Route Width in PCB Manufacturing: What the Finished Trace Actually Measures

Route width is treated as a design preference — wider for power, narrower for density — when in fact it is a manufacturing tolerance, and the number on the drawing is not the number the current sees. Etching forms a trapezoid rather than a rectangle, plating thickens the outer layers unevenly, and the same 100 µm line can finish at 77 µm average width and 30 % more resistance. This article sets out what width actually controls — current capacity, characteristic impedance and process capability — the arithmetic of the etched cross-section, and why widening a 50 Ω line is really a decision about dielectric thickness rather than about width.

Route Width in PCB Manufacturing at XCE PCB's high-frequency PCB factory

Three Jobs, One Number

A trace width is asked to do three unrelated jobs at once, and the three set limits in different directions. Current capacity wants more copper. Impedance control wants an exact ratio of width to dielectric thickness. Process capability wants a width wide enough to etch, plate and cover with solder mask. The original version of this article listed those jobs but gave no numbers, and two of its statements are the wrong way round.

JobWhat sets the limitDirectionThe number that matters
Current capacityCross-section, copper weight, allowed temperature rise, layer positionWider is better, with diminishing returnsAmperes per millimetre of width at a stated rise
ImpedanceRatio of width to dielectric thickness for a given DkWidth is fixed once the stack is chosenW/h, and the tolerance on both
ManufacturabilityCopper thickness, etch factor, mask resolutionNarrower is harderMinimum width per copper weight; finished-width tolerance

Because the three constraints apply to the same geometric feature, a change made for one of them is always paid for by another. That is why the useful question is never “how wide should this trace be” but “which of the three is binding here”.

The Finished Width Is Not the Drawn Width

Etching removes copper isotropically, so it cuts sideways as well as down. The result is a trapezoid: the top width is set by the resist image, the bottom width by the top width minus twice the undercut, and the undercut is roughly the copper thickness divided by the etch factor. With an etch factor of 1.5 on 1 oz copper, that is about 23 µm per side.

Copper weightThicknessUndercut per side at EF 1.5Drawn 100 µm finishes asAverage width / resistance change
0.5 oz17.5 µm12 µm76 µm at the base88 µm / +14 %
1 oz35 µm23 µm54 µm at the base77 µm / +30 %
2 oz70 µm47 µm6 µm at the baseNot achievable — the line is undercut away

The 2 oz row is why heavy copper has a minimum line width instead of a preferred one. A 100 µm image on 70 µm foil with a 1.5 etch factor barely exists at the base, and any real process would either widen the image or reduce the etch factor by using a different chemistry. The practical floors, with the process under control, are about 0.075 mm on 0.5 oz, 0.1 mm on 1 oz and 0.15 mm on 2 oz — and the numbers on a drawing should be finished widths, with the CAM department applying the compensation.

Outer layers add a second complication. Pattern plating deposits 20–25 µm of copper on the walls and the surface, so an outer trace drawn at 100 µm top width ends up 50–60 µm thick with sidewalls that are closer to vertical than an inner layer’s. That is good for resistance and bad for fine-line control: the same etched image is 0.1 mm on an inner layer and 0.13 mm on an outer one, which is why impedance coupons are layer-specific and cannot be shared.

Current Capacity: Amperes per Millimetre, and Why Inner Layers Are Halved

Current capacity is a thermal question wearing an electrical costume. A trace heats because of I²R, and the steady-state temperature rise depends on how much copper there is and how well the heat escapes. IPC-2221 tabulates the answer for a 10 °C rise; IPC-2152 is the more recent and generally less optimistic model.

Copper weight1 mm wide, outer layer, 10 °C riseSame trace on an inner layerCross-section
1 oz2.3 Aabout 1.1 A0.035 mm²
2 oz4.0 Aabout 2.0 A0.070 mm²
6 oz8.8 Aabout 4.4 A0.210 mm²

The halving on inner layers is not a safety factor invented by conservative engineers; it is the absence of convection and radiation on both faces, with only conduction into the laminate available. It matters most in exactly the boards where it is most often forgotten: multilayer power distribution, where the plane carrying the current is buried.

Two consequences follow for width. First, capacity scales with the product of width and thickness, so doubling copper weight and halving width is roughly neutral thermally but very different for impedance and for etch capability. Second, resistance and therefore voltage drop scale with length over width, so a 2 A rail with a 5 % drop budget needs a width derived from length, not from the temperature table — the table tells you when the trace gets hot, not when the load browns out.

Impedance Fixes the Ratio, Not the Width

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

Dielectric thickness50 Ω width (Dk 3.38)Width for 0.075 mm tolerance as % of nominalWhich build it suits
0.102 mm0.24 mm31 %Thin HDI cores; tolerance dominates
0.203 mm0.47 mm16 %Fine-pitch digital
0.254 mm0.59 mm13 %The common microwave stack
0.508 mm1.18 mm6.4 %Low-loss long runs
0.813 mm1.89 mm4 %Antenna feeds, high-power RF

The right-hand column is the one to look at. Etch tolerance is roughly a fixed number of micrometres, so its effect on impedance shrinks as the line gets wider: ±25 µm is 10 % of a 0.24 mm line and only 1.3 % of a 1.89 mm line. Translated into impedance, a 10 % width change moves Z0 by roughly 4–5 % at W/h ≈ 2, which is why thin-dielectric 50 Ω designs need tighter etch control or a relaxed impedance window, and why the same fabricator can hold ±5 % on one stack and only ±10 % on another without doing anything differently.

Widening a 50 Ω Line Is Really a Decision About Thickness

The instinct on a lossy link is to widen the trace. For a controlled-impedance line that is not available: width is locked to dielectric thickness by the ratio, so a wider trace means a thicker laminate. The trade can be quantified at 10 GHz on a Dk 3.38 stack, where the dielectric contribution is about 0.13 dB/in and the copper contribution about 0.19 dB/in at 0.254 mm.

Dielectric50 Ω widthCopper lossDielectric lossTotal at 10 GHz
0.254 mm0.59 mm0.19 dB/in0.13 dB/in0.32 dB/in
0.508 mm1.18 mm0.095 dB/in0.13 dB/in0.225 dB/in
0.813 mm1.89 mm0.063 dB/in0.13 dB/in0.193 dB/in

Two readings follow. The dielectric term does not move with thickness, because it is a loss per unit of electrical length rather than per unit of physical length; only the copper term responds to width. And the returns diminish quickly: doubling the thickness buys 0.095 dB/in, tripling it buys a further 0.03 dB/in, while the board gets three times thicker and the via aspect ratios with it. The point where the two contributions are equal — about 0.35 mm on this laminate — is the natural stopping place for a loss-driven stack, unless the mechanical envelope demands otherwise.

This is also why the copper thickness argument is weaker than it looks. At 10 GHz the current occupies the outer 0.66 µm of a 35 µm foil, so doubling to 2 oz reduces the conductor loss by a few percent while widening the trace by the same factor reduces it by half. Width is the effective lever, but it is bought with dielectric thickness, not with copper weight.

Width Is Not Uniform Across a Panel

Etching happens in a spray chamber, and the etch rate depends on fresh chemistry reaching the surface. In a dense array of fine lines the local etchant is depleted and the copper load is high, so lines etch narrower than in an open area; at the panel edge the spray is stronger, so they etch wider. The result is that the same drawn width can finish 10 % apart across one panel.

Three controls address it. Etch compensation is applied per layer in CAM, at not less than twice the copper thickness for heavy foil, and it is applied to the drawn image rather than to the finished number. Panel layout places the impedance coupons in a copper-balanced area, not in the open border. And the process is qualified by measuring finished width on coupons from the centre and the edge, not by measuring one point.

For a designer the practical consequence is a rule about tolerance budgets rather than about etch: on a 0.254 mm stack, ±25 µm of etch, ±10 % of dielectric thickness and ±0.05 of Dk together account for most of a ±10 % impedance window, and the widths on the drawing should leave room for all three rather than assuming one.

Where Width Does Not Do What It Is Credited With

Three claims about width survive in most articles and do not survive measurement.

ClaimWhat actually happens
A wider trace reduces crosstalkAt fixed spacing, a wider trace couples more, because the facing area grows. What reduces crosstalk is spacing relative to dielectric thickness, and a reference plane between the aggressor and victim. A wider line on a thicker dielectric can be better, but the improvement comes from the dielectric, not the width
A wider trace improves heat dissipationPartly. It lowers the temperature rise for a given current because there is more copper to spread heat into, but the heat is generated in that same copper. A trace is not a heat sink for other components; thermal vias and planes are
A wider trace reduces EMIIndirectly at best. Emission is driven by loop area and common-mode current. A wider trace changes loop area only marginally; a return plane and a shorter loop change it by tens of decibels

The pattern is the same in all three: width is a local geometric variable being asked to solve a field problem. It cannot, and the fixes that work are the ones that change the field — spacing, reference planes and loop geometry, which is what trace width and spacing deals with as a pair.

What to Put on the Drawing

Five items make width a controllable specification rather than a preference:

1. Finished width per layer, with tolerance — not the drawn width, and not a single nominal number. 2. Copper weight per layer and finished outer copper, since outer traces are plated 20–25 µm thicker than the base foil. 3. Impedance target per layer with the coupon geometry, so the width and the impedance cannot drift apart. 4. Minimum width justified by the process — 0.075 mm on 0.5 oz, 0.1 mm on 1 oz, 0.15 mm on 2 oz. 5. Etch compensation responsibility, stated as applied by the fabricator, with the acceptance measurement defined.

With those five, a width dispute becomes an arithmetic check on a coupon. Without them, it becomes a conversation about whose number was nominal.

Conclusion

Route width carries three responsibilities and they conflict. Current capacity scales with cross-section and roughly halves on inner layers; impedance fixes the ratio of width to dielectric thickness, so a 50 Ω line on 0.254 mm laminate is 0.59 mm wide whether the designer wants that or not; and process capability sets a floor that rises with copper weight, from about 0.075 mm on 0.5 oz to 0.15 mm on 2 oz. The number that reaches the etcher is not the number the current sees, because etching forms a trapezoid: a 100 µm line on 1 oz foil finishes with an average width of about 77 µm and 30 % more resistance if the etch factor is 1.5.

The most useful reframing is this: on a controlled-impedance board, width is not a free variable at all. It is determined by the dielectric thickness, and the real decision is the stack. That is why a loss problem is usually solved by a thicker dielectric or a lower-Df material rather than by a wider line, and why the answer to “can you make this trace wider” is often “only by making the board thicker”.

XCE PCB etches to those windows on both conventional and high-frequency builds: 2 to 72 layers, finished widths from 0.075 mm, 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 at the centre and the edge of every panel, because a single measurement is not a process. The current-carrying tables behind the widths are in copper thickness in high-frequency PCBs.

Frequently Asked Questions

How wide should a trace be for 2 A?
On an outer layer with 1 oz copper and a 10 °C rise, IPC-2221 gives about 2.3 A for a 1 mm wide trace, so 2 A needs close to 0.9 mm; on an inner layer the capacity roughly halves and the width doubles. If the constraint is voltage drop rather than temperature, calculate from resistance instead: 1 oz copper is 0.49 mΩ per square, so a 0.5 mm wide, 100 mm long trace is 0.1 Ω and drops 0.2 V at 2 A.
Does the finished trace come out narrower than the design?
Yes, on inner layers by a predictable amount. Etching cuts sideways as well as down, forming a trapezoid whose base is narrower than the top by roughly twice the copper thickness divided by the etch factor. A 100 µm line on 1 oz foil at an etch factor of 1.5 finishes with about 54 µm at the base and 77 µm average, which is about 30 % more resistance than the nominal width implies. Outer layers are then plated 20–25 µm thicker, so the same image finishes wider.
Why can I not just widen the trace to reduce loss?
Because a controlled-impedance line has a fixed ratio of width to dielectric thickness. On Dk 3.38, 50 Ω means W/h ≈ 2.33, so a 0.254 mm dielectric gives 0.59 mm and a 0.508 mm dielectric gives 1.18 mm. Widening therefore means thickening the laminate. The copper term falls from 0.19 to 0.095 dB/in at 10 GHz between those two stacks, but the dielectric term stays at 0.13 dB/in and the board doubles in thickness.
What is the minimum trace width for 2 oz copper?
About 0.15 mm, and the limit is etch undercut rather than lithography. At 70 µm thickness and an etch factor of 1.5, each side undercuts by roughly 47 µm, so a 100 µm image would leave almost nothing at the base while a 150 µm image leaves about 56 µm. Heavier copper also needs wider spacing, and the etch compensation applied in CAM is typically at least twice the copper thickness.
Does a wider trace reduce crosstalk?
No, not at fixed spacing — a wider trace presents more facing area to its neighbour and couples more. What reduces crosstalk is the ratio of spacing to dielectric thickness and the presence of a reference plane. A design that widens traces while keeping the same spacing and a thin dielectric usually measures worse than the narrow version it replaced.
How much does width variation affect impedance?
Roughly 4–5 % of impedance for a 10 % change in width at W/h around 2. Because etch tolerance is fixed in micrometres, its effect depends on the line width: ±25 µm is about 10 % of a 0.24 mm line on a 0.102 mm dielectric and about 1.3 % of a 1.89 mm line on 0.813 mm. This is why thin-dielectric 50 Ω designs need tighter etch control or a wider impedance window.
RF & Microwave PCB Fabrication

Need Width and Impedance Held on the Same 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