Introduction and Function of Trace Width and Spacing in High-Frequency PCB Processing: Width Sets Impedance, Spacing Sets Coupling

Trace width and trace spacing are usually introduced together, as two numbers with the same job. They are not. Width is tied to the dielectric thickness by the impedance ratio, so on a controlled-impedance board it is not a free variable at all; spacing is the only one of the two that an engineer can change to control coupling, and it is the one that is usually specified as a rule of thumb rather than a number. This article separates the two, gives the coupling numbers against the spacing-to-height ratio, and explains why the familiar “3W” rule and the RF “3h” rule are not the same rule and cannot be substituted for one another.

Microstrip and Stripline Design - high-frequency PCB manufactured by XCE PCB

Two Numbers, Five Jobs

The introduction and function of trace width and spacing is usually written as though the two travel together: wider traces and wider gaps are “better”, narrower ones allow more density. In a controlled-impedance design that framing breaks immediately, because the two parameters are constrained in opposite ways. Width is fixed once the stack is chosen. Spacing is free, and it is the only geometric lever an engineer has over how much one trace talks to another.

ParameterWhat it actually setsWho controls itFreedom
WidthImpedance (through W/h), current capacity, minimum manufacturable lineThe stack, not the routerAlmost none on a controlled-impedance layer
SpacingCoupling and crosstalk, differential impedance, process floor, clearanceThe routerConstrained only by density and by the process floor

Read the right-hand column and the practical consequence is clear: a design review that asks “can we widen this trace” is asking about the laminate, while a review that asks “can we open this gap” is asking about the layout. The first has a cost, the second usually has none.

Width: A Ratio Disguised as a Dimension

For a microstrip, characteristic impedance depends on the ratio of width to dielectric height, not on width by itself. On a Dk 3.38 laminate the 50 Ω solution is W/h ≈ 2.33, so the width follows the dielectric thickness whether the designer wants it to or not.

Dielectric height50 Ω width (Dk 3.38)Edge-to-edge gap at s = 3hTrace pitchTraces per 10 mm
0.102 mm0.24 mm0.31 mm0.55 mm18
0.203 mm0.47 mm0.61 mm1.08 mm9
0.254 mm0.59 mm0.76 mm1.35 mm7
0.508 mm1.18 mm1.52 mm2.70 mm3

The last two columns are the ones that matter in a layout review, because they show the real cost of the 3h rule: on a 0.254 mm stack, honouring s = 3h means seven 50 Ω traces per 10 mm of board, and on a 0.508 mm stack it means three. Density and isolation are paid for from the same budget. Width itself, and the etched trapezoid that decides the finished width, is covered separately in route width in PCB manufacturing.

Spacing Function One: Coupling and Crosstalk

Coupled noise falls steeply with the ratio of gap to dielectric height, because the field that leaks from one trace to its neighbour is concentrated in the dielectric immediately under the trace. Doubling the gap does not halve the coupling; it drops it by 8–10 dB.

Gap s / height hNear-end crosstalk (typical)What it is good enough for
1About −6 dBNothing that carries a clock; only short stubs inside a connector fan-out
2About −14 dBLow-speed digital, non-critical analogue
3About −20 dBThe usual RF and mixed-signal floor
4About −26 dBSensitive receivers, high-resolution converters
6About −34 dBBeyond this the coupling is below the noise floor of most measurements

The numbers are typical for a 50 Ω microstrip on 0.254 mm laminate and will move by a few decibels with Dk, trace width and the presence of a solder mask, so they are a design guide rather than a specification. What does not move is the shape of the curve: the useful range of the lever is between s = h and s = 4h, and beyond 4h additional gap buys very little while costing density.

Two qualifications matter in practice. First, crosstalk is a function of the parallel run length as well as the gap, so a 5 mm parallel run at s = 3h is a much smaller problem than a 50 mm one; the way to fix a long parallel run is usually to change the routing, not the gap. Second, the far-end (forward) crosstalk in a microstrip is small because the air and the dielectric carry different field fractions; in a stripline, where the field is symmetric, the forward component does not cancel and the total is higher at the same spacing.

Spacing Function Two: Differential and Coplanar Impedance

Where two traces are driven as a pair, the gap is an impedance parameter rather than a crosstalk parameter, and it trades directly against width.

Pair configurationGapDifferential impedance (100 Ω design)Width needed
Loose coupling3w or moreAbout 100 Ω with each line near 50 ΩWide; two independent 50 Ω lines
Medium coupling2wAbout 97 Ω at the loose-coupling widthSlightly narrower
Tight coupling1wFalls to about 90 Ω unless width is reducedNarrower by 10–20 %
Coplanar waveguideg on both sides, same faceZ0 falls 10–15 % as g goes from 3h to 1hNarrower; needs ground stitching

Tight coupling is popular because it saves width and because a tightly coupled pair is more tolerant of a nearby aggressor. The price appears at discontinuities: a pair that is tightly coupled in the routing but loosely coupled through a via field or a connector footprint has an impedance bump at exactly the place where the signal is most likely to be reflected. The usual fix is to keep the gap constant through the transition and to add ground vias at the pair’s edge, rather than to retune the routing impedance.

Coplanar designs add a different dependency: the gap to the ground tracks on the same face sets the impedance, and the ground has to be stitched often enough to remain one conductor. A coplanar line with ground stitching every 5 mm at 10 GHz is stitching at roughly λg/3, which is not a short at all.

Spacing Function Three: Process Floor and Clearance

Below a certain gap the board simply cannot be etched reliably, and that floor is set by copper thickness rather than by lithography. Etchant has to flow into the gap and reaction products have to flow out; in a 0.05 mm gap between 35 µm of copper, it cannot, and the result is a copper sliver or a short that appears only on part of the panel.

Copper weightPractical minimum gapWhy
0.5 oz (17.5 µm)0.075 mmEtchant exchange still works at a 4:1 gap-to-thickness ratio
1 oz (35 µm)0.10 mmBelow about 3:1 the gap etches incompletely at panel edges
2 oz (70 µm)0.15 mmSliver risk and lateral etch both rise with thickness

The second floor is electrical rather than chemical. IPC-2221 tabulates minimum spacing against voltage; for externally uncoated conductors it steps from 0.1 mm in the 31–50 V band to 0.2 mm in the 101–150 V band and upward from there, with coated and internal conductors allowed closer. The step is not linear and it is not intuitive: the 24 V to 48 V change in a telecom design crosses one of those steps, and a layout that was compliant at 24 V is not at 48 V. The table and the arithmetic are in concepts of electricity in PCB.

The third floor is reliability rather than either: at high humidity and bias, closely spaced conductors on a biased surface can grow a conductive filament (CAF) through the laminate. The mitigation is spacing and a laminate qualified for the voltage, not a coating alone.

The 3W Rule and the 3h Rule Are Not the Same Rule

This is the most common source of confusion in the introduction and function of trace width and spacing, and it is worth stating plainly: the digital “3W” rule measures centre-to-centre distance in units of trace width, and the RF “3h” rule measures edge-to-edge distance in units of dielectric height. They coincide for one specific geometry and diverge everywhere else.

RuleMeasured betweenUnitTypical geometry it impliesWhere it comes from
3WTrace centresTrace widthEdge gap = 2W; for W = 2h that is 4hDigital design practice for clock nets
3hTrace edgesDielectric heightCentre distance = W + 3h; for W = 2h that is 5hMicrowave layout practice
s = wTrace edgesTrace widthEdge gap equal to the line widthDense digital routing; crosstalk 10–15 dB worse than 3h

Applied to a 0.254 mm stack with a 0.59 mm line, 3W means an edge gap of 1.18 mm while 3h means 0.76 mm. The digital rule is the stricter one here, and it is stricter by 55 %. On a thin stack the relationship reverses. This is why a rule quoted without its units is not a specification, and why a fabricator asked to “keep 3W” will ask which one you mean.

What Etching Does to Both Numbers at Once

Etching removes copper sideways from the trace and from the gap, so a single process variable moves width and spacing in opposite directions: the trace gets narrower and the gap gets wider. Both changes raise impedance, which is why a panel that is over-etched reads high on the coupon and still passes a width check.

NominalUndercut per side (1 oz, EF 1.5)Finished traceFinished gapImpedance effect
W 0.20 / s 0.1523 µm0.154 mm0.196 mm+8 % (narrower line, wider gap)
W 0.59 / s 0.7623 µm0.544 mm0.806 mm+2 %
W 1.18 / s 1.5223 µm1.134 mm1.566 mm+1 %

The first row is why fine-line and tight-gap designs are qualified on finished dimensions rather than on drawn ones, and why the compensation has to be applied to the image in CAM. The mechanism itself is in PCB etching.

Six Rules That Survive Contact With a Fabricator

1. State gaps as edge-to-edge finished dimensions with a tolerance, and state which rule you are applying — 3W or 3h. 2. Set the gap from s/h, not from habit: 3h for RF and mixed-signal, 4h where a sensitive receiver shares the layer. 3. Keep the gap constant through transitions — a pair that changes coupling at a via field has an impedance bump there. 4. Respect the copper-thickness floor: 0.075 mm on 0.5 oz, 0.1 mm on 1 oz, 0.15 mm on 2 oz. 5. Check the voltage band, not the nominal voltage — the IPC steps are not linear and 48 V is not 24 V. 6. Put the impedance coupon where the density is representative, because a coupon in an open border etches differently from the array in the middle of the panel.

Those six are checkable. The rules that are not checkable — “keep spacing as wide as possible”, “avoid tight coupling” — are the ones that get ignored on the third revision.

Five Claims That Do Not Survive

ClaimWhat actually happens
A wider trace reduces crosstalkAt a fixed edge gap a wider trace presents more facing area and couples more. What reduces crosstalk is the gap-to-height ratio and a continuous reference plane
Wider spacing always improves the designIt costs density, and beyond s = 4h the return is negligible. The money is better spent on stack height or on routing the parallel run shorter
3W and 3h are interchangeableThey measure different quantities from different references and differ by 55 % on a 0.254 mm stack with a 0.59 mm line
Tighter coupling always means better signal integrityIt reduces width and improves immunity to aggressors, but it makes the pair more sensitive to discontinuities and to the via field
Spacing only matters for high frequencyIt also sets the process floor, the voltage clearance and CAF margin. A 1 MHz power board can fail on a 0.05 mm gap that cannot be etched

Where the Gap Is Chosen For You: Fan-Out

In one place the designer does not get to choose the spacing, and that is the escape from a fine-pitch array. The gap is whatever is left after the pads, and it is the number that decides whether the design needs HDI.

Take a 0.8 mm pitch array with 0.4 mm pads: the channel between two pads is 0.4 mm, so one 0.15 mm trace with 0.125 mm either side fits inside the 0.1 mm process floor. At 0.65 mm pitch the channel is 0.3 mm, which leaves a 0.1 mm trace with exactly 0.1 mm gaps — at the floor, no margin. At 0.5 mm pitch the channel is 0.25 mm against 0.25 mm pads, and there is no room for a trace at all: the escape has to go through the pad (via-in-pad) or the board has to go to a 1+N+1 HDI build with 0.1 mm laser microvias.

That progression is the real reason HDI exists, and it is also why spacing is worth deciding early: once the array pitch is fixed, the gap and the trace width are both consequences, and the only remaining variables are the via strategy and the layer count. The escape arithmetic, with the density figures per channel, is worked through in HDI PCB performance and functions.

How Both Are Verified

Width is verified on the coupon by cross-section or by the impedance measurement itself, and on the panel by AOI against the CAM image. Spacing is verified differently and less well: AOI checks the drawn gap, and there is no routine measurement of finished edge-to-edge gap on a production panel because the feature is between two conductors rather than on one.

That asymmetry is why the gap should be specified with a generous margin against the process floor, and why a design that sits exactly on 0.1 mm at 1 oz is asking for a yield problem it will never see in the AOI data. The test methods that do catch it — coupon TDR, VNA and cross-section — are described in high-frequency PCB testing, and the impedance design side in impedance control in PCB manufacturing.

The layout conventions that produce a stack where width and gap are both defined — microstrip versus stripline, and what each does to the coupling — are set out in microstrip and stripline design.

Conclusion

Width and spacing are introduced together and do different jobs. Width is a ratio in disguise: on a Dk 3.38 laminate, 50 Ω means W/h ≈ 2.33, so a 0.254 mm dielectric gives a 0.59 mm line and a 0.508 mm dielectric gives 1.18 mm whether the layout wants it or not. Spacing is the free variable, and it buys coupling: about −6 dB of near-end crosstalk at s = h, −20 dB at 3h and −34 dB at 6h, with almost nothing gained beyond 4h.

The two rules everyone quotes are not the same rule. 3W is centre-to-centre in units of width; 3h is edge-to-edge in units of dielectric height; on a 0.254 mm stack with a 0.59 mm line they differ by 55 %. And both numbers move during etching, in opposite directions, which is why a fine-line design has to be qualified on finished dimensions.

XCE PCB holds both: finished widths from 0.075 mm, gaps to the copper-thickness floor, impedance to ±10 % on coupons taken from the centre and edge of every panel, 2 to 72 layers, panels to 600×1200 mm, samples in as fast as 24 hours, inspected to IPC-A-600G and IPC-6015. Send the stack and the density requirement together — request a quote or WhatsApp +86 180 3341 4657.

Frequently Asked Questions

What is the difference between the 3W rule and the 3h rule?
3W measures centre-to-centre distance in units of trace width, and comes from digital design practice for clock nets. 3h measures edge-to-edge distance in units of dielectric height, and comes from microwave layout. On a 0.254 mm stack with a 0.59 mm line, 3W means an edge gap of 1.18 mm and 3h means 0.76 mm — the digital rule is 55 % stricter there, and the relationship reverses on thin stacks.
Does a wider trace reduce crosstalk?
No, not at a fixed edge gap: a wider trace presents more facing area and couples slightly more. Crosstalk is controlled by the gap-to-height ratio and by a continuous reference plane. A design that widens traces while keeping the same gap and a thin dielectric usually measures worse than the version it replaced.
How much spacing is enough in a high-frequency layout?
Edge gap of three times the dielectric height is the usual RF floor, giving about −20 dB of near-end crosstalk for a 50 Ω microstrip. Four times the height gives about −26 dB for sensitive receivers. Beyond six times the height the coupling is below the measurement floor of most setups and additional gap only costs density.
What is the minimum gap a fabricator can etch?
It is set by copper thickness rather than by imaging: about 0.075 mm on 0.5 oz, 0.10 mm on 1 oz and 0.15 mm on 2 oz. Etchant has to flow into the gap and reaction products out, so below roughly a 3:1 gap-to-copper ratio the gap etches incompletely and leaves slivers or shorts on part of the panel.
Does tighter differential coupling improve signal integrity?
It reduces the pair width by 10–20 % and makes the pair less susceptible to a nearby aggressor, but it also makes the pair more sensitive to discontinuities: a pair that is tightly coupled in routing and loosely coupled through a via field has an impedance bump exactly where reflections matter. Keep the gap constant through transitions and stitch ground at the pair edge.
Why does an over-etched panel still pass a width check?
Because etching widens the gap while narrowing the trace, and both changes raise impedance. A line can finish inside its width tolerance and still read high on the impedance coupon. This is why fine-line and tight-gap designs are qualified on finished edge-to-edge dimensions and on impedance, not on drawn width alone.
RF & Microwave PCB Fabrication

Widths, Gaps and Impedance Held on One Panel

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

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