Rule 3: No reference plane gaps
The immediate vicinity of devices is littered with vias. Power nets break out into local decoupling and then drop to power planes, often with multiple vias to minimize inductance and increase current carrying capacity, while control buses drop to inner layers. All of these breakouts end up thoroughly cramped near the device.
Each of these vias incurs a keepout region, larger than the via diameter itself, on inner ground layers to provide manufacturing clearance. These keepout regions easily create discontinuities on the return current path. Compounding the situation, several vias close together can create a ground plane trench, hidden from view on the top layer CAD view. Figure 2 shows an example of a situation where ground plane clearance from two power plane vias create overlapping keepout regions and create a discontinuity in the return path. Return current is forced to divert around the ground plane keepout regions, creating the now-familiar problem of a radiating inductive path.
Figure 2: Keepout regions in the ground layer surrounding vias may overlap, forcing return current to far from the signal path. Even without overlap, the keepout region may form a mouse bite impedance discontinuity in the ground layer.
Even “friendly” ground vias bring an associated metal pad with minimum dimensions dictated by the board fabrication process. When the via gets too close to the signal line, it can create an encroachment that resembles a mouse bite chomped out of the topside ground clearance. Figure 2 includes an illustration of a mouse bite formation.
Because the keepout regions are created automatically by CAD software and vias are used so frequently on system boards, the initial pass of a layout almost always includes a few return path disruptions. During a layout review, trace every high-speed line, inspecting the associated return current layer for disruptions. A good rule is to move any vias that create a ground layer disruption within any region closer than the top layer ground clearance.
Rule 4: Keep differential lines differential
The return current path is so critical to signal line performance that it should be considered part of the signal path. At the same time, differential pairs are often not tightly coupled, and return currents might flow through an adjacent plane. The two return currents must be routed through electrically equivalent paths.
Even when two lines in a differential pair are not tightly coupled, proximity and shared design constraints keep the return currents on the same plane; really keeping spurs low demands even better matching. Any intentional structures, such as ground plane cutouts underneath a differential component, should be symmetrical. Similarly, length matching may introduce squiggles in the signal line. The return current does not follow the squiggles. Any length matching in a differential line should be reflected in the other differential line.
Rule 5: No clocks or control lines near RF signal lines
Clock and control lines are sometimes treated like benign neighbors, because they operate at slow speed or even near DC. However, their switching characteristic is nearly a square wave, generating distinct tones at odd harmonic frequencies. The radiated energy of a square wave has less to do with its fundamental frequency than the sharpness of its edges. In digital system design, knee frequency estimates the highest frequency harmonic that must be considered, it is calculated through fknee = 0.5 / tr, where tr is rise time. Notice that rise time, not signal frequency, matters. But square waves with sharp corners also have strong high order odd harmonics that might fall at just the wrong frequency and couple onto an RF line, violating an aggressive transmit mask.
Clock and control lines should be separated from RF signal lines with an internal ground layer or topside ground pour. If it is not possible to isolate signals with ground, route lines so that they cross at a right angle. Because magnetic flux lines radiated from the clock or control line form radial cylindrical contours around the flow of current in the aggressor line, they will not induce a current in a receptor line. Slowing a rise time decreases the knee frequency and helps reduce the impact of an aggressor, but a clock or control line can also act as a receptor line. The receptor line can still act as a conduit for routing spurs into a device.
Rule 6: Isolate with ground between high-speed lines
Microstrip and stripline mostly couple to adjacent ground layers. Mostly. Some flux lines still emanate horizontally and terminate in adjacent traces. The tone on one high-speed line or differential pair ends up on the next trace. But a ground pour on the signal layer gives flux lines a lower impedance destination, keeping tones out of neighboring traces.
Clusters of traces carrying identical frequencies, as found routed out of a clock distribution or synthesizer device, may run adjacent, because aggressor tones are already present on the receptor line. But the lines in the grouping eventually spread out. When they do, include a ground pour between the dispersing lines and drop in a via where they begin to spread so that the induced return current can travel back along the nominal return current path. In Figure 3, the via at the end of the ground island allows the induced currents to move onto the reference plane. Space additional vias along the ground pour no more than every tenth of a wavelength so that the ground does not become a resonant structure.
Figure 3: A via in the topside ground where differential lines separate provides a path for return current to flow.