Lumped or distributed
Where the boundary between lumped and distributed behaviour falls. Enter a frequency and you get the longest trace you can still treat as lumped. Enter a trace length and you get the highest frequency at which it still counts as lumped.
SI prefixes accepted (4k7 / 1M / 10m / 220). Upper-case M = mega, lower-case m = milli
Advanced — Dielectric thickness, Copper thickness, PCB relative permittivity
History
Formula
Limiting length: L = 0.25 / (2π × f × √(L₀ × C₀))
(L₀, C₀: inductance and capacitance per unit length)
Design notes
A component behaves as a lumped element as long as it is much smaller than the wavelength of the signal. At high frequency, parts and traces approach the wavelength and you can no longer ignore propagation delay or the phase shift along them — that is the distributed regime.
Rule of thumb: below a tenth of a wavelength you can design lumped.
On FR-4 (εr = 4.3) at 1 GHz:
→ wavelength on the board ≈ 145 mm → limiting trace length ≈ 14.5 mm
Circuits that need distributed treatment: RF, high-speed serial buses (PCIe Gen4 and up), microwave filters.
When you need this
This answers "does this net need termination?". It gives the boundary — in both length and frequency — between treating a trace as a plain wire (lumped) and as a transmission line (distributed).
The criterion
If a trace is short compared with the signal wavelength, the voltage is essentially the same everywhere along it. That is the lumped regime and no termination is needed. Once the length approaches the wavelength, voltage varies along the trace — the distributed regime — and reflections and termination must be considered.
This calculator uses an electrical length of βl = 0.25 radians (about 14°) as the boundary, roughly λ/25. That is more conservative than the λ/10 figure textbooks often quote. Beyond the length reported here, treat the net as a transmission line.
The right input is edge rate, not clock frequency. A 10MHz clock with a 1ns edge contains content up to roughly 350MHz (f = 0.35/tr). That is what causes reflections. Enter 0.35/tr, not the clock rate.
Guideline by edge rate
| Rise time | Equivalent frequency | Length where termination matters |
|---|---|---|
| 10 ns | 35 MHz | about 180 mm |
| 5 ns | 70 MHz | about 89 mm |
| 2 ns | 175 MHz | about 35 mm |
| 1 ns | 350 MHz | about 18 mm |
| 0.5 ns | 700 MHz | about 8.9 mm |
| 0.2 ns | 1.75 GHz | about 3.6 mm |
Current logic families routinely have sub-nanosecond edges, which means anything over about 20mm is a transmission line. "It is a slow bus, it will be fine" stops being true as soon as the device generation changes.
If it is distributed
- Series termination (damping resistor). 22–33Ω right at the driver, so that the driver output impedance plus the resistor matches the trace. Easiest for point-to-point and adds no static current.
- Parallel termination. A resistor matching the trace impedance at the receiver, to ground or supply. Cleanest waveform, but it draws current continuously.
- Thevenin termination. Resistors to both rails, terminating to a mid-level. Easier to bias than parallel termination, but still draws current.
- Simply make it shorter. The most reliable fix. If rearranging the placement gets you under 20mm, the question disappears.
Frequently asked questions
Should I enter the clock frequency?
References usually say λ/10. Is this different?
If I am only slightly over, will it still work?
Does the same apply to differential pairs?
Standards and references
- IPC-2141A — When to treat a trace as a transmission line, and termination design.
- Howard Johnson, "High-Speed Digital Design" — The rise-time-based lumped/distributed criterion.
Last updated: 2026-08-29