Critical Via-Stub Resonance in mm-Wave RF PCB Design

A 1.5 mm via stub can resonate at 28.8 GHz. Its reflection returns 180° out-of-phase after only 17.3 picoseconds. That frequency sits directly inside modern Ka-band hardware. It also overlaps important 5G Infrastructure frequencies.

The via may look electrically insignificant inside Altium or Allegro. Electromagnetically, it is an open-ended transmission-line resonator.

The hidden conflict inside multilayer RF hardware

Modern RF products demand increasing integration density. Massive MIMO panels need more beamformer channels. Open RAN radios combine RF, digital, and power electronics. Phased Array Radar systems require tightly matched signal paths. LEO payloads demand compact, thermally efficient multilayer assemblies.

Those requirements often produce thicker and more complex PCB stack-ups. Thicker boards simplify routing, shielding, and thermal spreading. However, they also create longer unused via barrels.

That unused copper becomes a via stub.

At low frequencies, the stub appears mainly capacitive. At microwave frequencies, its behaviour becomes distributed. At resonance, the open stub transforms into a near short-circuit. The result can be a severe insertion-loss notch.

Premium Rogers substrates cannot remove this geometry-driven resonance. Neither can stronger amplifiers, calibration, or improved impedance matching.

The governing physics

Consider a plated-through via connecting two internal signal layers. Copper extending beyond the destination layer remains electrically open. That remaining section becomes an open-circuit transmission-line stub.

Its input impedance is approximated by: Zstub = −jZₛ cot(βl)

where:

• Zₛ is the stub’s characteristic impedance.

• β is the propagation constant.

• l is the unused barrel length.

At the first quarter-wave condition: βl = π/2

Therefore: cot(π/2) = 0 and Zstub → 0

The physical open circuit becomes a short circuit electrically. That short appears in shunt across the intended transmission path.

The first resonant frequency is approximately: f₁ ≈ c / (4l√εeff)

Higher resonances occur at odd multiples: fₙ ≈ (2n + 1)c / (4l√εeff), where n = 0, 1, 2…

This quarter-wave mechanism produces destructive interference. The signal enters the stub and reflects from its open end. The round trip creates 180° phase reversal at resonance. The returning wave then opposes the forward signal.  [Ref. 1]

Quantifying a 1.5 mm via stub

Assume:

• Stub length: l = 1.5 mm

• Effective permittivity: εeff = 3

• Propagation velocity: vp = c/√εeff

Then: f₁ ≈ 28.85 GHz

The round-trip delay becomes: tround-trip = 2l√εeff/c

Therefore: tround-trip ≈ 17.33 ps

The half-period at 28.85 GHz is also 17.33 ps.

That creates the required 180° phase relationship. The unused barrel therefore becomes a Ka-band resonator. This is not a theoretical edge case. It falls inside 3GPP’s established FR2 operating region.

The relationship between stub length and resonance is unforgiving:

• 2.0 mm → approximately 21.64 GHz

• 1.5 mm → approximately 28.85 GHz

• 1.0 mm → approximately 43.27 GHz

• 0.5 mm → approximately 86.54 GHz

• 0.38 mm → approximately 113.87 GHz

Samtec recently reported a comparable 0.38 mm structure. Its estimated quarter-wave resonance was approximately 110 GHz. That agreement validates the first-order design estimate. [Ref. 2]

Pain Point 1: The insertion-loss notch can appear unexpectedly

A connector launch may perform perfectly below 10 GHz. The same launch may collapse near 28 GHz. The failure often appears suddenly.

S₂₁ develops a deep, narrow-frequency notch. S₁₁ rises sharply near the same frequency. Phase also rotates rapidly around resonance. Group-delay variation then increases.

The resulting behaviour can disrupt:

• broadband impedance matching

• modulation error-vector magnitude

• channel-to-channel phase tracking

• digital predistortion

• phased-array calibration

• pulse fidelity in Radar systems

• wideband Ground station receivers

The first resonance is not the only concern. Higher odd-order resonances remain possible:

f₃ ≈ 3f₁, f₅ ≈ 5f₁

For the 1.5 mm example:

• First resonance: approximately 28.85 GHz

• Third-order resonance: approximately 86.54 GHz

• Fifth-order resonance: approximately 144.24 GHz

This matters beyond present Ka-band systems. Package and PCB dimensions are becoming wavelength-significant above 70 GHz. 

Samtec identifies this transition as a critical barrier for future interconnects. Radiation and crosstalk then emerge from physically small discontinuities. [Ref. 2]

The system-level consequence

Every decibel matters inside mm-Wave link budgets. Atmospheric loss already constrains Satellite Communications. Wide-scan Phased Arrays already suffer projected-aperture loss. Power amplifiers already operate near thermal limits.

A resonant via transition consumes margin without adding functionality. Increasing PA power does not correct the underlying phase distortion. Adding calibration cannot recover a complete spectral notch.

The correct solution must begin inside the PCB geometry.

Pain Point 2: Resonance depends on more than drill depth

The first-order equation appears simple. The physical via structure is not.

The effective permittivity depends on:

• laminate dielectric constant

• resin and reinforcement distribution

• z-axis material behaviour

• pad diameter

• anti-pad diameter

• neighbouring return vias

• differential-via spacing

• copper roughness

• local cavities

• adjacent reference planes

For single-ended vias, εeff often approaches laminate Dk. However, geometry can shift the effective value.

Differential vias can show substantially different modal permittivity. A field solver is therefore required for final prediction. [Ref. 1]

The resonance sensitivity follows: f₁ ∝ 1/√εeff

Therefore: Δf/f ≈ −½ · Δεeff/εeff

A 10% error in εeff creates roughly 5% frequency error.

For the 28.85 GHz example:

5% frequency movement ≈ 1.44 GHz

The predicted notch could therefore move significantly. It may enter or exit the operational channel.

Manufacturing length tolerance creates another shift

Resonance also follows: f₁ ∝ 1/l

Consider a nominal 1.5 mm stub.

A variation from 1.4 mm to 1.6 mm produces:

• 1.4 mm → approximately 30.91 GHz

• 1.6 mm → approximately 27.04 GHz

That creates nearly 3.9 GHz of resonance movement.

One design can therefore produce different failures across production.

Rogers substrates improve control but not immunity

Rogers RO3003G2 specifies approximately:

• Process Dk: 3.00 ± 0.04 at 10 GHz

• Design Dk: 3.07 at 77 GHz

• Dissipation factor: 0.0011 at 10 GHz

Those properties support predictable mmWave transmission lines. However, the correct Dk must match the modelling method.

Using 3.00 instead of 3.07 shifts the example resonance. The estimate moves from 28.85 GHz to approximately 28.52 GHz. That difference already exceeds 300 MHz.

The material improves repeatability. It does not eliminate via-stub resonance. Low-loss materials can also preserve resonant energy. That may create sharper resonant behaviour. Loss reduction must therefore accompany correct transition design. [Ref. 3]

Pain Point 3: The stub interacts with return-path geometry

A signal via never operates alone. Its electromagnetic structure includes:

• signal barrel

• pads

• antipads

• reference-plane clearances

• nearby ground vias

• plane cavities

• package transitions

• connector grounds

• enclosure interfaces

The signal current moves vertically through the via. The return current must also change reference planes. Without nearby ground vias, that return path expands.

The loop inductance then increases. Common-mode conversion can rise. The transition may excite parallel-plate or cavity modes. This creates two simultaneous failure mechanisms:

1. Stub resonance

2. Return-path discontinuity

Back-drilling only addresses the unused barrel. It does not automatically fix the return-current geometry.

Likewise, adding ground vias does not remove the signal stub. Both structures require electromagnetic optimisation.

Why this matters for phased arrays

Hybrid beamforming reduces the number of full RF chains. It does not remove interconnect sensitivity. Each remaining channel still requires controlled amplitude and phase. Via tolerances can create frequency-dependent channel differences.

Those differences affect:

• beam pointing

• sidelobe levels

• null depth

• axial ratio

• cross-polarisation

• wide-angle calibration

A common resonant defect can reduce total array bandwidth. Random resonance variation creates channel mismatch. Both become harder during wide-scan operation.

Why this matters for Aerospace & Defense

Radar channels require stable group delay. Fast pulses depend on broadband phase linearity. A narrow resonance creates ringing in the time domain. That ringing can contaminate weak target returns.

A resonant via structure can increase common-mode excitation. That current may couple into cables and enclosure seams. MIL-STD-461 governs equipment-level EMI emissions and susceptibility. Its requirements cannot replace correct PCB electromagnetic design. [Ref. 4]

Why this matters for Satellite Communications

LEO payloads combine dense RF and digital assemblies. Rad-Hard layout practices improve radiation resilience. They do not prevent transmission-line resonance.

Multipactor mitigation addresses vacuum-discharge risks. It does not correct buried via-stub behaviour.

Ground stations face similar problems. High-gain receivers require low-loss, phase-stable transitions.

An unexpected Ka-band notch can directly reduce G/T.

The strategic engineering solution

Leading RF organisations control via stubs architecturally. They do not wait for prototype failure.

1. Route toward the far side whenever possible

Layer selection determines the remaining stub length. Routing deeper through the board can shorten unused copper. The best destination layer is not always the nearest layer. The electromagnetic barrel length matters more than visual convenience.

Every critical transition needs a documented stub-length calculation. Altium and Allegro rules should flag excessive unused barrel. However, rule checking cannot predict full modal behaviour.

2. Use blind or buried vias where justified

A blind via ends at its destination layer. It therefore avoids a through-barrel stub. This can provide the cleanest mm-Wave transition. However, blind vias introduce other considerations:

• sequential-lamination cost

• registration tolerance

• aspect-ratio restrictions

• reliability requirements

• stacked-versus-staggered micro-vias

• thermal-cycle durability

Rigid-Flex RF designs require additional caution. The rigid-to-flex interface can change reference geometry. A stubless via still needs a continuous return path.

3. Back-drill only with controlled residual length

Back-drilling removes unused plated copper mechanically. Keysight identifies back-drilling as a direct resonance remedy. Its demonstrations compare stubbed and back-drilled transitions using VNA data. [Ref. 5]

However, back-drilling is not a binary process. A residual stub always remains. The fabricator must control:

• drill depth

• depth tolerance

• drill diameter

• layer registration

• copper clearance

• breakthrough margin

• remaining stub length

One design guide reports residual stubs around 5–10 mils. Actual capability depends on the selected manufacturer.

For εeff = 3, those lengths predict:

• 10 mils, 0.254 mm → approximately 170 GHz

• 5 mils, 0.127 mm → approximately 341 GHz

Those resonances sit far above Ka-band. However, transition capacitance still remains. The back-drilled geometry must still be simulated. [Ref. 6]

4. Optimise the complete via field structure

Increasing anti-pad diameter reduces shunt capacitance. However, excessive clearance disrupts the return-current path. 

Smaller pads reduce capacitance. However, fabrication reliability can limit pad reduction.

Ground-via spacing controls the return-loop geometry. However, excessive periodicity can create additional modes. 

The transition therefore requires multi-variable optimisation. Relevant variables include:

• barrel diameter

• pad diameter

• anti-pad diameter

• ground-via radius

• ground-via count

• stub length

• plane spacing

• dielectric constant

• copper thickness

• connector geometry

This is not a two-dimensional impedance problem. It is a three-dimensional electromagnetic structure.

5. Use HFSS and SI/PI analysis correctly

A lumped via model may work below resonance. It becomes unreliable near distributed behaviour.

HFSS 3D Layout can model:

• vias

• pads

• anti-pads

• reference planes

• cavities

• return vias

• transmission-line launches

Ansys training examples specifically compare long vias and removed stubs. The workflow includes both S-parameters and TDR analysis. [Ref. 7]

A robust process should include:

1. Initial analytical resonance estimate.

2. Full-wave via-transition extraction.

3. Broadband S-parameter review.

4. Electric- and magnetic-field inspection.

5. Mixed-mode analysis for differential structures.

6. Time-domain reflectometry.

7. Tolerance and stack-up variation.

8. Correlation against manufactured coupons.

Do not inspect only S₂₁ magnitude. Also review:

• S₁₁

• insertion-loss phase

• group delay

• TDR impedance

• differential-to-common-mode conversion

• nearby-via coupling

• current density

• field confinement

6. Validate the manufactured structure

A successful simulation validates the CAD geometry. It does not validate the fabricated barrel. Production validation should include:

• impedance coupons

• via-transition test vehicles

• microsection inspection

• controlled VNA calibration

• fixture de-embedding

• back-drill-depth verification

• lot-to-lot comparison

• temperature testing

For high volume mm-Wave programmes, test multiple samples. One excellent transition does not prove manufacturing capability. Procurement teams should request statistical evidence.

The B2B procurement implication

A PCB drawing should not simply state “back-drill required.” That instruction leaves critical engineering undefined. A robust RF procurement package should specify:

• maximum residual stub length

• back-drill direction

• back-drill diameter

• depth tolerance

• keep-out around adjacent structures

• target operating bandwidth

• acceptable S₁₁ and S₂₁

• coupon geometry

• microsection sampling

• production acceptance criteria

For Massive MIMO and Phased Array hardware, also request:

• channel-to-channel phase variation

• transition loss across frequency

• temperature-dependent phase

• repeatability across assemblies

• common-mode conversion

• array-level calibration residuals

The cheapest PCB may not produce the lowest system cost. A failed mm-Wave transition can force:

• board redesign

• additional calibration

• amplifier margin

• reduced operating bandwidth

• lower manufacturing yield

• delayed regulatory approval

The strategic lesson is simple.

A via is not merely a plated hole.

At mm-Wave, it is a transmission line, resonator, and mode transformer.

The remaining question is more difficult:

Should every mm-Wave RF procurement package mandate maximum residual stub length, measured transition S-parameters, and production microsection evidence?