PCB Design Rule Check: What DRC Catches, and the Four Things It Cannot See

A design rule check is the cheapest gate in the whole PCB flow and the easiest to over-trust. It compares a layout against a list of numbers, and it will report a board as clean while that board is unmanufacturable, out of impedance tolerance, or resonant at the frequency it was designed for. The reason is not that DRC is weak — it is that DRC only knows the rules that were entered, and on a high-frequency board the four parameters that decide whether the product works are not rule-based at all. This article separates the three checks that all get called “DRC”, lists the rules worth building a deck around, and names the four things no rule deck can see.

PCB Design Rule Check: What DRC Catches, and the Four Things It Cannot See

Three Checks Are Called DRC

The confusion starts with the name. In a design review someone says “did you run DRC?” and three different activities answer:

GateWho runs itWhat it comparesWhat it catches
1. DRC in the CAD toolDesignerLayout against the rule deckClearance, width, annular ring, mask and silkscreen violations — geometric conflicts
2. DFM / CAM reviewFabricatorThe data against real process capabilityEtch compensation, acid traps, slivers, panel utilisation, drill-to-copper, registration margin
3. Electrical / RF verificationDesigner or SI engineerThe layout against the electrical intentImpedance, return path, stub resonance, coupling, launch geometry

A clean gate 1 says nothing about gates 2 and 3. Most of the expensive mistakes in this industry are boards that passed DRC, were manufactured correctly, and did not work — because nobody ran the third gate at all, or ran it against a stack-up that the fabricator then changed.

DRC Only Knows the Rules You Entered

A design rule check is a comparator, not an expert. It takes the numbers in the deck and reports where the layout disagrees with them. Two consequences follow, and both cost money:

  • A deck that is too generous passes boards the shop cannot build. The classic version is a default 6 mil clearance setting on a design that actually uses 4 mil in a BGA escape. DRC reports nothing, the fabricator quotes a premium or rejects the panel, and the schedule absorbs a re-spin.
  • A deck that is too tight forces unnecessary re-spins. Importing a 0.1 mm annular ring rule onto a board that a capable shop would build with 0.05 mm creates hundreds of false violations, and the team learns to ignore the report — which is worse than not running it.

The rule deck therefore has to be derived, per material and per shop, from a capability sheet that states minimum trace and space, minimum annular ring, minimum mask dam, drill sizes and aspect ratio, copper-to-edge, and registration tolerance. Where a design needs to go tighter than that, the exception belongs in a written note to the fabricator, not in a quietly relaxed deck.

The high-frequency case needs its own deck. PTFE laminates are dimensionally less stable than FR-4, drill smaller aspect ratios well and too-large ones badly, and behave differently in etching, so a deck copied from an FR-4 project will be simultaneously too tight on some rules and too loose on others. The material-specific process limits are in high-frequency PCB versus standard PCB.

The Rules Worth Building a Deck Around

These are the ones that actually produce rejections and yield loss. Values are typical starting points; every one of them should be replaced by the number on the fabricator’s capability sheet.

RuleTypical valueWhat happens when it is violated
Min trace / space0.1 mm on 1 oz outer copper; 0.15 mm on 2 ozEtch undercut removes copper from both sides of the trace; a 0.1 mm line on 2 oz can lose 40 % of its width and change impedance
Annular ring0.05 mm minimum, 0.15 mm preferredDrill break-out or a ring that survives drilling but cracks at reflow — the classic intermittent via
Solder mask dam≥0.1 mm between padsMask lifts or bridges, leaving exposed copper between fine-pitch pins
Copper to board edge≥0.3 mm (0.5 mm for routed edges)Burrs and exposed copper at the profile; shorts at the panel edge after routing
Hole to copper≥0.25 mmResin void or CAF path between a plated hole and an inner layer
Aspect ratio≤6:1 on PTFE; per shop otherwiseIncomplete plating and voids in the barrel centre on deep holes
Silkscreen over padZero overlapNon-wetting on the pad and a solder defect that looks like a paste problem
Thermal relief on planes4 spokes, 0.25–0.3 mmCold joints on through-hole parts, or a plane that never reaches the barrel

Two of these deserve a comment because they are routinely set wrong. Annular ring is not one number: the finished ring is the pad minus the drill, and it shrinks again with registration and drill wander, so a nominal 0.05 mm on a 12-layer board is a different risk than on a 2-layer one. And minimum trace width on heavy copper is an electrical rule as much as a geometric one — 2 oz copper etches with roughly 28 µm of undercut per side, which is why fine lines on heavy copper are a cost and yield problem before they are an impedance problem.

The First Thing DRC Cannot See: Impedance

Impedance is not a rule in the layout. It is a property of the cross-section — trace width, dielectric thickness, Dk, copper thickness and the presence of mask — and the layout tool has none of that unless a stack-up has been built and impedance targets have been calculated against it.

So DRC can happily pass a 0.2 mm trace that is 50 Ω on one stack-up and 61 Ω on another, and it will never notice. Four inputs change the answer more than the trace width does:

  • Dielectric thickness tolerance. ±10 % on the core is the single largest contributor to impedance error, worth roughly ±6 % on a 50 Ω line — before the width is even considered.
  • Etch tolerance. ±25 µm on the finished width is ±2.8 % on a 0.254 mm trace and ±5.5 % on 0.127 mm, which is why fine-line impedance is harder than it looks.
  • The bondply in a hybrid stack. A 0.10 mm bondply layer omitted from the model moves a 50 Ω line to roughly 61 Ω. It is the most common single error in RF stack-ups.
  • Solder mask. A mask layer over a microstrip pulls impedance down by 1–3 Ω; a test coupon measured without mask therefore reads high relative to the finished board.

The control is not a stricter DRC, it is a constraint system: define the impedance targets in the schematic or constraint manager, calculate them against a named stack-up, and let the tool hold the width. Then verify the stack-up did not change between design and fabrication. The calculation itself is worked through in common impedance designs and layer stacks.

The Second: The Return Path

A DRC deck contains no notion of current returning. A trace crossing a split in the reference plane is geometrically perfect — correct width, correct clearance, correct annular rings — and electrically broken, because the return current has to detour around the gap, which raises inductance and radiates.

Three patterns account for most of it: a signal crossing a plane split between two ground pours; a reference plane change at a via with no stitching via nearby; and a connector or filter region where the ground is stitched only at the edges. None of them is a rule violation, and all three are visible in a plot of the ground layer, which is why a hybrid stack is reviewed as a cross-section rather than as a set of layers (see stack-up design).

The rule that does exist — stitching via spacing — is frequency-dependent and usually set far too loose. A ground via fence has to be a small fraction of a wavelength apart to look like a wall: at 10 GHz in a PTFE stack, λg is around 20 mm, so a 5 mm pitch is roughly λ/4 and already leaky; at 28 GHz the same 5 mm pitch is closer to λ/2 and provides almost nothing.

The Third: Stub Resonance and Via Transitions

A through via on a thick board carries a stub below the layer where the signal leaves it. The stub is an open-circuited transmission line, and it resonates when its length reaches a quarter wavelength — which means the via behaves as a short at one frequency and as a notch at another, regardless of what the DRC report says.

The numbers are easy to work out and rarely worked out: a 1.6 mm stub resonates near 33 GHz in a PTFE stack (εeff ≈ 2.2) and near 23 GHz in FR-4 (εeff ≈ 4.3). On a 2.4 mm board the same calculation lands at 22 GHz and 15 GHz. This is why a design that works on a 0.8 mm stack fails when the mechanical team thickens the board: nothing in the rule deck changed, and the resonance moved into the band.

The fixes are structural, not geometric: back-drill the stub, use blind or buried vias, or route the high-frequency layers close to one side of the stack so the stub is short. Each of them is a stack-up decision made before layout, which is exactly why the third gate cannot be a post-layout check.

The Fourth: Everything Defined by the Material

The last category is the one that catches RF designers most often: parameters that live in the laminate data sheet, not in the layout.

  • Dk tolerance and lot-to-lot variation. A ±0.05 Dk window on 3.38 is ±1.5 %, which shifts a filter’s centre frequency by the same 1.5 % — 150 MHz on a 10 GHz design. No rule deck expresses that.
  • TCDk and phase over temperature. The drift is real, calibrated against, or designed around; none of it is a DRC item.
  • Copper roughness. At high frequency the conductor loss depends on the foil profile, and the profile is chosen at fabrication, not drawn in layout.
  • Finish thickness. Electroless nickel sits inside the skin depth above roughly 10 GHz; the finish is an electrical decision that appears nowhere in the CAD data.

These are handled the only way they can be: by writing them into the fabrication note. The ten parameters ranked by how often they decide a design are listed in ten critical parameters for high-frequency PCBs.

What to Send With the Gerbers

The gap between gate 1 and gate 2 is closed by documentation, not by more checking. Five items cover it:

ItemWhat it must containWhat it prevents
Stack-up drawingMaterial per layer, dielectric thickness with tolerance, copper weight, Dk used in the modelThe fabricator substituting a standard core that changes impedance
Impedance tableTarget, tolerance, which layers, single-ended or differential, test coupon requirementA board that is manufacturable and out of tolerance
Finish and mask calloutFinish type with thickness per the relevant IPC spec, mask type and colourAn ENIG nickel thickness nobody checked against the loss budget
DRC report + netlist compareThe actual report, plus a netlist comparison from the Gerbers back to the schematicOpen and short circuits — which is netlist compare, not DRC
Panel and coupon requirementsPanel size, array style, impedance coupon per lotNo way to prove the impedance after the fact

The netlist comparison item is worth stating plainly because the original article credited DRC with preventing opens and shorts. It does not. A geometric rule check has no concept of connectivity; a netlist compare between the layout and the schematic, or between the Gerbers and the netlist, is what finds a broken net. Both checks belong in the release gate, and they are different checks.

Building the Rule Deck in Five Steps

  • Get the capability sheet first. Not the marketing page — the numbers the CAM department actually works to, per material class.
  • Encode it once, per material. An FR-4 deck and a PTFE deck differ on minimum width, annular ring, aspect ratio, panel utilisation and drill sizes.
  • Put impedance in constraints, not in DRC. Target values with tolerances, calculated against a named stack-up, held by the router.
  • Add the connectivity check. Netlist compare on every release, in both directions.
  • Review the grounds as a plot, not a report. One minute looking at the reference planes catches more than an hour of rule violations, because the return path is invisible to the checker.

Run in that order, the deck stops being a formality. The measure of a good rule set is not that it reports zero violations — it is that the fabricator’s engineering query comes back empty.

Five Claims in the Original Article That Do Not Hold

ClaimWhat is actually true
“DRC prevents short and open circuits”Connectivity is a netlist comparison, not a geometric check. DRC can flag two nets closer than the rule allows; it cannot tell you a net is broken
“DRC optimises signal integrity”DRC enforces geometric limits. Signal integrity depends on impedance, return path, stub length and material properties — none of which the rule deck contains
“DRC enhances thermal management”At most it checks thermal relief geometry and copper area. Thermal performance is a stack-up and copper-distribution question, evaluated separately
“DRC ensures compliance with IPC standards”It ensures compliance with the rules you entered. IPC class compliance is a fabrication and inspection question, assessed against IPC-6012 and the applicable class on the finished board
“Real-time feedback means problems are caught early”Real-time feedback catches geometric errors during routing. The expensive ones — stack-up, impedance, return path, stub resonance — are decided before routing starts and are not visible in real time at all

Conclusion

A PCB design rule check is a comparator against a list of numbers, and it is worth exactly what the list is worth. Build the deck from the fabricator’s real capability, per material, and it removes a whole class of re-spins. Copy a default deck and it produces either false confidence or a report nobody reads.

Then run the two gates it cannot replace. Netlist compare for connectivity. A stack-up and return-path review — with the impedance table calculated against a named cross-section and the stub length checked against the highest frequency in the design — for everything that decides whether the board works. On a high-frequency project those two reviews are where the engineering actually happens; DRC just keeps the geometry honest.

XCE PCB runs a DFM review on every high-frequency order before release: stack-up and impedance model checked against the quoted material, annular ring and mask dam re-checked at the shop’s qualification level, and a fabrication note returned with the confirmation. 2 to 72 layers, impedance ±10 %, panels to 600×1200 mm, samples in as fast as 24 hours. The full sequence is in the twelve-step manufacturing overview.

Frequently Asked Questions

What does a PCB design rule check actually verify?
It compares the layout geometry against a list of numeric rules — minimum trace and space, annular ring, solder mask dam, copper-to-edge distance, hole-to-copper clearance, silkscreen overlap. It reports where the layout disagrees with those numbers. It does not check connectivity, impedance, return paths or material properties, so a clean DRC report is a necessary but far from sufficient condition for a working board.
What is the difference between DRC and DFM?
DRC runs in the CAD tool against the designer's rule deck. DFM is the fabricator's review of the released data against real process capability: etch compensation, acid traps, slivers, drill-to-copper margins, registration, panel utilisation. A board can pass DRC and still be rejected or built at a premium because the deck did not match what the shop can actually do.
Why does my board pass DRC but fail impedance testing?
Because impedance is set by the cross-section, not by the layout. Dielectric thickness tolerance (±10 % is worth about ±6 % on a 50 Ω line), etch tolerance, a bondply layer missing from the model, and solder mask all move the result. A 0.10 mm bondply omitted from the stack-up calculation can take a 50 Ω line to roughly 61 Ω while every geometric rule passes.
How often should the design rule deck be updated?
Whenever the fabrication capability changes, and at minimum once per material class per project. An FR-4 deck and a PTFE deck differ on minimum trace width, annular ring, aspect ratio and drill sizes, so a deck carried over from an FR-4 project is usually both too tight and too loose at the same time. Derive it from the shop's capability sheet, and record any intentional exception in writing.
Can DRC catch a via stub resonance?
No. Stub resonance is a property of the stack-up: a through via carries a stub below the layer where the signal leaves it, and it resonates near a quarter wavelength. A 1.6 mm stub resonates around 33 GHz in a PTFE stack and 23 GHz in FR-4, so the fix — back-drilling, blind or buried vias, or moving the RF layers close to one surface — is a stack-up decision made before layout.
Is netlist comparison part of DRC?
No, and this is a common misattribution. DRC is geometric; it has no concept of connectivity. A netlist compare between layout and schematic, or between the Gerbers and the netlist, is what finds a broken net or a short between two nets. Both checks should be part of the release gate, and they are separate tools.
RF & Microwave PCB Fabrication

Want a DFM Review Before the Gerbers Are Released?

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

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