At what length does a trace become a "transmission line"?

Does the voltage finish changing before the edge makes a round trip?

Drive a signal into one end of a trace and it takes a propagation delay t_pd to reach the other end — roughly 6–7ps/mm for FR-4 microstrip, and around 7ps/mm for stripline. A 50mm trace is 300–350ps one way.

If the signal's rise time t_r is much longer than that round trip (2·t_pd), reflections bounce back and forth many times and settle out while the edge is still changing, so the whole trace can be treated as a single voltage. That's the condition for lumped-element treatment. If t_r is shorter than the round trip, there's a moment when the near end and far end sit at different voltages, and reflections show up as visible artifacts in the waveform. That's distributed behavior — a transmission line.

A common rule of thumb: lumped-element treatment holds if the trace delay is 1/6 or less of the rise time; a more conservative design uses 1/10. In terms of length, that's a critical length ℓ = t_r / (6 × t_pd).

Clock frequency is beside the point

Even a 1MHz signal, if driven by a device with a 1ns edge like a 74LVC part, carries edge bandwidth of 0.35 / t_r ≈ 350MHz. What triggers reflections is not the 1MHz fundamental — it's that 350MHz-equivalent edge.

So "it's a slow signal, the trace doesn't matter" is wrong; the correct statement is "it's a slow-edged signal, the trace doesn't matter." Modern MCUs and logic ICs often have faster edges than their predecessors, to save power, so the same part number from ten years ago may have had a longer critical length.

The real symptoms are overshoot and undershoot at the receiver, ringing, and double-clocking as the waveform crosses back and forth through the threshold. It's not unusual for an intermittent "count is sometimes one too many" bug on a 1MHz clock to trace back to ringing on a 30mm trace.

Past the critical length, terminate

Once a trace exceeds its critical length, something has to tame the reflections. For a simple point-to-point trace on a board, the easiest fix is a series termination at the driver (a damping resistor): pick the resistor so it, plus the driver's own output resistance (roughly 10–30Ω for CMOS), sums to the trace's characteristic impedance (around 50Ω). That's why 22–33Ω values show up so often.

Series termination works by halving the amplitude at the driver, then doubling it back to full swing when it reflects off the open far end. The correct full amplitude only appears once, right at the far end — any receiver tapped in along the way sees a period at half amplitude. It doesn't suit a trace with multiple receivers.

For multiple receivers, or when the far end needs the correct amplitude at all times, use parallel termination at the receiver end (a resistor matching the trace's characteristic impedance, to GND or to a termination voltage). That draws continuous DC current, raising power consumption and demanding more from the driver.

Before deciding whether termination is needed at all, check whether the trace can simply be shortened below its critical length, or whether a slower-edged driver would do. Termination treats the symptom — a short enough trace needs none.

For a sine wave (RF), it's 1/10 of a wavelength

Everything above concerns digital edges. For RF or high-frequency analog signals treated as sine waves, compare the trace length with the wavelength instead. The rule of thumb is 1/10 of a wavelength or less for lumped-element treatment.

FR-4 microstrip has an effective dielectric constant around 3, so its wavelength is about 1/√3 ≈ 0.58 of the free-space value: roughly 1.7m at 100MHz and 17cm at 1GHz, giving λ/10 values of 17cm and 17mm respectively. At 2.4GHz, λ/10 ≈ 7mm — a regime where even a few-mm feed line to an antenna needs impedance matching.

The digital critical length and the RF λ/10 rule are really the same idea, once you convert edge bandwidth 0.35 / t_r into a frequency: a 1ns edge ≈ 350MHz ≈ λ/10 of 48mm. That's looser than the 1/6 rule's 26mm because a sine wave is a single frequency, while an edge carries even higher-frequency content.

The decision procedure

1. Read the rise time t_r off the driver's datasheet. If it isn't listed, use the transition time of another output in the same family, or measure it (with a scope that has enough bandwidth).
2. Get t_pd from the trace length and layer, and calculate the one-way trace delay. The trace length to delay calculator does this directly.
3. If one-way delay × 6 is shorter than t_r, treat it as lumped; if longer, treat it as a transmission line. The lumped/distributed calculator implements exactly this check.
4. If it's a transmission line, first see whether the trace can be shortened; if not, settle on a characteristic impedance and terminate.
5. On the prototype, check the far-end waveform with an active probe or a short GND connection, and confirm any overshoot stays within the receiving IC's absolute maximum rating (e.g. VCC + 0.5V).