When does a trace become a transmission line?

Two different pictures of the same wire

In every introductory circuits class, wires are perfect. Connect a resistor to a battery with two wires, and Kirchhoff's laws assume the voltage at every point along each wire is the same — instantaneously, everywhere. That is the lumped model, and for DC circuits and low-frequency signals it works beautifully. It is what every schematic implicitly assumes.

But voltage changes do not teleport. A transition at one end of a copper trace travels as an electromagnetic wave at a finite speed — roughly 6 inches per nanosecond in a typical PCB. Until that wave reaches the far end, the far end does not know the driver has switched. For a 10 ns edge on a 1-inch trace, the delay is about 170 ps — negligible. For a 100 ps edge on the same trace, the delay is a significant fraction of the edge itself, and the voltage is measurably different at the two ends of the wire at the same instant.

When that happens, the wire is no longer acting like an ideal connection. It has become a transmission line — a structure with its own impedance, its own propagation delay, and reflections at every point where the impedance changes. The question this page answers is: for your trace length and your edge rate, which picture do you need?

The race that decides it

When the driver switches, an edge sets off down the line. It reaches the far end after one trip time, T_d. If the far end is not terminated, it reflects, and arrives back at the driver after 2T_d — the round trip.

Now the important part. The driver is not switching instantaneously either; it takes t_r to get from low to high. So there are two possibilities:

• The echo comes back while the driver is still switching. It blends into the edge that is still being produced. The waveform looks slightly distorted, and nothing about it reads as ringing. You can treat the wire as a wire.
• The echo comes back after the driver has finished. It lands on a waveform that had already settled, so it shows up as a distinct step — overshoot, then a reflection off the driver, then another step, and so on. That is ringing, and it is unmistakable.

So the test is a comparison between 2T_d and t_r. The usual form of it:

Note that nothing in that expression is a length until you convert it. The same six inches is a plain wire for a 20 ns edge and a badly behaved transmission line for a 100 ps one. A rule quoted in inches with no rise time attached is a rule that happens to be right for one era of silicon.

Putting numbers on it

Propagation delay comes straight from the dielectric the wave travels through: t_pd ≈ 84.7 · √D_k,eff ps/inch. A stripline is fully buried, so it sees the laminate's full D_k — around 4.2 for standard FR-4, giving roughly 173 ps/inch. A microstrip has air above it, so its effective D_k is lower, around 3.0, and it runs faster at about 147 ps/inch. Surface traces genuinely are quicker, which is why clock nets sometimes end up on an outer layer.

Feed a 100 ps edge into that: l_crit = 100 / (6 × 173) ≈ 0.1 inch. Two and a half millimetres. Essentially every routed connection on a modern board is electrically long, and the useful question is no longer “is this a transmission line?” — it is “which of these hundreds of transmission lines can I afford not to control?”

Why this decides how your board behaves

Crossing the threshold does not break anything by itself. What it changes is which failures become possible:

• Overshoot and undershoot. An unterminated line can drive a receiver pin well above the rail. That is a reliability question as much as a signal one — most input structures have an absolute-maximum spec that repeated overshoot eats into.
• Double clocking. If the ringing crosses the receiver's threshold on the way back down, a single edge is seen as two. On a clock or a strobe this is a hard functional failure, not a margin one.
• Settling time eating your setup budget. Even ringing that never crosses a threshold has to decay before the receiver samples. On a slow bus there is usually time; on a fast one the ringing is the timing failure.
• Crosstalk with a long tail. A ringing aggressor keeps coupling into its neighbour for as long as it rings, well after the edge that caused it.

What to do about it

Work in picoseconds, not inches. Convert every net's length to delay once, early, and keep the number. 2T_d against t_r is a comparison you can do in your head; inches against a remembered rule of thumb is not.

Decide by consequence, not by rule. Almost every net is electrically long, so the threshold cannot be your filter. Sort instead by what a reflection would cost: clocks and strobes first, because double clocking is functional; then anything sampled near the edge; then everything with slack to spare. A reset line that rings for 3 ns and is read once at power-up needs nothing.

If it is marginal, slow the edge rather than adding a resistor. Reducing t_r is the only lever that moves the threshold itself, and on a control net it is free. See edge rate, not clock rate.

Go deeper — where the /3 comes from, and when the rule stops applying

The divisor is a convention, not physics. Some houses use /2, some /6, and the tighter your overshoot specification the more conservative it needs to be. What sets it is how much of the reflected step lands on a settled waveform rather than on a moving one.

If the round trip is a fraction k of the rise time, then the echo arrives when the edge is a fraction k of the way through its transition, and roughly (1 − k) of the reflected amplitude adds to a waveform that is still moving. With 2T_d = t_r/3, about a third of the echo is absorbed into the edge and two thirds appears as overshoot — which for a typical CMOS driver into an open end is enough to be visible but usually not enough to be a problem. Tighten the rule to /6 and you are asking for the overshoot to be barely visible at all.

Three cases where the rule quietly stops applying:

• Multi-drop nets. The rule assumes one driver and one receiver on a uniform line. Add taps and every stub becomes its own reflection site with its own round trip. The relevant delay is no longer the end-to-end one.
• Capacitive loading. A receiver pin is not an open circuit; it is a few picofarads. That capacitance slows the edge locally and lowers the effective impedance of the line near it, which both changes the reflection and makes the measured T_d longer than the geometric one.
• Lossy lines. The rule assumes the echo comes back at full strength. On a long, lossy channel it does not — loss attenuates the round trip twice over, and a 16-inch backplane trace can be electrically very long and still barely ring. This is why loss and reflection trade against each other, and why the worst ringing is usually on short mismatched nets rather than long ones.

A last point about what the threshold does not tell you. Crossing it says reflections are now visible as distinct events. It says nothing about whether they are large. Amplitude is set by the reflection coefficients at the two ends, and those depend on impedances, not on length. A well-matched 20-inch trace is electrically enormous and perfectly quiet; a badly matched half-inch stub at the wrong frequency is short and catastrophic. Length decides whether you see the reflection separately; impedance decides how big it is.

In the real world

The rule's practical value today is not in deciding which nets are transmission lines — they all are — but in giving you a defensible way to say which ones you have chosen not to worry about. That distinction matters at review time. A fixed length rule is easy to state, while “we terminated the nets where a reflection could double-clock a receiver, and here is the list” connects the decision to an observable failure mechanism.

The failure mode to watch for is an unclassified net: a strap, a jumper, a test point, a route to a connector that was added late. Those are the ones that get routed without a reference plane underneath, and they are electrically long for the same reason everything else is.

Related

• Signal Bandwidth: Why Edge Rate Matters — where t_r comes from and how to slow it
• Transmission-Line Reflections and Ringing — what happens once you are over the threshold
• Transmission-Line Termination Schemes — the fixes, and their price
• Transmission-Line Characteristic Impedance — what decides the size of the reflection
• Lab A — travelling waves — watch the round trip happen