Reducing RF Effects in PCB Interconnect Design

Below a few hundred megahertz a board can usually be treated as a set of ideal connections. Above that, the same copper begins to behave like a transmission line with delay, loss and coupling attached, and the layout decisions that were harmless at lower frequencies start to dominate performance. Understanding RF effects is not only a concern for radios; the fast edges of modern digital interfaces reach the same territory, and a board that ignores them will show ringing, radiation and intermittent link failures.

How RF Effects Show Up on a Board

The symptoms are recognisable once you know what to look for. A clock line that rings at the receiver, an antenna that performs worse on the product than it did on the bench, a differential pair that fails an eye mask only after the cable is connected. All of these point to the same underlying physics: signals are waves, and any change in the impedance along their path creates a reflection.

Coupling is the second mechanism. Two traces running in parallel exchange energy, which raises crosstalk, and a trace running over a split in the reference plane forces the return current to take a long detour, which radiates. Neither problem is about the signal trace itself; both are about controlling the path that current takes on its way back to its source.

Return Paths and Ground Plane Discipline

Every high frequency signal has a return current that follows the path of least impedance, which at high frequency means the reference plane directly beneath the trace. Interrupting that plane forces the return current around the obstacle, and the resulting loop behaves like a small antenna. Keep the reference plane continuous under every fast trace, and never route a fast signal across a plane split.

Where a solid plane is not available, a ground stitch via on either side of the crossing keeps the return loop small. The same reasoning applies to the transition between a microstrip and a stripline in a multilayer stack, which our notes on microstrip and stripline routing explain layer by layer. The rule is short and memorable: follow the current, not the trace.

Via Stubs and Layer Transitions

A via that connects two inner layers leaves an unused section of barrel below the lower layer, and that stub behaves like a resonant transmission line. At low frequencies it is invisible; at several gigahertz it creates a notch in the through response. Back drilling removes most of the stub, and blind or buried via constructions avoid it by design.

Transitions also introduce their own inductance and capacitance. A signal via without an adjacent return via makes the return current detour, and a large pad or antipad adds capacitance that shows up as a small reflection. Where many fast signals change layers, place the return vias in a regular pattern so the return path is predictable, and keep the antipad geometry consistent across the board.

Impedance Matching and Termination

A trace carries a characteristic impedance set by its width, its height above the reference plane and the dielectric constant of the laminate. Where every point along the route has the same impedance, reflections vanish and the design becomes predictable. Where the impedance changes, whether at a connector, a layer transition or a change in trace width, part of the signal reflects back.

Termination is the second half of the same idea. Series resistors at the driver, parallel terminations at the receiver, or differential terminations across a pair absorb energy and prevent it from bouncing. Choose the scheme from the topology rather than by habit, and place the components as close to the relevant pin as the layout allows, because the stub from the pin to the resistor is itself part of the RF problem.

Shielding and Separation

Shielding is not only a metal can over a radio. Most of the benefit comes from layout discipline: keeping aggressor and victim apart, running ground between them where they must be close, and giving sensitive analogue or RF nodes their own local reference. A ground stitch fence along both sides of a noisy net is often more effective, and far cheaper, than a shield component.

Where a metal shield is fitted, it has to be bonded to ground at frequent intervals, otherwise it becomes a resonator rather than a barrier. Our overview of EMI suppression design principles covers the same tools from the emissions side, and the techniques overlap almost completely because both problems are about controlling current loops.

Connectors, Cables and Interconnect Boundaries

Many RF effects that appear on the bench are actually created at the boundary between the board and the outside world. A connector pin with a long internal stub, a pigtail drain wire instead of a proper shield termination, or a cable shield bonded at one end only will all degrade performance in ways no amount of board tuning can fix. Specify the connector as part of the RF design rather than as a mechanical afterthought.

Inside the board, treat the interface between fast digital and radio sections as a boundary with its own rules. Keep the two regions physically apart, provide a clean return path under the crossover, and filter any signal that must cross. The techniques for routing high frequency traces and data buses apply to both sides of that boundary.

Design Checks Before Release

In any interconnect design review, walk the fast nets one at a time and confirm three things: the reference plane is continuous, the impedance is controlled along the whole route, and the return path is short at every layer transition. These checks cost an afternoon and remove most of the surprises that appear during EMC testing.

Add a review of the stackup against the actual fabricator capabilities, since an impedance target that cannot be manufactured is a documentation error rather than a design. Where the project includes a radio, keep the antenna keep-out and matching network on the same checklist, because those are the parts where a small layout error has the largest measurable consequence.

FAQ

At what frequency do RF effects start to matter? The practical answer is that they matter when the rise time of the signal is short compared with the propagation delay of the trace, which for fast logic means traces longer than roughly one sixth of the wavelength of the highest significant frequency. Many modern digital interfaces cross that line at clock rates that would have seemed modest a decade ago.

Does a ground plane alone solve RF problems? It solves the return path problem, which is the largest single contributor, but it does not control impedance discontinuities, stub resonances or connector transitions. Those need attention to geometry and to termination. A continuous plane is the foundation rather than the whole solution, and it is worthless if the plane is split under a fast net.

How do I know whether a via stub will hurt my design? Estimate the stub length, which is the distance from the lower connected layer to the bottom of the barrel, and compare it with the wavelength of the highest frequency of interest. As a rule, keep the stub shorter than one tenth of the wavelength in the dielectric. Beyond that, back drilling or a blind via construction is worth the process cost.