Electrical Characteristics in High-Frequency PCBs: What to Measure and How

Electrical characteristics in high-frequency PCB manufacturing are usually presented as a list of material properties, which is where the trouble starts: a list cannot tell you how a number was obtained, at what frequency, or what it is blind to. Dk quoted from a resonator at 10 GHz and Dk quoted from a clamped stripline at 1 MHz can differ by more than the tolerance a design is built on. This article takes the seven quantities that decide whether a high-frequency board works — Dk, Df, characteristic impedance, insertion loss, dielectric strength, insulation resistance and the temperature drift of Dk — and sets out for each one the method, the acceptance criterion, and the failure the measurement cannot see. Two entries that appear in most such lists are deliberately moved out, with reasons.

Electrical Characteristics in High-Frequency PCBs

A List Is Not a Specification

The earlier version of this page listed seven properties under the heading “electrical performance” and seven more under “functions”, gave no number for any of them, and included one entry that is not an electrical characteristic at all. That structure is common and it is not harmless: a designer who reads “low Dk reduces loss” will specify a low-Dk laminate to fix a loss problem that is caused by Df, and a designer who reads “CTE is an electrical characteristic” will look for CTE on the wrong drawing.

The useful version of the list answers four questions per line: what is the quantity, how is it measured, at what frequency or condition, and what does the measurement not see. The table below is that list, and the rest of the article takes each row in turn.

QuantityMethodWhat it controlsWhat the measurement cannot see
DkSplit-post dielectric resonator at 10 GHz; clamped stripline for thin coresWidth for a given impedance, guided wavelength, delayThe anisotropy of the material, and the difference between design and process values
DfSame fixture as Dk, same frequencyDielectric loss per unit of electrical lengthLoss in the copper, the finish, or the mask
Z0TDR on a coupon, with the reference plane definedReflections, return lossLoss; a line can be 50 Ω and unusable
Insertion lossVNA, TRL-calibrated, de-embeddedEverything, in one numberWhich term is responsible without a second measurement
Dielectric strengthASTM D149 on the laminateVoltage withstand of the dielectricDefects, voids, and the effect of a PTH barrel through it
Insulation resistanceIPC-TM-650 2.5.17, before and after humidityLeakage between netsElectrochemical failure that appears only under bias
TCDkResonator over temperaturePhase drift over the operating rangeMoisture uptake, which shifts Dk the other way

Two entries that most lists include are missing on purpose: coefficient of thermal expansion, which is a dimensional property whose electrical consequence is a plated-barrel failure (section 6), and electromagnetic compatibility, which is a property of the layout and the enclosure rather than of the laminate (section 8). Keeping them in an electrical table is how a specification ends up asking a material to do a job the geometry has to do.

Dk: Two Numbers, Ten Per Cent Apart

Dk is not one number per material. Laminates are published with a process value — what the laminate measures, with tolerance, in the fixture the supplier qualifies — and often with a design value, which is derived from a different method and is what a field solver expects. For most materials the two are close. For some they are not.

LaminateProcess DkDesign DkDfWhich value to use where
RO30033.00 ± 0.043.000.0010Interchangeable
RO4003C3.38 ± 0.053.550.0027Design value for simulation; process value for acceptance
RO4350B3.48 ± 0.053.660.0037Same: a 5 % gap
RO301010.2 ± 0.3011.200.0022A 10 % gap — using the wrong one puts the impedance 5 % out
RO321010.2 ± 0.5010.80.0027Glass-reinforced; the tighter tolerance is the reason to pay for it
RT/duroid 58802.20 ± 0.022.200.0009Interchangeable, and the tightest tolerance of the group
TLY-52.20 ± 0.022.200.0009PTFE with microglass; same electrical target as 5880
RF-303.00 ± 0.043.000.0014 at 1.9 GHzNote the frequency: not a 10 GHz figure

The 3010 row is the one worth remembering. Its design value of 11.20 against a process value of 10.2 is a 10 % difference, and because impedance scales roughly as 1/√Dk at fixed geometry, choosing the wrong one moves a 50 Ω line by about 5 % — half the usual acceptance window, before any fabrication tolerance is added. The rule is unglamorous but effective: simulate with the design value, accept on the process value, and state on the drawing which one the coupon is being judged against.

The second thing the table does not show is frequency. Dk falls slowly with frequency in every polymer system, and a value quoted at 1 MHz is not the value at 10 GHz. RF-30 illustrates the honesty problem: its Df is quoted at 1.9 GHz, which is a legitimate number but not comparable with a 10 GHz figure from another supplier. Comparing two datasheets without checking the measurement frequency is the most common analytical error in laminate selection.

Df and the Only Loss Formula Worth Memorising

Dielectric loss in dB per inch follows αd ≈ 2.6 · fGHz · √Dk · Df. The formula is worth memorising because it settles two arguments at once: loss rises linearly with frequency, and it rises with the square root of Dk but linearly with Df. That asymmetry is why a high-Dk material can be a low-loss material.

LaminateDk / Df1 GHz10 GHz28 GHzLine width for 50 Ω at 0.254 mm
FR-44.3 / 0.0200.11 dB/in1.08 dB/in3.02 dB/in0.48 mm
RO4350B3.48 / 0.00370.018 dB/in0.18 dB/in0.50 dB/in0.60 mm
RO4003C3.38 / 0.00270.013 dB/in0.13 dB/in0.36 dB/in0.59 mm
RO301010.2 / 0.00220.018 dB/in0.18 dB/in0.51 dB/in0.20 mm
RT/duroid 58802.20 / 0.00090.0035 dB/in0.035 dB/in0.097 dB/in0.79 mm

The 3010 row is the argument made numerically: at 10 GHz its dielectric loss per inch is the same as 4350B’s, because the larger √Dk is offset by the smaller Df, yet the 50 Ω line is a third of the width. High Dk buys area, not loss. Stated per unit of electrical length the rule is sharper still: two materials with the same Df lose the same amount over the same number of wavelengths, whatever their Dk, because the shorter guided wavelength of the high-Dk material cancels its higher per-inch loss.

This is the error behind the sentence “a lower Dk enables reduced loss”, which appears in the earlier version of this page. Dk sets delay and width; Df sets loss. The two travel together in commercial materials, which is why the confusion survives, but the counter-example above is a production material and not a thought experiment.

Characteristic Impedance Is Measured on a Coupon, Not Calculated

A field solver gives the impedance of an idealised cross-section. The board delivers something else, and the difference is the sum of four controllable errors. A coupon measures the sum.

Source of errorMagnitudeEffect on Z0Control
Dielectric thickness tolerance±10 % on 0.254 mm±6 %The dominant term; specify prepreg build, not just total
Etch tolerance±25 µm±2.8 % at 0.254 mm; ±5.5 % at 0.127 mmEtch compensation in CAM; measure finished width
Solder mask25–40 µm over the trace−1 to −3 ΩJudge coupons with mask if the product is masked
Forgotten bond ply0.10 mm of prepreg50 → 61 Ω (+22 %)Count every layer in the stack, including the one that only glues
Void in the bond10 µm+2.3 ΩPress cycle; void inspection on thick stacks

Two practices follow. First, a coupon is only meaningful if it is built from the same stack, the same foil and on the same panel as the product; a coupon on the border of a different design proves nothing. Second, the measurement has to state its reference plane: a TDR launched through a connector with 1 mm of uncontrolled line moves the first 100 ps and can shift a reading by several ohms, which is why the acceptance is defined at a de-embedded reference plane and not at the connector. The fabrication side of impedance control is set out in impedance control in high-frequency PCB manufacturing.

Insertion Loss: One Number That Contains Everything

Insertion loss is the only measured quantity that includes the dielectric, the copper, the finish and the mask, and it is therefore the only one that can settle a dispute about a link that does not work. It is also the easiest to measure badly. Three conditions decide whether the number means anything.

Calibration. A TRL calibration moves the reference plane to a defined point on the coupon; a response calibration does not, and typically reports 0.2–0.5 dB of fixture loss as though it were board loss. On a 100 mm line at 10 GHz that difference is larger than the material effect being investigated. Coupon geometry. The line must be long enough for the loss to exceed the measurement noise — 20 wavelengths or more is the usual rule — and wide enough that the launch discontinuity is a small fraction of the total. De-embedding. A 2×-thru structure is now standard and removes the launch; without it, two labs will disagree by 10 % on the same board.

The number that comes out should be compared with the two-term budget from section 3 plus the copper term. If the measured total is 0.32 dB/in on a 0.254 mm RO4003C microstrip at 10 GHz, then 0.13 is dielectric and about 0.19 is copper — which is the check that tells you whether the foil grade in the stack is the one that was ordered, and is the reason copper foil roughness and Df have to be specified together rather than separately. The process side of the measurement is in testing steps in high-frequency PCB production.

CTE: An Electrical Failure With a Mechanical Cause

Coefficient of thermal expansion appears in most electrical-characteristics lists and does not belong there. It is a dimensional property: parts per million per degree Celsius, measured by thermomechanical analysis, and it changes nothing about how a signal propagates. What it changes is whether the plated barrel of a through hole survives a reflow, and the mechanism is worth stating because the numbers are large.

LaminateCTE, x / y / z (ppm/°C)Z-axis growth, 1.6 mm over 215 °CConsequence
FR-4, Tg 17017 / 17 / 4617–24 µmBarrel strain within the ductility of standard plating
RO4003C11 / 14 / 46≈ 11 µm (4350B reference)Lower in-plane expansion, similar z
RO301013 / 11 / 16Lowest of the groupAll three axes below copper; the ceramic filler does the work
TLY-526 / 15 / 217≈ 75 µmNeeds high-elongation plating and a relaxed aspect ratio
RT/duroid 588031 / 48 / 237≈ 82 µmThe extreme case; plating ductility is the design constraint

Read the last column rather than the third: a PTFE board expands 7–8 times as much through its thickness as a hydrocarbon one, and the plated barrel has to stretch to follow. That is a mechanical requirement on the plating — 20 % elongation or better — and a limit on aspect ratio, not a term in the signal budget. The electrical number that captures the same physics more usefully is the temperature coefficient of Dk, which is section 9. Glass transition is a separate question again, covered in glass transition temperature in PCB manufacturing.

Dielectric Strength and Insulation Resistance

These two are the safety and leakage entries, and both are specified on the laminate datasheet by methods that say little about a finished board. Dielectric strength is quoted in kV/mm from ASTM D149 on a specimen of the base material; on a finished board the relevant geometry is a 0.1 mm dielectric between two layers, where a single void, a scratch or a drilled hole edge reduces the withstand far below the bulk value. The design response is spacing rules and layer-to-layer creepage, not a material upgrade.

Insulation resistance is quoted as volume resistivity in MΩ·cm, measured dry. What fails in the field is not the dry value but the wet one: ionic contamination from poor rinsing, flux residue under a component, or a bias applied across a humid surface for a thousand hours. These are electrochemical failures, and the tests that predict them are biased humidity tests and CAF resistance, not a resistivity reading at incoming inspection. The same reasoning applies to surface treatment residues and to the solder mask, which is why the process articles carry the burden here rather than the material table.

EMC Is a Property of the Geometry, Not of the Laminate

Listing electromagnetic compatibility as an electrical characteristic of a PCB material is a category error that leads to real decisions — typically, buying a lower-Df laminate to fix a radiation problem. Emissions are driven by loop area, by the impedance of the return path and by common-mode current on cables. A laminate cannot shield: it is a dielectric between two conductors, and the conductors are what radiate.

What the board contributes is the return path. A continuous reference plane under a trace, a via fence along a microstrip that leaves the ground region, and a stack in which every signal layer is adjacent to a plane are the three measures that move emissions by tens of decibels. Their cost is layer count and layout discipline, not material. The one place where a material property enters is loss, and loss only reduces the amplitude that reaches the radiating structure — a second-order effect compared with loop geometry.

The Characteristic Most Lists Leave Out: TCDk

A board that meets its impedance target at 25 °C and drifts in phase at 85 °C has an electrical characteristic nobody specified. Dk is a function of temperature, and the coefficient varies in both sign and magnitude across the laminate families — from −425 ppm/°C on a high-Dk ceramic-filled PTFE to −125 on 5880 and +50 on 4350B.

LaminateTCDkDk change over 100 °CPhase drift, 100 mm at 10 GHz
RO4003C+40 ppm/°C+0.4 %≈ 3°
RO4350B+50 ppm/°C+0.5 %≈ 4°
RO6035HTC−66 ppm/°C−0.66 %≈ 5°
RT/duroid 5880−125 ppm/°C−1.25 %≈ 10°
RO6010.2LM−425 ppm/°C−4.25 %≈ 35°

The arithmetic behind the last column: phase over a fixed length scales with √Dk, so a fractional Dk change produces half as much fractional phase change, and 100 mm at 10 GHz on a Dk 2.2 laminate is about 1,644° of electrical length. Ten degrees of drift over a 100 °C excursion is enough to break a phased array’s beam pointing unless it is compensated; 35° is not a drift but a redesign. Note also the sign: a negative coefficient and a positive one cannot be interchanged in a stack that has to hold phase, even if their Dk at room temperature is identical.

Coupon Design Decides Whether Any of This Is Measurable

Every number in this article comes off a coupon, and the coupon is where most programmes fail. Four rules cover the common cases. The coupon must be built on the production panel, with the same stack, foil and finish. The line must be long enough that the quantity being measured is larger than the uncertainty — for loss, 20 wavelengths; for impedance, long enough to separate launch from line. The reference plane must be defined and the launch de-embedded, because a connector transition is not part of the product. And the coupon should exist in more than one place on the panel, because etch rate and dielectric thickness vary across 600 mm, and a single reading in the middle is a sample of one.

The last rule matters more on large panels. A 600×1200 mm sheet etched in a spray chamber will not deliver the same finished width at the centre and the edge; the difference is process, not error, and it has to be inside the tolerance rather than inside the argument. Position the impedance coupon in a copper-balanced area, and report centre and edge separately.

What to Write on the Drawing

Six lines convert this article into a purchase specification. 1. Dk basis, design value for simulation and process value for acceptance, with the measurement frequency. 2. Df limit and frequency, since a 1.9 GHz figure and a 10 GHz figure are not comparable. 3. Impedance target per layer, with tolerance and coupon location. 4. Insertion loss budget at a named frequency, where the design can support it — it is the only number that includes the finish. 5. TCDk or a phase-drift limit for anything that has to hold phase over temperature. 6. The finish, because at 10 GHz a 3–6 µm nickel layer matters more than the copper under it.

A drawing with those six is testable. One with a list of adjectives is not, and the difference shows up as a first-article argument rather than as a working product.

Conclusion

Electrical characteristics in high-frequency PCB manufacturing are worth listing only when each entry carries a method, a frequency and an acceptance value. Dk has two published values that can be 10 % apart, so the drawing has to say which one is being judged; Df rather than Dk sets loss, and the formula αd = 2.6 f √Dk Df shows why a Dk 10.2 material can lose no more per inch than a Dk 3.48 one; impedance is the sum of a ±6 % thickness term, a ±2.8 % etch term and a mask offset, measured on a coupon from the same panel; and insertion loss is the only quantity that includes the finish, which at 10 GHz is the dominant conductor term.

Two entries belong elsewhere. CTE is a dimensional property whose electrical consequence is a strain in the plated barrel — 82 µm of z-axis growth on a 1.6 mm PTFE board against 11 µm on a hydrocarbon one — and EMC is decided by loop area and return path rather than by any laminate. Removing them from the electrical list is not pedantry: it points each problem at the discipline that can actually solve it, which for the CTE problem is plating ductility and for the EMC problem is stack design.

XCE PCB measures all seven quantities on production coupons rather than quoting them: 2 to 72 layers, impedance ±10 % with TDR coupons at centre and edge, panels to 600×1200 mm, samples in as fast as 24 hours, inspected to IPC-A-600G and IPC-6015. The physics behind the numbers, rather than the measurements, is in concepts of electricity in PCB, and the stripline and microstrip geometries that turn a Dk into a width are in microstrip and stripline design.

Frequently Asked Questions

Which Dk value should be used in a simulation?
The design value where one is published. RO3010 is the clearest case: 10.2 ± 0.30 is the process value that acceptance is judged against, while 11.20 is the design value the field solver expects. Using the process value in simulation gives a wider 50 Ω line than the board needs and shifts impedance by about 5 %, half the usual window. RO4003C and RO4350B have a smaller but still real gap, 3.38 against 3.55 and 3.48 against 3.66.
Does a lower Dk reduce insertion loss?
Not by itself. Dielectric loss is 2.6 f √Dk Df dB/in, so loss is linear in Df and only square-root in Dk. RO3010 with Dk 10.2 and Df 0.0022 has the same 10 GHz dielectric loss per inch as RO4350B with Dk 3.48 and Df 0.0037 — about 0.18 dB/in — while its 50 Ω line is a third of the width. Per unit of electrical length, two materials with the same Df lose the same amount whatever their Dk.
Why is CTE not an electrical characteristic?
Because it does not change how a signal propagates. CTE is a dimensional property measured by TMA, and its electrical consequence is indirect: a PTFE laminate expands 237 ppm/°C through the thickness, so a 1.6 mm board grows about 82 µm across a 215 °C reflow, and the plated barrel has to stretch to follow. That is a requirement on plating ductility and aspect ratio, not a term in the loss or impedance budget.
How is insertion loss measured on a high-frequency coupon?
With a VNA using a TRL calibration so the reference plane sits at a defined point on the coupon, on a line at least 20 wavelengths long at the highest frequency of interest, with the launch removed by 2×-thru de-embedding. Without de-embedding two labs can disagree by 10 % on the same board, and a response calibration typically reports 0.2–0.5 dB of fixture loss as if it were board loss.
What impedance tolerance can actually be held?
The sum of the errors decides it. Dielectric thickness at ±10 % moves a 50 Ω line by ±6 % and is the dominant term; etch at ±25 µm adds ±2.8 % on a 0.254 mm stack and ±5.5 % on 0.127 mm; solder mask pulls 1–3 Ω off. ±10 % overall is achievable when each term is controlled, which is why a stack drawing should specify prepreg build per layer rather than only total thickness.
Why does TCDk matter for radar and phased arrays?
Because phase over a fixed length scales with √Dk, so a temperature coefficient becomes a beam-pointing error. Over a 100 °C excursion, 100 mm of line at 10 GHz drifts about 3° on RO4003C, 10° on 5880 and 35° on RO6010.2LM. A positive and a negative coefficient cannot be substituted for each other in a stack that has to hold phase, even when their room-temperature Dk is identical.
RF & Microwave PCB Fabrication

Need the Numbers Proven on a Coupon, Not Quoted from a Datasheet?

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

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