The published version of this article listed dielectric constant, dissipation factor, impedance, CTE, voltage withstand, EMC and conductivity, and called the list “concepts of electricity”. Seven of those eight are real quantities, but they are answers, not concepts — and one of them, the coefficient of thermal expansion, is not an electrical property at all. The concepts underneath are few and can be stated in a paragraph: charge is conserved, so current flows in closed loops; the loop, not the trace, is the circuit element; a conductor’s cross-section sets its resistance and its shape sets its inductance; the electric field that carries the signal lives in the dielectric; and above a few hundred kilohertz a board is no longer electrically small. Every number in this article follows from those five statements, including the ones that decide whether a 10 GHz link survives the laminate.
The List Was of Answers, Not Concepts
A parameter list is easy to publish and hard to use, because nothing on it tells a designer which number to change when a link fails. Sorting the original eight items by what they physically are makes the gap obvious.
| Item as listed | What it physically is | Where it belongs |
|---|---|---|
| Dielectric constant (Dk) | A property of the field in the laminate: how much energy the polarisation stores per volt, which sets phase velocity | Field, not current |
| Dissipation factor (Df) | The fraction of that stored energy lost per radian | Field, not current |
| Characteristic impedance (Z0) | A ratio of voltage to current for a wave, fixed by geometry and Dk, dissipating nothing | Transmission line |
| CTE | Thermal expansion of the laminate — a mechanical property that strains plated holes | Not electrical at all |
| Voltage withstand | Breakdown of the dielectric or of the air gap, a field-strength limit | Insulation |
| EMC | The far field radiated by the currents that were not cancelled in a loop | A consequence |
| Conductivity | Free-electron density and mobility in the copper | Conduction |
| Signal integrity, reliability, power management … | Outcomes of the seven items above, not separate properties | Results |
Two of those rows are worth arguing about, because they are where most published explanations go wrong. CTE appears in almost every Chinese-language “PCB electrical performance” article, and it has no place there: it matters because a plated hole is a copper cylinder in a laminate that expands five to ten times faster, which is a mechanical fatigue problem, covered in glass transition temperature and Z-axis expansion. And EMC is not a property to be improved alongside others; it is what the loop geometry does to the current that the other parameters allowed.
Current Flows in a Loop, and the Board Is Part of It
Charge is conserved. A driver pushes electrons into a trace, and the same number must leave the far end and return to the driver through whatever conductor is available — a plane, a second trace, a cable shield, or the chassis. There is no sink called ground into which charge disappears; “ground” is the name of the return conductor, and it is as much a part of the circuit as the trace.
This sounds like pedantry until a board is laid out as if it were not true. A signal crossing a plane split has a return path that must detour around the gap; the loop area grows from a fraction of a square millimetre to tens of square millimetres, inductance rises with it, and the link fails at a frequency nobody predicted from the schematic. The geometry that matters is the loop, and the loop includes both layers.
One order of magnitude is worth keeping in mind. A 1 A signal at 1 GHz moves 1 nC per cycle, and that charge makes the round trip through the plane every 1 ns. The plane is not a static reference; it is a conductor carrying the same current as the trace, at the same frequency, in the opposite direction.
Where the Return Current Goes: Least Resistance, Then Least Inductance
Return current spreads out at DC and concentrates under the trace at high frequency, and the crossover is not a mystery — it is the frequency at which the inductive reactance of the return path exceeds its resistance. Setting 2πfL equal to R gives f = R/(2πL), and for traces on 1 oz copper the answer lands in the hundreds of kilohertz.
| Trace width (1 oz) | Resistance | Inductance (approx.) | Crossover f = R/(2πL) |
|---|---|---|---|
| 0.2 mm | 2.46 mΩ/mm | 0.28 nH/mm | 1.4 MHz |
| 0.5 mm | 0.98 mΩ/mm | 0.28 nH/mm | 0.56 MHz |
| 1.0 mm | 0.49 mΩ/mm | 0.28 nH/mm | 0.28 MHz |
The practical reading: below a few hundred kilohertz the return current uses the whole plane, because spreading out lowers resistance; above about 1 MHz it has already moved to the strip directly beneath the trace, because that is the path of least inductance. By 10 GHz roughly 80 % of it is within three dielectric thicknesses of the trace centreline, and on a 0.254 mm stack that is a band about 0.76 mm wide. A plane with a slot in that band is not a plane for that signal.
There is a second consequence that decides when the word “ground” stops being usable. In a dielectric of Dk 3.38 the guided wavelength is 184 mm at 1 GHz and 18.4 mm at 10 GHz. A 100 mm line at 10 GHz is 5.4 wavelengths long, so the potential of the reference plane differs from one end to the other by design, and a probe clipped to two points of the same plane will measure a voltage between them. The plane is a transmission line, not an equipotential.
Resistance: One Number That Does Not Care How Wide You Draw It
Copper’s resistivity is 1.72 × 10−8 Ω·m, which for 1 oz (35 µm) foil gives a sheet resistance of 0.491 mΩ per square. The square is the useful unit because a square of copper has the same resistance whatever its size: a 1 mm wide, 1 mm long square and a 10 mm wide, 10 mm long square are both 0.49 mΩ. Resistance then scales with the number of squares in series, which is length divided by width.
| Copper weight | Nominal thickness | Sheet resistance at 20 °C | Resistance of a 0.2 mm wide, 50 mm long trace |
|---|---|---|---|
| 0.5 oz | 17.5 µm | 0.98 mΩ/sq | 0.25 Ω |
| 1 oz | 35 µm | 0.49 mΩ/sq | 0.12 Ω |
| 2 oz | 70 µm | 0.25 mΩ/sq | 0.06 Ω |
Two corrections are needed before that table is used for anything real. The first is temperature: copper’s coefficient is 0.00393 per K, so a trace at 100 °C carries about 30 % more resistance than the same trace at 25 °C, and it is the hot resistance that sets voltage drop and Joule heating. The second is that plated outer layers are not 1 oz: the base foil is 1 oz but pattern plating adds 20–25 µm, so finished outer copper is 50–60 µm and the resistance is correspondingly lower than the table. The current-carrying tables behind those numbers, and the derating that goes with them, are in copper thickness in high-frequency PCBs.
Skin Effect: When the Copper Stops Being 35 µm Thick
Alternating current crowds toward the surface of a conductor, and the depth at which the current density has fallen to 1/e of its surface value is the skin depth. For copper it follows δ ≈ 66 µm/√(f in MHz).
| Frequency | Skin depth in copper | 1 oz copper, thickness ÷ skin depth |
|---|---|---|
| 1 MHz | 66 µm | 0.5 — thinner than the copper, no crowding yet |
| 10 MHz | 20.9 µm | 1.7 |
| 100 MHz | 6.6 µm | 5.3 |
| 1 GHz | 2.09 µm | 17 |
| 10 GHz | 0.66 µm | 53 |
| 28 GHz | 0.39 µm | 90 |
The consequence for a fabricator is blunt, and it is the single most common misunderstanding in copper-weight discussions: at 10 GHz the current occupies the outer 0.66 µm of a 35 µm foil, so the effective resistance is roughly 30–50 times the DC value, and doubling the copper to 2 oz changes the conductor loss by very little — well under 10 % once roughness is included. What does change it is width, because the current spreads sideways along the surface, and surface treatment, because a 3–6 µm nickel layer is several skin depths thick and carries the current in a metal with twice the resistivity.
Surface roughness belongs in the same paragraph. At 0.66 µm skin depth, a foil with 1–2 µm peaks forces the current to follow a longer path, and measured insertion loss rises 20–50 % against smooth foil on the same laminate. This is not a detail to leave to the laminate supplier; it is the reason the copper profile has to be specified along with the Dk.
Impedance Is Geometry, and It Dissipates Nothing
The original article described characteristic impedance as “the impedance encountered by a signal travelling on a guide wire”, which is close enough to be dangerous, because it invites the reading that a 50 Ω trace resists current like a 50 Ω resistor. It does not. For a wave, Z0 = √(L/C), where L and C are the inductance and capacitance per unit length of the cross-section; a wave entering the line sees that ratio, and if the line is terminated in the same value, no energy comes back.
For a 50 Ω microstrip on a 0.254 mm laminate of Dk 3.38 (effective Dk 2.67) the numbers are 272 nH/m of inductance and 109 pF/m of capacitance. A metre of that trace has 0.49 Ω of resistance and carries a signal whose impedance is 50 Ω; the two are not the same quantity and do not add. What impedance does is set the voltage-to-current ratio of the wave, and what a discontinuity in it does is reflect a fraction of the wave, given by (Z1−Z2)/(Z1+Z2). That is why a 10 % impedance step costs about 0.1 dB of return loss — harmless — while a 25 % step costs about 1 dB, which is where link budgets begin to notice.
Because Z0 is a ratio of L to C, it is controlled by geometry: trace width, dielectric thickness, and the distance to the reference plane. That is why the fabricator’s lever is the stack-up, and why the tolerances that matter are the ones on laminate thickness and etch, not on copper purity. The build side of it, including the ±10 % window and what moves impedance in production, is in impedance control in high-frequency PCB manufacturing, and the layer geometry in stack-up design.
The Dielectric Carries the Signal Current Too
No conduction current flows through the laminate, and yet the signal current in the plane is connected to the current in the trace through the dielectric. Between the two conductors the field is time-varying, so there is a displacement current density J = ∂D/∂t in the dielectric, and its integral over the cross-section equals the signal current. The dielectric is not a passive gap; it is the medium in which the field, and therefore the energy, travels.
Two consequences follow directly. The phase velocity is v = c/√εeff, so Dk sets electrical length: at Dk 3.38 the guided wavelength is 184 mm at 1 GHz and 18.4 mm at 10 GHz, and a 100 mm line is 5.4 wavelengths at the upper frequency. And Df sets how much of the energy in that field is lost per unit of electrical length, which is why loss is quoted in dB per inch at a frequency rather than as a material constant. A low-Dk material shortens the line for the same delay, but it does not by itself reduce the loss per unit delay — only Df does.
Voltage: Air Breaks Down at 3 kV/mm, and Spacing Comes From a Table
Dielectric breakdown is a field-strength limit: dry air at sea level holds about 3 kV/mm, laminate datasheets quote 20–40 kV/mm for FR-4 (measured on a thin specimen in oil, which is why the figure overstates what a humid board survives), and the practical limit is almost always the surface path rather than the bulk. IPC-2221 turns this into a table of minimum conductor spacing as a function of peak voltage and conductor location.
| Peak voltage between conductors | B1: inner layers | B2: outer, uncoated | B4: outer, polymer coated |
|---|---|---|---|
| ≤30 V | 0.05 mm | 0.10 mm | 0.05 mm |
| 31–50 V | 0.10 mm | 0.60 mm | 0.13 mm |
| 51–100 V | 0.10 mm | 0.60 mm | 0.13 mm |
| 101–150 V | 0.20 mm | 0.60 mm | 0.40 mm |
| 151–300 V | 0.20 mm | 1.25 mm | 0.40 mm |
| 301–500 V | 0.25 mm | 2.50 mm | 0.80 mm |
| >500 V | +0.0025 mm/V | +0.005 mm/V | +0.00305 mm/V |
Read as physics rather than as paperwork, the table says three things. Internal conductors need a fifth of the external spacing because they are sealed in laminate and cannot arc through air. The 31–50 V band is a step, not a slope: bare outer copper jumps from 0.10 mm to 0.60 mm, which is why a product that migrates from 24 V to 48 V keeps a layout that is no longer compliant, with nothing in the schematic to flag it. And the table is in peak volts, not RMS: 230 V AC peaks at 325 V and lands in the 301–500 V band, so it needs 2.5 mm, not the 1.25 mm the RMS figure suggests. Above 3050 m altitude the B2 column no longer applies, because the air is thinner.
EMC Is What the Loop Does to the Current You Could Not Cancel
Radiated emission from a board scales with loop area, current and frequency squared; the loop is the antenna, and the differential currents in a trace and its return cancel only to the extent that they are close together. This is the same geometry that decided the return path in section 3, seen from the outside, which is why “improving EMC” and “controlling the return path” are not two tasks.
Three numbers make the mechanism concrete. Doubling the dielectric thickness doubles the loop area of a microstrip and costs about 6 dB of emission at the same current. Routing a signal across a plane split multiplies the loop area by ten or more, which is 20 dB. And a cable leaving the board carries common-mode current driven by the board’s plane potential differences; that is why ferrites and chokes work on cables while the board itself looks clean on a near-field probe.
Which Parameters Are Electrical, and Which Only Look Like It
Most arguments about “electrical performance” are arguments between domains. Putting each parameter in its own column, with the article that covers it, removes the ambiguity.
| Parameter | Domain | What it actually governs | Covered in |
|---|---|---|---|
| Dk, Df | Field | Phase velocity, loss per unit delay | critical parameters |
| Z0 | Transmission line | Reflection at discontinuities | impedance designs and stacks |
| Copper thickness, roughness | Conduction | DC drop, conductor loss above 1 GHz | copper thickness |
| Surface finish | Conduction | 0.1–0.3 dB/in at 10–28 GHz through the barrier metal | surface treatment |
| Tg, Td, CTE | Thermo-mechanical | Hole-wall strain, delamination, warpage | Tg and Z-axis expansion |
| Spacing, creepage | Insulation | Breakdown and tracking | This article, section 8 |
| Return-path continuity | Both | Impedance discontinuities, emissions, crosstalk | what DRC cannot see |
The value of the split is diagnostic. A link that fails with a clean eye at the transmitter and a closed eye at the receiver, on a board whose impedance measures in tolerance, is not a Dk problem; it is usually a return path, a stub, or a connector launch. A board that passes every electrical test and fails thermal cycling is not an electrical problem at all.
Five Numbers to Put on a Drawing
Because the geometry is what governs the electricity, the drawing has to carry the geometry. Five items cover most of it, and each one is a number a fabricator can be held to:
1. Reference plane for every signal layer, stated per layer, including which plane is the return for which signal, and a note that no signal crosses a split. 2. Impedance target and tolerance with the test coupon geometry, for example 50 Ω ±10 % microstrip, 0.30 mm line on 0.254 mm laminate. 3. Copper weight per layer and the finished outer copper, because 1 oz base plus plating is not 1 oz. 4. Minimum spacing against the highest sustained voltage, quoted with the column of IPC-2221 that was used. 5. Surface finish and its thickness window, because above 10 GHz the finish is part of the conductor.
None of those five is a material “grade”. That is the point: a board is electrically specified by geometry and window, and material grades are how the geometry is achieved, not what it is.
Corrections to the Original Article
The published version contained six statements that would mislead a designer, and they are listed here rather than quietly removed, because the same six appear across the industry.
| Original claim | Correction |
|---|---|
| CTE is one of the PCB’s electrical properties | It is a mechanical property. It matters electrically only through hole-wall strain and layer-to-layer registration; the failure it causes is fatigue, not loss |
| Characteristic impedance is “the impedance a signal encounters on a trace” | Z0 is a ratio of L to C fixed by the cross-section, and it dissipates nothing. A 50 Ω line has 0.49 Ω/m of resistance, which is a different quantity and does not add to 50 |
| A lower Dk means faster transmission, therefore better | It means shorter electrical length and less delay, which is sometimes the goal and sometimes not. Loss per unit delay depends on Df, not on Dk |
| EMC is a performance parameter alongside the others | It is the radiated result of loop area and common-mode current. You do not add EMC; you remove loop area |
| Voltage withstand is a material property of the laminate | It is a field-strength limit of the whole path — bulk, surface and air gap — and the surface path usually fails first, especially in humidity and contamination |
| Better electrical performance improves productivity | Not a property of the board at all. What shortens a design cycle is a stack-up whose impedance, return paths and design rules were agreed before layout started |
Conclusion
Electricity in a PCB is one loop, one field and a handful of ratios. The current that leaves a driver returns through a plane, and which part of the plane it uses depends on frequency: the whole plane at DC, a strip under the trace above a few hundred kilohertz, and a band about three dielectric thicknesses wide at 10 GHz. The copper sets resistance through its cross-section and, above 1 GHz, through its surface — at 10 GHz the current uses 0.66 µm of a 35 µm foil, so more copper stops helping while more width and smoother foil keep helping. The field in the dielectric sets phase velocity and loss, which is why Dk and Df appear in every loss conversation. And the spacing table is a field-strength limit that steps at 31 V, not a formality.
XCE PCB builds high-frequency and mixed-dielectric boards to those terms: 2 to 72 layers, impedance ±10 %, panels to 600×1200 mm, samples in as fast as 24 hours, inspected to IPC-A-600G and IPC-6015. Coupons are measured on the finished panel, not on a test vehicle from another lot. If a link is failing and the impedance reads in tolerance, send the stack-up and the return-path plan — that is usually where the answer is. The fabrication sequence that produces the geometry is in the manufacturing overview, and the electrical measurements that verify it are in signal integrity in high-frequency PCBs.
Frequently Asked Questions
What are the basic electrical concepts in a PCB?
Does current return through the ground plane or the path of least resistance?
Why does 2 oz copper not reduce loss at 10 GHz?
Is characteristic impedance the same as resistance?
What spacing do I need between conductors at 48 V?
Why does the dielectric matter if no current flows through it?
Need a Stack-Up That Keeps Its Return Path?
Send us your Gerber files and requirements for a manufacturability review and a tailored quotation.
sales1@xcepcb.com
+86 180 3341 4657