Via Step Response & Rise-Time Degradation Calculator
Model a PCB via as a lumped LC discontinuity: excess capacitance and inductance, self-resonant frequency, 10–90 % rise-time degradation, and peak reflection for a given edge rate.
Lumped LC via (IPC-2141A C) · Johnson–Graham rise-time degradationAPI slug via-step-response
Inputs
The edge through the via
sample point
Standard values of Via Length
Select a row to use it| Via Length | Output Rise Time | Edge Slowdown | Peak Reflection | Status |
|---|---|---|---|---|
| 400ps | 0.0099% | 0.64% | PASS | |
| 400ps | 0.011% | 0.7% | PASS | |
| 400.1ps | 0.013% | 0.7% | PASS | |
| 400ps | 0.0087% | 0.6% | PASS | |
| 400ps | 0.0028% | 0.34% | PASS |
- Via InductanceL
- 1.139nH
- Via CapacitanceC
- 571.7fF
- Excess CapacitanceCexc
- +116fF
- Excess InductanceLexc
- −290pH
- Via Impedance √(L/C)Zvia
- 44.6Ω
- LC Self-Resonancef₀
- 6.236GHz
- Signal Knee Frequencyfknee
- 1.25GHz
Formula
H. Johnson & M. Graham, "High-Speed Digital Design", Prentice Hall 1993; IPC-2141A (via capacitance)
| Symbol | Quantity | Your value |
|---|---|---|
| L | Via barrel inductance (nH) | 1.139 nH |
| C | Pad-to-antipad capacitance (pF) | 571.7 fF |
| Cexc | Capacitance in excess of a matched section (pF) | +116 fF |
| Z₀ | Impedance of the line the via sits in (Ω) | 50 Ω |
| tᵣ | 10–90 % rise time (ps) | 400 ps |
| f₀ | LC self-resonant frequency of the via (GHz) | 6.236 GHz |
| fknee | Knee frequency, 0.5/tᵣ (GHz) | 1.25 GHz |
Assumptions
5 items
- 01The via is a lumped L and C: L = 0.2·h·(ln(4h/d) + 0.5) nH with h and d in mm (Johnson and Graham), and C = 0.0554·εr·h·D₁/(D₂ − D₁) pF from the pad and antipad diameters (IPC-2141A).
- 02Against the line impedance Z₀ the via is an excess capacitance C − L/Z₀² or an excess inductance L − C·Z₀²; it slows the edge by 2.2·(Z₀/2)·C_ex or 2.2·L_ex/(2Z₀), added to the input rise time in quadrature.
- 03The peak reflection is Z₀·C_ex/(2·t_r) or L_ex/(2·Z₀·t_r), for a ramp edge.
- 04The lumped model holds only well below the via's self-resonance 1/(2π√(LC)); a warning names a resonance within 3 times the knee frequency 0.5/t_r. Stubs, the return path between planes and the dielectric's loss are not modelled.
- 05The slowdown is judged against a budget, at most (10 % by default), which the slowdown warning follows; the drawing is the input and output edges as Gaussian edges, whose rise times add in quadrature as the model's do; the standard values are board thicknesses 0.8 to 3.2 mm.
Worked example
A PCIe Gen4 edge through a 1.6 mm via
A 30 ps PCIe Gen4 edge on a 50 Ω line crosses a full 1.6 mm via with a 0.3 mm drill, 0.6 mm pad and 1 mm antipad in εr 4.3. How much does it slow the edge?
| Step | Expression | Result |
|---|---|---|
| Barrel inductance | 0.2 × 1.6 × (ln(4 × 1.6 / 0.3) + 1/2) | 1.139 nH |
| Pad capacitance | 0.0554 × 4.3 × 1.6 × 0.6 / (1 − 0.6) | 571.7 fF |
| Excess capacitance | 0.5717 − 1.139 × 1000 / 50² | +116 fF |
| Edge degradation | 2.2 × 50 / 2 × 0.116 | 6.381 ps |
| Output edge | √(30² + 6.381²) | 30.67 ps |
| Slowdown, within the 10 % budget | (30.67 − 30) / 30 × 100 | 2.2 % |
How it works
A via is short compared with the wavelength of any signal it is likely to carry, so it is modelled not as a transmission line but as a lumped discontinuity: one series inductance from the plated barrel and one shunt capacitance between the pad and the surrounding antipad. The barrel inductance follows the short-round-conductor form nH with and in millimetres, and the pad-to-antipad capacitance follows IPC-2141A as pF. Note that inductance grows with barrel length but only logarithmically with diameter — shortening a via helps far more than fattening it.
The crucial point is that neither nor alone tells you whether the via hurts. An ordinary section of transmission line also has inductance and capacitance; what a receiver actually sees is the imbalance between them. Subtract what a matched section would have contributed and you get the excess: and . Exactly one of these is positive, and it identifies the character of the discontinuity. A via with has zero excess and is electrically invisible no matter how large and are individually.
From the excess the timing cost follows directly. A small shunt capacitance is charged through the source impedance in parallel with the load, , giving a first-order time constant ; converting to a 10–90 % edge multiplies by . A small series inductance gives by the same argument. Cascaded first-order edges combine approximately in quadrature, so . The consequence is that the same via is harmless on a slow edge and severe on a fast one: a 6 ps degradation costs 0.2 ps on a 100 ps edge but doubles a 6 ps edge.
Reflection behaves differently again. The peak reflected amplitude from a small discontinuity is — inversely proportional to rise time, so it grows in direct proportion to edge rate rather than in quadrature. Finally, and resonate at . This structural self-resonance is distinct from the resonance of an unused barrel stub, and once the signal knee frequency approaches it the lumped model becomes optimistic.
Practical tips
- Aim for Z_via close to the line impedance rather than for minimum capacitance. The excess-capacitance output crossing zero is the optimum, and it is directly visible as you adjust the antipad
- Remove non-functional pads on layers the via does not connect to. This is usually the largest single reduction in via capacitance available and costs nothing in fabrication
- Back-drill long vias on high-speed nets. Shortening the barrel cuts inductance and capacitance together, which moves the self-resonance up rather than trading one for the other
- Check the knee frequency against the self-resonance before trusting any lumped via model. If the knee is within about a third of f0, only a 3D field solver will give a reliable answer
- For differential pairs, model each via of the pair separately at the single-ended impedance — a 100 ohm differential pair has 50 ohm single-ended lines, so use 50 ohms here
Common mistakes
- Judging a via by its raw capacitance — 0.57 pF sounds alarming, but four fifths of it is matched by the barrel inductance and only the 0.12 pF excess is a discontinuity. Compare against L/Z0-squared, not against zero
- Maximising the antipad diameter on the assumption that less capacitance is always better — push C below L/Z0-squared and the via flips inductive and the reflection climbs again. Growing the default antipad from 1.0 to 2.0 mm makes the reflection worse, from 2.9 percent to 7.3 percent at a 100 ps edge
- Designing to clock frequency rather than edge rate — a 100 MHz clock with a 50 ps edge stresses a via far harder than a 500 MHz clock with a 500 ps edge, because both reflection and degradation track the rise time
- Confusing this LC self-resonance with the via stub quarter-wave resonance — they are different mechanisms with different frequencies, and the stub resonance is usually the lower and more damaging of the two
- Using board thickness for the barrel length on a back-drilled via — after back-drilling the signal only traverses the remaining stub-free length, and using the full thickness overstates both L and C substantially
Frequently asked questions
6 questions
Why does the via barely matter at 400 ps but dominate at 20 ps?
Rise times combine in quadrature. A 6 ps degradation on a 400 ps edge gives 400.05 ps, a 0.01 percent effect. The same 6 ps on a 20 ps edge gives 20.9 ps, a 4.5 percent effect. Reflection is worse still because it scales as 1 over the rise time directly, not in quadrature, so it grows twenty-fold over that same range.
What is excess capacitance and why not just use C?
A matched transmission line of the same physical length also has capacitance, and that portion costs nothing because it is part of a properly terminated line. Only the amount beyond L divided by Z0 squared is a discontinuity. A via with square root of L over C equal to 50 ohms has zero excess and is electrically invisible in a 50 ohm system.
How do I make a via less capacitive?
Enlarge the antipad, remove non-functional pads on unused layers, and shrink the pad to the minimum annular ring the fabricator allows. But stop at C equal to L divided by Z0 squared — going further makes the via inductive and the reflection rises again. Watch the excess-capacitance output cross zero.
Is this the same as via stub resonance?
No. The quarter-wave stub resonance comes from an unused length of barrel acting as an open transmission line, producing a deep narrow insertion-loss null. The f0 here is the lumped LC self-resonance of the via structure itself. Both should stay well above your knee frequency, and the stub resonance is usually the lower one.
When does the lumped model stop being valid?
When the via is no longer electrically short compared to the signal — practically, when the knee frequency reaches about a third of the LC self-resonance, or when the computed reflection exceeds roughly 30 percent. Past that point L and C interact as a distributed structure. The calculator warns in both cases.
Why does the capacitance formula use the full barrel length?
IPC-2141A treats the via as a coaxial section spanning the layers it passes through, so the barrel length sets the plate area. If your via only crosses two planes rather than a full stack-up, use the length between those planes instead of the board thickness — the result scales linearly with it.
Shop Components
As an Amazon Associate we earn from qualifying purchases.