Via Stub Resonance and Backdrill Calculator
Enter the board thickness, the layer your signal leaves on and the data rate. The calculator returns the stub length, where it resonates, how close that is to your signal band, and how much barrel a backdrill would need to remove.
| Propagation velocity in the via | |
| Longest stub that still meets your ratio | |
| Backdrill depth required (from the far side) |
Runs entirely in your browser. Nothing is uploaded, stored, or sent anywhere.
What a via stub does
A plated through-hole goes all the way through the board whether or not your signal does. If the signal enters at the top and leaves on an inner layer, the barrel below that exit layer carries no useful current — but it is still connected, still conductive, and still an open-circuited transmission line hanging off the signal path.
An open stub is not harmless. At the frequency where it is a quarter wavelength long, the open end reflects back to the junction inverted, and the stub presents a short circuit to the through path. The channel gets a deep notch at that frequency, and a broad loss penalty approaching it. Below about a quarter of that frequency the stub is just a small capacitance and nobody notices, which is why stubs were a non-issue for decades and became a design constraint the moment serial links passed a few gigabits.
v = c / √Dk ·
fres = v / (4 · Lstub) ·
fNyquist = data rate / 2Reading the ratio
What matters is not the resonant frequency on its own but its distance from the signal’s content. Nyquist — half the NRZ data rate — is the fundamental of the fastest alternating pattern the link carries, and a common working rule is to keep the stub null at least three times above it. That is a rule of thumb, not a specification: it buys margin for the third harmonic and for the fact that the null is broad, not sharp.
The defaults show why this became a problem. A 1.6 mm board with the signal exiting near the top leaves a 1.2 mm stub, resonating around 31 GHz — entirely irrelevant at 1 Gbps and comfortable at 10. Push the rate to 25 or 56 Gbps and the same stub that was invisible becomes the dominant channel impairment. Nothing about the board changed.
The three fixes, in order of preference
- Choose the exit layer. The cheapest stub is the one that does not exist. Route the high-rate pairs so they leave near the far side of the board, and the barrel below them is short by construction.
- Backdrill. Drill the unused barrel away from the far side with a slightly larger bit after plating. It works well and is routine at high rates, but it costs money, needs a stated stub-length tolerance (typically a few tenths of a millimetre of remaining stub), and the fabricator must be told — it is not a default.
- Blind or buried vias. They remove the problem entirely and cost considerably more, so they earn their place on dense HDI designs rather than as a stub fix alone.
What this estimate simplifies
- The barrel’s impedance is set by the antipad and the surrounding plane cavity, and it is generally not the same as the trace impedance — so the notch is neither infinitely deep nor perfectly placed.
- Dk is not constant with frequency, and the effective value a via sees differs from the value quoted for a trace on the same material.
- Backdrilling leaves a residual stub by design. Ask your fabricator what tolerance they hold before assuming it goes to zero.
- Pad and antipad capacitance, plated-barrel roughness and the launch discontinuity all shape the real response and are absent here.
Where this fits
Signal integrity is one of our primary service lines. Deciding stub budgets, backdrill depths and exit layers is exactly what a pre-layout constraint study settles — one to two weeks, before routing starts, when all of it is still free to change. Post-layout, the same questions get answered against the real geometry in Ansys SIwave or Siemens HyperLynx.
More free tools
Each of these runs entirely in your browser. Nothing is uploaded, stored or sent anywhere, and none of them needs an email address.