, and the gap is not a defect. It is the reason loss-critical designs finish with OSP or immersion silver.

De-embedding is the other half of the measurement. A 100 mm coupon measured through two connectors at 10 GHz carries a launch discontinuity that can exceed the loss of the line itself, so a 2×-thru or TRL calibration kit built on the same stack is not optional. Without it the number is the test fixture’s, not the laminate’s. The full test ladder — which test is run at which production stage — is set out in testing steps in high-frequency PCB production.

Return Loss, Crosstalk and Jitter: the Passive-Board Reality

Return loss is S11 and insertion loss is S21; they are not two views of one number. A −10 dB return loss means about 10 % of the incident power is reflected, and the remaining 90 % continues down the line — some of it to be absorbed, some to reflect again. For a design budget, −10 dB across the band is the usual floor for a controlled-impedance interconnect, −15 dB where a connector is involved.

Crosstalk is measured as NEXT and FEXT on a coupled pair, in a four-port configuration, and the pass window depends on the standard rather than on physics: −30 dB differential at 5 GHz is a common requirement for a digital link, while an RF board often cares more about the isolation of the RF path than about a digital neighbour. What the coupon cannot show is mode conversion inside a connector, which is why crosstalk measured on a coupon and crosstalk measured in the chassis often disagree by 10 dB.

Jitter is where the original text drifted furthest. Total jitter is dominated by the transmitter and the reference clock; a passive board contributes deterministic jitter through loss and skew, and the useful board-level measurement is skew between an intra-pair pair and between lanes, in picoseconds, not a jitter number in UI. A coupon can measure skew; it cannot measure a bit error rate, because there is no receiver. The parameters that drive these numbers are collected in ten critical parameters for high-frequency PCBs.

Thermal Shock and Cycling: Which Test Catches Which Failure

“Thermal resistance testing” was the heading in the original, and the three tests listed under it were thermal shock, high-temperature ageing and CTE. None of them measures thermal resistance, which is a steady-state quantity in K/W computed from thickness over conductivity — 0.20 W/m·K for RT/duroid 5880 against 0.69 W/m·K for RO4350B means a 0.254 mm layer of the first is 3.5 times the thermal resistance of the second, and no chamber test will tell you that.

What the chamber does tell you is whether the via survives the mismatch between copper and laminate in the z direction. The tests are not interchangeable.

TestConditionDurationPass criterionFailure it reveals
Thermal shock / cycling−55 to +125 °C, air to air100 cycles, 15 min dwellNo delamination; resistance change ≤10 %Barrel cracking from z-axis expansion
Solder float288 °C, 10 s3 passesNo delamination, no barrel pull-outMoisture and resin outgassing
Reflow simulation260 °C peak, 3 passes3 cyclesNo delamination at 25×Lead-free assembly damage
High-temperature ageing125 °C in air1000 hPeel strength retained ≥80 %Oxidation, resin degradation

One hundred cycles is a qualification number, not a life prediction. What it establishes is that the plated barrel can accommodate the expansion of the laminate in the z direction; whether it will do so for twenty years is a different question, answered by the CTE and the ductility of the plating rather than by the cycle count. For boards destined for a via-in-pad or a 0.2 mm thin core, the sample should be the product, not a coupon, because the failure depends on the aspect ratio of the hole.

The hole itself is described in properties and functions of the PCB through hole, and the reflow side in PCBA manufacturing process.

Z-Axis Expansion: Measure the Material, Verify the Via

CTE is a material property measured by TMA on a bare laminate, and it is the one number on this checklist that is better read from a data sheet than measured on a coupon — every lot of the same laminate will agree within a few percent, and the test destroys a sample to confirm what the certificate already says. What is worth measuring is the consequence: how far the stack moves between room temperature and reflow.

LaminateCTE X / Y, ppm/°CCTE Z, ppm/°CExpansion, 1.6 mm over 25–260 °C
RT/duroid 588031 / 48237+82 µm
Taconic TLY-526 / 15217+75 µm
RO4003C11 / 1446+16 µm
RO4350B10 / 1232+11 µm
RO301013 / 1116+5.5 µm

An 82 µm expansion against an 11 µm one is the whole reason PTFE boards need a different plating specification: the copper barrel has to stretch, so elongation at fracture becomes an acceptance criterion and 20 % is the number to ask for on a 5880 stack. It is also the reason the aspect ratio limit for PTFE is around 6:1 rather than the 10:1 that FR-4 tolerates — a deeper hole concentrates the same expansion over a longer barrel, and the plating has less reserve.

Peel Strength: the Adhesion Number Behind Every Rework

Peel strength is measured by pulling a 1 in wide strip of foil at 90° and reporting force per width. It is the number that decides whether a board survives a rework, a connector press or a thermal cycle without the pad lifting, and it varies by a factor of five between laminate families — which is why a single acceptance limit for all high-frequency material is meaningless.

RT/duroid 5880 is specified at 5.5 N/mm (about 31 pli). RO4003C is 1.05 N/mm (6 pli), RO4350B 0.88 N/mm (5 pli), and Rogers 6035HTC 1.38 N/mm (7.9 pli). The PTFE figure is not a property of PTFE: it is the result of sodium-naphthalene or plasma treatment before plating, without which the same laminate measures below 1 N/mm. A supplier quoting 5 N/mm on a PTFE board is describing the treatment, not the resin.

The test also has a time dimension that a single value hides. Peel strength falls after reflow and falls further after ageing; the acceptance criterion for a board that will be reworked twice should be the value after three reflow passes, not the incoming certificate. If the specification says “1 N/mm” and the measurement is made on virgin material, it is testing the wrong object.

Surface Finish: Thickness and Solderability, Not “Smoothness”

The original checklist asked for surface flatness, thickness uniformity and freedom from oxidation, using a microscope and a flatness gauge. Flatness is not an acceptance criterion for a finish; thickness and solderability are. The measurable quantities are the coating thickness by XRF confirmed by micro-section, the solderability by wetting balance to J-STD-002, and — on a high-frequency board — the loss penalty the finish imposes.

FinishThicknessEffect at 10 GHzWhere it belongs
OSP0.2–0.5 µmLowest; organic, no metal in the skinLoss-critical RF, fine pitch
Immersion silver0.1–0.25 µmSecond lowest; thin silver, no nickel barrierHigh-frequency, high-volume
ENIGNi 3.0–6.0, Au 0.05–0.10 µmNickel dominates; Rs about 2.04× copperMixed-signal, wire bond, fine pitch
Immersion tinabout 1.0 µmIntermediate; press-fit and reflowConnector press-fit areas

The reason ENIG costs loss is arithmetic rather than chemistry. At 10 GHz the skin depth in copper is 0.65 µm and in nickel 1.33 µm; a 3 µm nickel layer is more than two skin depths thick, so the current runs in nickel and the surface resistance doubles. On a 0.18 dB/in laminate that is a visible fraction of the total, and it is why the finish is chosen with the loss budget rather than after it. The full comparison of processes is in surface treatment processes in PCB manufacturing.

Cleanliness and Insulation Resistance: the Test Nobody Runs

Ionic contamination is measured by extracting the residues from a board into a known volume of a water–isopropanol mixture and measuring the conductivity, reported as micrograms of sodium chloride equivalent per square centimetre. The acceptance limit is 1.56 µg/cm² per IPC-TM-650 2.3.25, and it is the same limit for a high-frequency board as for any other — but the consequences are worse, because a conductive residue on a 0.1 mm gap is a leakage path across a resonator, and moisture uptake of 0.02 % in a PTFE laminate does not dilute it.

Surface insulation resistance is the slower companion test: a test pattern is biased at 100 V in 85 °C / 85 % RH for 500 to 1000 h while the resistance is logged. It is the only test on this checklist that addresses electrochemical migration and CAF, and it is almost never specified on an RF board because it takes a month. Where it matters — medical, automotive under-hood, anything with a fine-pitch via in a humid environment — it is worth the month.

Warpage, Hole Wall and Cross-Sections: the Destructive Set

Three measurements cannot be made non-destructively and are therefore sampled: warpage, plated thickness on the barrel, and the cross-section of the hole. Warpage for a surface-mount board is held to 0.75 % of the diagonal by IPC-6012, measured on a granite plate with the board unrestrained; a 200 mm board therefore has 1.5 mm of allowance, and a 0.2 mm thin high-frequency core will use most of it after reflow.

The cross-section is the only direct evidence of plating quality. A 25 µm nominal plating on the barrel gives a finished hole 50 µm smaller than the drill, so a 0.3 mm drill finishes at 0.25 mm and the tolerance on the drill, not on the plating, sets the finished diameter. The section is read for average thickness, for voids (5 % of the barrel area is the usual allowance), for etchback or resin smear at the inner-layer interface, and for layer-to-layer registration, which on a high-frequency multilayer is held to about ±25 µm.

Sampling is where cost lives. One coupon per layer per lot for TDR, one test board per lot for the destructive set, and the destructive samples taken from a production panel rather than a dedicated test panel — a separate panel is processed with more care than the product and flatters the result. The plating side of this is covered in plating in high-frequency PCB manufacturing.

What a Bare Board Test Cannot Prove

Four items on the original list are not board-level tests, and no amount of instrumentation will make them so. Gain, harmonic distortion and noise figure require an active device. Radiated emissions require a complete product in its enclosure. Thermal resistance in K/W is a calculation from thickness and conductivity, not a measurement. Long-term reliability is an extrapolation from accelerated tests and a physical model, not a result.

The distinction matters commercially as much as technically. A test report that lists a gain measurement on an unpopulated board is not conservative, it is wrong, and it invites a customer to accept a board on the strength of a number that describes something else. The honest checklist stops where the object stops: the board can be shown to be dimensionally, electrically and materially what was ordered, and nothing beyond that.

Sequencing the Plan: Order, Sampling and Cost

The order that costs least is: coupon electrical tests first, because they are non-destructive and can be repeated; finished-board electrical test second, because it catches artwork and process escapes that the coupon never sees; destructive sampling last, because each sample is consumed. Running the cross-section first is the most common sequencing error, and it removes the evidence before the electrical number exists.

The cost is dominated by the destructive set, not by the electrical measurements. A TDR coupon takes minutes and can be run on every lot; a micro-section takes hours of skilled work and is run on a sample. That asymmetry is why the sampling plan should be written down before the first lot, with the sample size tied to the process capability rather than to habit — a process running at Cpk 1.3 can be sampled less often than one at Cpk 0.9, and the checklist should say which it is.

Finally, every window in this article assumes the coupon is built on the same stack, the same panel and the same process as the product. A coupon from a different panel is a different experiment, and reporting it as a product result is the one error that no amount of good instrumentation recovers from. The comparison of which requirements change when the board is not FR-4 is in high-frequency PCB and standard PCB differences.

Conclusion

A high frequency PCB testing checklist earns its place when every line carries a number, and the numbers here are deliberately material-specific: ±10 % impedance with a 7 % budget already spent, 0.18 dB/in at 10 GHz on RO4350B against 0.035 on 5880, 82 µm of z-axis expansion on PTFE against 11 on hydrocarbon, 1.05 N/mm of peel on RO4003C against 5.5 on 5880, Ni 3.0–6.0 µm under 0.05 µm of gold.

The four tests the original article was missing are the four that cost the most to run and reveal the most: the micro-section, the ionic extraction, the SIR soak and the thermal cycle to 100 passes. A board can pass every coupon measurement and still fail all four, and the checklist is incomplete until they are on it.

Frequently Asked Questions

How many tests does a high-frequency PCB actually need before shipment?
Ten non-destructive or sampled tests cover a bare board: impedance on a coupon, differential impedance, insertion loss, return loss, crosstalk, thermal shock, z-axis CTE, peel strength, coating thickness and ionic contamination, plus warpage and a hole-wall cross-section on a sample. Anything beyond that involves an active device or a complete product and cannot be claimed from a bare board.
Can an impedance coupon represent the product board?
Only if the coupon is built on the same stack, the same panel and the same press cycle. A 35 ps TDR resolves about 3 mm on a Dk 2.67 line, so a 50 mm coupon reports an averaged impedance and cannot see a via antipad or a connector launch. Coupon data proves the stack; only a TDR on the finished board shows the launches.
Why can the ±10 % impedance window be used up before measurement?
Because it is a budget. Dielectric thickness at ±10 % contributes about ±6 %, a ±25 µm etch tolerance about ±2.8 % on a 0.254 mm stack, solder mask −1 to −3 Ω and Dk lot spread ±0.7 %. Combined by root sum square that is ±7 %, leaving 3 % for coupling and coupon edge effects.
Why is gain not measured on a high-frequency PCB?
Gain is a property of an amplifier, and a bare laminate is passive. The measurable board-level quantities are S21 insertion loss in dB per inch and S11 return loss. Harmonic distortion likewise needs a non-linear device, so a passive coupon will show none of it.
Is 100 thermal cycles enough to qualify a PTFE board?
It qualifies the plated barrel against the z-axis expansion of the laminate, which is 82 µm over 25–260 °C on a 1.6 mm RT/duroid 5880 stack against 11 µm on RO4350B. It is not a life prediction. Because the failure depends on aspect ratio, PTFE is limited to about 6:1 and the plating should be specified at 20 % elongation or better.
Does ENIG hurt high-frequency loss?
It can. At 10 GHz the skin depth is 0.65 µm in copper and 1.33 µm in nickel, so a 3–6 µm nickel layer carries the current and has roughly 2.04 times the surface resistance of copper. OSP and immersion silver avoid the nickel and are the right finishes when the loss budget is tight.
RF & Microwave PCB Fabrication

Need the Coupons Measured Before the Panel Ships?

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

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