Reducing RF Effects in PCB Interconnect Design

Once the signals on a board reach the gigahertz range, the interconnect stops behaving like a connection and starts behaving like a circuit. Corners reflect, surfaces resist more than their cross section suggests, finishes add loss where they were meant only to protect, and the material a trace is printed on becomes part of the electrical design. Reducing the RF effect in a design is largely a matter of removing the small deviations from a uniform transmission line that accumulate into a degraded channel.

Where the Effects Originate

Four mechanisms account for most of the loss and distortion.

Discontinuities — any change in geometry along the path — produce reflections, and each reflection takes energy out of the forward wave. The skin effect — the crowding of current near the surface of a conductor, which becomes more pronounced as frequency rises — raises the effective resistance of a trace above its direct current value. Adjacent traces exchange energy through the fields around them, which appears as crosstalk. And the edges of planes and the ends of unmatched lines radiate, which is both a loss and an emissions problem.

Every practical measure below addresses one of these four.

From the Device to the Board

The transition from a package to the board is the first place where the integrity of a channel can be lost, and it is often the most difficult to correct afterwards.

Interconnect density is the constraint. As device input and output counts rise, the pitch available for escaping them falls, and the board materials and processes become the limiting factor rather than the silicon. The practical measures are to escape on fine geometry where the package demands it, and to keep the transition as short as possible so that the discontinuity it represents occupies a small fraction of the channel.

Package style matters for a related reason. A leaded component introduces inductance in series with the signal, and at high frequency that inductance is a significant impedance. Surface mount components remove the lead and much of that inductance, and the same reasoning applies to board-level connectors: the shorter and more controlled the transition, the less the channel suffers.

Routing Inside the Board

The routing measures that follow are the familiar ones, applied consistently.

Corners are formed at forty-five degrees or with curves rather than at right angles, because a sharp corner is a local change in geometry that reflects part of the signal and concentrates the field. Where a corner is unavoidable, a small chamfer removes most of the effect.

Ground layer continuity is the foundation of everything else. A trace whose return path is directly beneath it behaves approximately as a transmission line; a trace whose return path has to detour around a slot does not. Providing continuous reference beneath fast signals, and placing stitching vias to tie reference planes together where they would otherwise be interrupted, is the single most effective measure available.

Vias are treated as the discontinuities they are. Each one adds capacitance and inductance, so routes that do not need to change layers should not, and where a change is required the return current needs a path across the same transition. Placing a via for the return next to the signal via provides it.

Material choice supports the rest. A laminate with a tightly controlled dielectric constant and low loss keeps the impedance predictable and reduces the energy absorbed by the dielectric along the route. The control matters as much as the value: an unpredictable dielectric constant makes the impedance vary between boards, which is more damaging than a consistently different one.

Etching and Cross Section

At high frequency the shape of the conductor matters as much as its nominal width, and shape is a fabrication property.

Etching removes copper from the sides of a trace as well as from the exposed surface, which produces a trapezoidal cross section rather than a rectangular one. The loss of width at the top of the trace raises the resistance, and the effect is proportionally larger on narrow traces. The plating in the hole and on the surface, the sidewall angle and the consistency of the etch across the panel all determine whether the finished trace has the impedance the model predicted.

This is why a design intended for microwave frequencies has to state its requirements in terms of the finished geometry rather than only the drawn one: total width tolerance, the acceptable undercut, and the condition of the sidewalls. Those requirements are process specifications as much as design ones, and they belong in the fabrication data.

Surface Finish

The finish on the copper is part of the conductor at high frequency, and its properties matter more than its purpose suggests.

Finishes based on nickel and gold, or immersion finishes, present a comparatively smooth and uniform surface. A finish applied by dipping the panel in molten solder and levelling it with hot air leaves a less uniform surface, which affects both the effective conductivity and the consistency of the impedance along the trace. The flatness matters for assembly as well, since a rougher surface is less suitable for fine-pitch placement.

The choice between finishes is described more fully in the article on surface finish selection, and the trade is one of several in which an assembly benefit and an RF benefit point in the same direction.

Solder Mask and the Dielectric Environment

Solder mask over a microstrip changes the dielectric environment of the trace, and the mask is applied in a thickness that is not tightly controlled. The result is a small, variable change in impedance along the trace, which is a source of the very reflection the rest of the design works to avoid.

Two practices reduce the effect. The first is to keep the mask off the critical traces, where the process allows it, using it for the areas between conductors rather than over them. The second is to account for the mask in the design model when it must be present, so that the predicted impedance includes it rather than being corrected later by measurement.

Leaving the Board

The last transition is from the board to a cable or an external system, and it is where a well-controlled channel is most likely to be spoiled.

A microstrip that reaches a coaxial connector has to change from a planar transmission line to a cylindrical one, and the transition has to preserve the impedance across the change. That requires the ground structure to be arranged deliberately around the transition rather than left to the general plane, and it requires the connection of the shield to be made at the point where the signal leaves the board so that the return current has somewhere to go that is not around the edge of the board.

Where the transition is well made, the reflection it produces is small and the radiation from the edge is contained. Where it is not, the result appears as a return loss that no amount of trace tuning on the board will fix.

Pulling the Measures Together

Most of the measures above belong to the routing and to the fabrication data rather than to components: continuous references, controlled geometry, minimal vias, a finish that suits the frequency, and specifications that describe the finished conductor rather than the ideal one. The routing behaviour of fast nets is covered further in the article on high-frequency routing, and the way impedance is specified, verified and tested through the fabrication process is set out in our guide to impedance control.

FAQ

Why is a 45 degree corner preferred over a right angle? Because the change in geometry at a sharp corner reflects part of the signal and concentrates the field. A gentler corner reduces both effects.

Does the surface finish matter electrically? Yes, at high frequency the finish is part of the conductor. A smoother, more uniform finish presents a more consistent surface than one applied by levelling molten solder.

Should solder mask be left off critical traces? Where the process allows it. Mask thickness varies, and over a microstrip that variability shows up as a change in impedance along the trace.