RobustPCBWhere Robust Design Starts
CalculatorsالعربيةUpgradeSign inSign up

What's new

Full changelogEngine 1.18.3

Via Stub & Back-drill

PCB & Copper

Resonance of an unused via barrel against the signal's Nyquist or knee frequency, and the longest acceptable stub.

Inputs

Board profile

Fills Dielectric constant from your board profile.

Unused barrel below the last layer the signal uses
Dk at the signal frequency. A via sees glass and resin, so allow about ±10 %.
Resonance ÷ reference frequency. 3 is a common target.

Results

1 passed
Stub resonance fres

quarter wave · 6.2× the Nyquist frequency
31.23GHz
Pass

Checked: fres ≥ 3.0 × the Nyquist frequency (R / 2) of 5 GHz
Longest acceptable stub lmax

for 3.0× the Nyquist frequency
2.498mm
Margin achieved

target 3.0 ×
6.2 ×
All results (2)
Reference frequency

Nyquist frequency (R / 2)
5GHz
Stub delay, one way ts

150 mm/ns in εr = 4.00
8.006ps
The stub resonates at 31.2 GHz, 6.2× the Nyquist frequency (R / 2) of 5 GHz. That clears the 3.0× target.

Round trip through the stub: 16 ps. Back-drilling leaves a short remnant, and the remnant is the new stub; ask the fab for its depth tolerance.

Stubs at the two ends of a link do not add: each makes its own notch. Connectors and packages can carry stubs too.

quarter-wave resonance

Parameter sweep

Vary one input over a range and see the answer and verdict at each step, as a table and a curve.Pro

Worst-case corners

Put a tolerance on each input and get the worst-case band around the answer.Pro

Give at least one input a tolerance above zero.

Cross-section

The principle

When a signal leaves a through via part-way down, the rest of the barrel is an open stub. Where the stub is a quarter wavelength, its reflection cancels the signal and cuts a loss notch. The notch is broad, so keep the resonance well above the highest frequency that matters.

Quarter-wave resonance
fres=vp4 ls=c4 lsεrf_{res} = \frac{v_p}{4\,l_s} = \frac{c}{4\,l_s\sqrt{\varepsilon_r}}
A λ/4 open stub looks like a short at its junction. Odd multiples also resonate; the first is the one in the band.
Stub delay
ts=lsεrct_s = \frac{l_s \sqrt{\varepsilon_r}}{c}
The reference frequency
fNyq=R2fknee=0.35trf_{Nyq} = \frac{R}{2} \qquad f_{knee} = \frac{0.35}{t_r}
Nyquist for an NRZ data rate R; the knee for a 10 to 90 % rise time.
Longest acceptable stub
lmax=c4 m frefεrl_{max} = \frac{c}{4\,m\,f_{ref}\sqrt{\varepsilon_r}}
m is the margin, fres ÷ fref. Three is a common target.
  • lsl_sstub length, the unused part of the barrel
  • vpv_ppropagation velocity in the dielectric
  • εrε_rdielectric constant at the signal frequency
  • RRdata rate, bits per second (NRZ)
  • trt_r10 to 90 % rise time
  • mmresonance margin, a ratio
More detail

The resonance is not a hard wall: loss grows as the stub approaches a quarter wave, and the eye closes before the notch reaches Nyquist. A 3× margin puts the notch beyond the third harmonic, where an NRZ spectrum has little energy left.

Removing the stub

  • Back-drill. The fab drills out the unused barrel from the far side. The remnant is the new stub; ask for the depth tolerance.
  • Route on the far layer. A top-to-bottom transition uses the whole barrel and has no stub. Choosing the layer pair costs nothing.
  • Blind or buried vias. No stub, at the cost of a more complex stack-up.
10 Gb/s through a 1.2 mm stub in ε_r = 4
v_p = c/2 = 1.5 × 10⁸ m/s, so f_res = 1.5 × 10⁸ / (4 × 0.0012) = 31 GHz, 6.3× the 5 GHz Nyquist frequency. Fine.
The same signal through a 3 mm stub
f_res = 12.5 GHz, only 2.5× Nyquist: the notch skirt is in the band. The longest stub for a 3× margin is c / (4 × 3 × 5 GHz × 2) = 2.5 mm; back-drill below that.
References: Bogatin, Signal and Power Integrity, Simplified, on via stubs and back-drilling. Johnson and Graham, High-Speed Signal Propagation, on the knee frequency and stub resonance. IPC-2141A on propagation velocity. Related: Controlled Impedance & Delay for the line the via joins.

Engine version ⁨1.18.3⁩