Impedance Matching: When It Actually Matters

Impedance matching is often applied as a blanket rule, with terminators added to nets that do not need them and controlled impedance demanded on boards where nothing runs fast enough to care. The result is extra components, extra board area and extra cost, in exchange for a benefit that is not measurable. Knowing when the effect actually appears turns the rule into an engineering decision.

This article explains the criterion that decides whether matching matters, where it should be applied, and where the effort is better spent elsewhere.

The Criterion Is Time, Not Frequency

Whether a trace behaves as a transmission line depends on how the propagation delay along it compares with the rise time of the signal. The accepted rule is that a trace shorter than about one sixth of the electrical length of the edge can be treated as a lumped connection, because the reflection returns while the edge is still transitioning and merely alters its shape slightly.

For a signal with a one nanosecond edge, that length is of the order of a few centimetres on a typical laminate. For a signal with a ten nanosecond edge, it is ten times longer. A one hundred megahertz clock with a slow edge and a short route behaves like a wire, while a twenty megahertz clock with a fast driver and a long route does not.

<img src="https://www.gopcba.com/wp-content/uploads/2026/05/Industrial-IOT-PCBA-1.jpg" alt="Series termination resistor at a driver output” />

Where Matching Is Necessary

Matching earns its cost where the trace is long relative to the edge, where several receivers load the same net, where the driver is fast and the receiver has a narrow input threshold, or where an interface specification requires a defined characteristic impedance. High speed serial links, memory buses, clock distribution and any net that leaves the board through a connector are the usual candidates.

It is also necessary where emissions are a concern. An unmatched line with a large reflection radiates the reflected energy as common mode noise, and the effect is not confined to the fundamental. A matched line is quieter as well as cleaner, which is an argument that applies even where the receiver would tolerate the reflection.

The same criterion applies to crosstalk, and it is worth separating the two effects. Coupling between adjacent traces depends on the mutual capacitance and inductance per unit length and on how long the traces run in parallel, while reflection depends on the impedance profile of a single line. A short parallel run between two nets can be perfectly acceptable on a board where a long unmatched line is not, and treating both problems with the same remedy usually costs more than it saves.

Where It Is Wasted

Short point to point nets between two devices that are adjacent on the board are the clearest case. Adding a series resistor to such a net changes the drive strength and the rise time, which may actually reduce the timing margin if the resistor was not part of the original driver specification. The same applies to low speed control lines, to enable pins and to status signals.

Demanding controlled impedance across an entire board is the other common over-application. Fabrication cost rises with the number of controlled layers, and the tolerance that can be held is finite, so the requirement should be attached to the nets that need it and to the layers on which they run. Marking every net as controlled impedance produces a more expensive board with the same performance. Where the budget is tight, the effort is better placed on the interfaces that are actually marginal, which is where a focus on low cost signal quality improvements pays off most.

Termination Choices

Where matching is needed, the method depends on the topology. A series resistor at the driver suits a point to point net with one receiver, dissipates no static power and adds no load to the driver beyond the resistor. A parallel termination at the receiver suits a bus and consumes current continuously. A Thevenin pair biases the line at mid supply and halves the static current at the cost of two resistors.

For differential pairs the matching is usually built into the geometry rather than added as a component. The pair is routed to a defined differential impedance, and the termination resistor at the receiver is placed as part of the interface specification. Getting the geometry right matters more than the terminator, because a mismatched pair with a perfect terminator still reflects at every discontinuity along its length.

Electrical Length And Physical Length

The length that matters is electrical, not physical, and the two are related by the effective dielectric constant of the structure. A microstrip trace on the outer layer has more of its field in air than a stripline trace buried between two planes, so it propagates faster per unit length. Two traces of the same physical length can therefore have different electrical lengths, and a matching calculation based on physical length alone carries an error.

This is why the same route on an inner layer and on an outer layer behaves differently, and why a design that changes the microstrip to stripline transition part way along a trace introduces a change of propagation velocity at that point. The discontinuity is small, and in a matched system it is one of many small effects that together consume the margin.

Verifying That The Effort Worked

A matched design should be measurable. The oscilloscope can show whether the edge is clean at the receiver and whether there is visible ringing at the transitions, and the eye diagram at the interface shows the accumulated result. Where a time domain reflectometer is available, the measured impedance profile can be compared with the intended one, which localises any remaining discontinuity to a specific location on the trace.

The characteristic impedance the fabricator measures on a coupon is the starting point of that comparison rather than the end of it. Coupon data confirms the process, the measurement on the board confirms the design, and the two can differ for reasons as simple as a plane split under a connector footprint.

A practical way to decide is to write the rise time and the route length for each net class into a table and apply the one sixth criterion mechanically. The nets that pass do not need matching, the nets that fail need the impedance controlled and, if the reflection is significant, terminated as well. The table takes twenty minutes and replaces a rule of thumb that is applied to everything and justified on nothing.

FAQ

Do I need matching on a twenty megahertz SPI bus? Usually the bus needs attention for crosstalk rather than for reflection, because the edge is fast but the route is short. Series resistors reduce the edge rate, which helps both problems at once, and that is often the entire fix.

Is a series resistor always a safe addition? No. It slows the edge and reduces the drive, which can violate the setup time at the receiver if the timing budget was already tight. It should be chosen from the driver impedance and the line impedance rather than by habit.

What happens if a controlled impedance board is built without the requirement? The impedances will be whatever the geometry and the laminate produce, which may still be acceptable for a slow design and will be a lottery for a fast one. The saving is real and so is the risk, and the decision should be made deliberately.