Drilling High-Frequency PCB: Where a Hole Stops Being a Hole and Becomes Three Specifications

A drilled hole carries three separate promises: it has to land inside its pad, it has to present a wall that copper can bond to, and it has to be short enough that the unused length does not resonate. On FR-4 all three are routine. On PTFE they are not, because the material removes heat twenty times more slowly than copper and softens before it cuts — so the drill can smear the wall instead of severing it, and no amount of plating will repair that. These are the steps and the numbers, with the parts of the standard description that do not survive contact with a fluoropolymer.

Drilling - high-frequency PCB manufactured by XCE PCB

Three Specifications in One Hole

Drilling is usually described as making connections between layers. Electrically that is right, and it hides the three tolerances a drill program actually has to satisfy.

  • Position. The hole must land inside the pad with a usable annular ring left over. This is a budget shared with the lamination shrink described in the lamination guide.
  • Wall. The wall has to be clean, correctly sized and free of smear, because the plating that follows bonds to what the drill left behind. A smeared PTFE wall plates, passes continuity, and fails in thermal cycling.
  • Length. The unused portion of a through via is a stub, and a stub is a resonator. At 1.6 mm it rings near 23 GHz, which is why back-drilling exists.

This is step six in the twelve-step sequence, after the press and before plating. It is also the step where the material stops behaving like a generic laminate: PTFE has a thermal conductivity around 0.20 W/m·K against 0.30 for FR-4 and 385 for copper, so friction heat stays at the drill edge.

The Numbers That Define a Hole

ParameterTypical valueWhy it matters
Finished holeDrill diameter − 2 × plating25 µm per side means a 0.30 mm finished hole is drilled at 0.25 mm
Aspect ratio≤6:1 on PTFE, up to 10:1 on FR-4Sets whether plating solution reaches the hole centre
Hole position±0.05 mmAgainst a 0.05 mm annular ring, one direction consumes the budget
Burr≤0.025 mmBurrs lift the pad and shadow the plating
Wall roughnessDeep grooves cannot be covered by a 0.3 µm seed
Hit count per toolA few thousand, material dependentWorn carbide burnishes PTFE instead of cutting it

The first row is the one that gets lost between design and fabrication. A drill file released with finished diameters produces holes 0.05 mm undersize on a 25 µm-per-side plating process, which closes the annular ring and can make a 0.30 mm via too small for the component lead. The convention has to be stated, exactly as the width convention has to be stated for etching.

Step 1 — Stack Height, Entry and Backing

Drilling is done through a stack of panels, and the stack is a registration decision before it is a productivity decision. Each additional panel in the stack adds position error, so a three-panel stack on a 0.30 mm tool is normal for FR-4 and a single or two-panel stack is normal for thin PTFE.

Entry and backing material do two jobs: they stop burrs on the copper at entry, and they stop the drill from pushing laminate away at exit.

  • Entry: aluminium foil 0.1–0.15 mm. It also dissipates heat from the drill tip, which matters more on PTFE than on FR-4.
  • Backing: 1.5–2.5 mm phenolic or composite. On soft fluoropolymer the exit burr is the dominant defect, and a rigid backing is the cheapest way to control it.
  • Pinning and datum. Panels are pinned to the same datum used in lamination and imaging. If the drill references a different datum, the registration budget is spent before the first hole.

Step 2 — Tool Geometry, and Why PTFE Needs a Sharp Edge

Carbide is universal; the geometry is where materials diverge. Two angles carry most of the difference.

Point angle. 130–140° is the working range for PCB work. A blunter point concentrates thrust, a sharper one wanders more on entry — on a 0.25 mm tool, wander is a measurable fraction of the annular ring.

Helix angle. 35–45°. The helix controls how fast the chip leaves the hole; on PTFE the chip is a soft ribbon that will re-weld to the wall if it stays in contact, so a free-cutting geometry matters more than a durable one.

Tool life is where PTFE punishes the process economically. RT/duroid 5880 is a micro-glass-fibre composite, and glass is abrasive. A tool that has drilled a few thousand hits in FR-4 still cuts FR-4; the same tool in PTFE has a rounded edge that burnishes the wall and produces smear. The rule that works is to change tools by hit count and by material, not by breakage.

Step 3 — Parameters: Two Numbers, Not One

Spindle speed and feed rate are usually quoted as a pair, which hides the two quantities that actually matter. Surface speed is what the cutting edge sees, and it is a function of diameter:

v = π × D × n  →  a 0.30 mm drill at 120,000 rpm runs at about 113 m/min; the same spindle speed on a 0.60 mm drill doubles it.

That is why a single spindle-speed recipe cannot be carried across a drill file: the same rpm is a different cutting condition at every diameter. The second number is chipload, the advance per revolution, typically 2–5 µm on small tools. Multiply the two and the feed rate falls out: 120,000 rpm at 3 µm per revolution is 0.36 m/min.

DrillSpindle speedSurface speedChiploadFeed
0.20 mm140,000 rpm88 m/min2 µm0.28 m/min
0.30 mm120,000 rpm113 m/min3 µm0.36 m/min
0.50 mm80,000 rpm126 m/min5 µm0.40 m/min
1.00 mm40,000 rpm126 m/min10 µm0.40 m/min

The values are starting points, and the rule behind them is the useful part: hold surface speed roughly constant as diameter changes, and keep chipload high enough to cut rather than rub. A low feed on a high spindle speed is the classic PTFE failure — the edge rubs, the wall heats, and the material smears.

Step 4 — The Heat Problem: Why PTFE Smears

Copper conducts heat at 385 W/m·K. FR-4 manages 0.30, RO4350B 0.69, RO3003 0.50, and unfilled PTFE such as RT/duroid 5880 about 0.20 — the worst of the family. A drill generating friction heat at the cutting edge has nowhere to send it, so the edge temperature climbs until the polymer softens.

PTFE has no true glass transition in the usual sense and melts around 327 °C. A dull edge on a low feed will reach that locally, and the result is not a cut surface but a smeared one: the polymer flows across the glass fibres, seals the wall, and hides the fibre ends that the plating chemistry needs to grip.

Smear is invisible after drilling. It is visible after plating only as a blister or a pull-away in cross-section, and in the field as a barrel that opens after thermal cycling. Three process answers, in order of effect:

  • Cut, do not rub. Adequate chipload with a sharp tool. This is the fix; everything else is mitigation.
  • Replace tools by hit count. A worn edge produces smear before it produces a broken hole.
  • Move the heat away. Air blast and vacuum extraction at the tool, aluminium entry foil as a heat spreader, and a pause between hits on small tools rather than a continuous high-speed run.

What does not belong here is liquid coolant. Production PCB drilling is dry: the panel is a laminated dielectric with a bond interface that absorbs water, and the same moisture that causes pink ring at lamination will cause delamination and outgassing at reflow. Cooling is air and vacuum, not flood.

Step 5 — Desmear: The Chemistry That Does Not Transfer

After drilling, the wall has to be prepared for plating, and this is the second place where a generic FR-4 process fails on PTFE. FR-4 desmear works by oxidising epoxy smear with alkaline permanganate: the resin is attacked, the smear is dissolved, and the glass is left with a microscopically rough surface for the seed layer.

PTFE has no epoxy in it. Permanganate has nothing to oxidise, so a standard desmear line returns a wall that is chemically unchanged — and a fluoropolymer surface has a surface energy so low that plated copper will not bond to it at all. The two routes that work are sodium naphthalene etching, which defluorinates the surface and leaves a carbonaceous, bondable layer, and plasma treatment, which does the same job in a vacuum chamber with oxygen or hydrogen process gas.

MaterialDesmear routeWhat it doesRisk if skipped
FR-4Alkaline permanganateOxidises and dissolves resin smearWedge voids, poor adhesion
RO4000 (4350B / 4003C)Permanganate, gentlerThermoset smear, thinner than FR-4Resin residue on the wall
Ceramic PTFE (3003 / 3010)Sodium naphthalene or plasmaActivates the fluoropolymer surfaceNo adhesion — barrel pulls out
Unfilled PTFE (5880 / TLY-5)Sodium naphthalene or plasmaRemoves smear and activatesSmear stays under the plating

The residue from sodium naphthalene is as much of a problem as the smear it removes: it is dark, it must be rinsed thoroughly, and sodium carried into the plating bath contaminates it. The working discipline is a dedicated line, a rinse that is verified rather than timed, and a water-break check on the wall — the same 30-second test described in the inner-layer guide.

Step 6 — Registration, Annular Ring and the Budget Nobody Owns

A via lands inside its pad only if four errors fit inside the ring: drill position, panel shrink, layer-to-layer registration, and pad definition. The numbers are individually small and collectively larger than the ring.

ContributionTypical valueWhere it comes from
Drill position±0.025–0.05 mmMachine accuracy plus stack height
Panel shrink0.12–0.30 mm over 600 mmPrepreg cure — compensated by scale factor
Layer registration±0.05–0.075 mmImaging and lamination
Remaining annular ring≥0.05 mmWhat is left after the three above

Once shrink is compensated per lot, the budget is mostly drill position and registration, and the two are correlated: a stack of three panels drilled on a machine with ±0.025 mm accuracy is not a ±0.025 mm process, because the panels move relative to each other. That is why thin PTFE builds are drilled one or two up even when the machine could take three.

Step 7 — What the Drill Decides About the RF Path

Two RF consequences are settled here and cannot be fixed later.

The stub. An unused through-via barrel is a transmission line shorted at one end. Its first resonance sits at a quarter wavelength in the filler medium: about 23 GHz at 1.6 mm, 18–19 GHz at 2.0 mm, 12 GHz at 3.2 mm. Below that frequency the stub adds capacitance and a notch appears above it. Back-drilling removes the stub with a slightly larger tool after plating, and its depth tolerance is what decides whether the remaining stub is short enough — a controlled-depth operation whose error budget is measured in tenths of a millimetre.

The wall. A rough wall is a longer conductor path and a poorer bond. At millimetre wave the current in a plated barrel is confined to a skin depth of about 0.65 µm at 10 GHz and 0.23 µm at 77 GHz, so the surface the drill leaves is literally the surface the signal uses.

Laser, Mechanical, or Both

Mechanical drilling covers through holes from about 0.2 mm upward. Below that, and for blind vias in HDI builds, laser drilling is used: UV or CO2 depending on the dielectric, with the depth controlled by energy and pulse count rather than by a physical stop.

The split matters for high-frequency work because the two processes leave different walls. A laser-drilled via in PTFE has a tapered wall and a heat-affected rim; it plates well but the taper changes the via inductance slightly and the rim has to be desmeared like any other. A mechanically drilled via in the same material has a straighter wall and a smear risk that the laser does not have. Neither is universally better — they fail differently, and the stack-up should say which one each via class expects.

Where Drilling Goes Wrong

SymptomRoot cause in this stepFix
Barrels open after thermal cyclingSmear under the plating on PTFESharp tool, higher chipload, Na-naph or plasma desmear
Vias off-centre on one edge of the panelCure shrink not compensated, or a three-up stackPer-lot scale factor; drill one or two up
Holes undersize after platingDrill file released with finished diametersState the convention; drill = finished − 2 × plating
Intermittent opens, coupon passesWall roughness deeper than the seed layer can coverControl chipload and hit count; inspect cross-section
Nodules and burrs lifting padsNo entry foil or a soft backingAluminium entry, rigid backing, deburr
Narrow band of loss at 20–25 GHzStub resonance from an unbackdrilled through viaBack-drill to depth, or route the signal on an inner layer

Drilling Checklist

#CheckPass condition
1Drill versus finished diameter declaredDrill = finished − 2 × plating thickness
2Aspect ratio inside the material limit≤6:1 on PTFE; ≤10:1 on FR-4
3Stack height matches the toleranceOne or two up for thin PTFE
4Entry foil and rigid backingAluminium 0.1–0.15 mm; backing 1.5–2.5 mm
5Tool geometry suited to the resin130–140° point, 35–45° helix, sharp
6Surface speed and chipload both setv held roughly constant across diameters; 2–5 µm/rev
7Tool change by hit countRecorded per material, not by breakage
8Desmear route matches the resinPermanganate for FR-4; Na-naph or plasma for PTFE
9Position and ring verified±0.05 mm position with ≥0.05 mm ring
10Stub strategy statedBack-drill depth or inner-layer routing for RF vias

Conclusion

Drilling looks like the most mechanical step in the sequence and turns out to be the one where material science has the last word. The three specifications a hole carries — position, wall and length — are decided by three different things: the datum and the stack, the heat at the cutting edge, and the layer the signal is routed on.

The claims worth discarding are the ones that assume a generic laminate: that PCB drilling uses liquid coolant (it is dry, and liquid is a defect source on a laminated dielectric), that soft PTFE wants slow spindle speeds (it wants a sharp edge and a real chipload, or it smears), and that a standard permanganate desmear prepares a fluoropolymer wall (it has nothing to oxidise).

XCE PCB drills Rogers, Taconic and F4B builds with diameter-specific parameters, hit-count tool changes, sodium naphthalene or plasma desmear for PTFE, and back-drilling for RF through vias: 2–72 layers, impedance ±10%, panels to 600 × 1200 mm, 5,000 m² of production space and 80+ engineers, with samples in as fast as 24 hours. The wall these tools leave is metallised in the plating step.

Frequently Asked Questions

Do PCB drills use coolant?
No. Production PCB drilling is dry — cooling and chip removal come from air blast and vacuum extraction at the tool. Liquid coolant would be absorbed at the laminate interfaces and act as a moisture source, which is the same mechanism behind pink ring at lamination and outgassing during reflow.
Should PTFE be drilled at lower spindle speed?
No, the opposite. PTFE needs a sharp edge and a real chipload so the material is cut rather than rubbed. Low feed on a high spindle speed is the classic smear recipe, because friction heat has nowhere to go: PTFE conducts heat at about 0.20 W/m·K against 385 for copper, and softens near 327 °C.
How is the drill diameter chosen?
Finished hole plus two plating thicknesses: a 0.30 mm finished via on 25 µm per side is drilled at 0.25 mm. The drill file must state whether it carries drill or finished diameters, because a file released with finished sizes produces undersized holes and eats the annular ring.
Why is aspect ratio limited on high-frequency laminates?
Because plating solution has to reach the middle of the barrel. Around 6:1 is the practical limit on PTFE, against 10:1 on FR-4, and the limit is set by the plating chemistry rather than by the drill: beyond it the copper at the hole centre is thin and can be discontinuous.
Can permanganate desmear be used on PTFE?
No. Permanganate works by oxidising epoxy smear, and PTFE contains no epoxy, so the wall comes out chemically unchanged — and a fluoropolymer surface has too little surface energy for plated copper to bond to it. Sodium naphthalene etching or plasma treatment is required, followed by a thorough rinse.
What causes a loss spike around 20–25 GHz?
Stub resonance. An unused through-via barrel is a shorted transmission line whose quarter-wave resonance lands near 23 GHz at 1.6 mm thickness, 18–19 GHz at 2.0 mm and 12 GHz at 3.2 mm. Back-drilling removes the stub, with the remaining length set by the depth tolerance of that operation.
RF & Microwave PCB Fabrication

Need PTFE Holes Without Smear?

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

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