Return Path Design Across Split Planes

Every signal that travels on a trace returns on a reference. At low frequency the current takes the path of least resistance, which may be a long way around, but at high frequency the return current follows directly beneath the trace, because that is the path of least inductance. That behaviour is what makes a continuous reference plane so valuable, and what makes a split plane so damaging to a fast signal that crosses it.

This article explains what a return current needs, what a split does to it, how much a slot costs, and how the layout avoids the problem.

What The Return Current Needs

The return current forms a loop with the signal, and the inductance of that loop is what sets the impedance of the trace. The return path is not a single line but a distribution across the plane, concentrated under the trace and spreading with distance. Anything that interrupts that distribution forces the current to take a detour, and the detour adds inductance and changes the impedance of the trace for its length.

The plane also provides the second half of a transmission line. A trace over a plane is a microstrip, and its impedance depends on the distance to that plane. If the plane is removed or replaced by a different plane at a different potential, the geometry changes and the impedance steps at that point. Both effects are consequences of the same physical arrangement, and both are corrected in the same way.

What A Split Does

A split is a gap in a plane, made to separate a noisy ground from a quiet one, to separate analog from digital, or to divide a supply into voltage domains. The split itself is not the problem; the problem is a signal routed across it. The signal trace passes over the slot, the return current cannot follow, and it is forced to go around the end of the slot or to find another plane.

The detour has two consequences. The added inductance makes the trace look like it has a series inductor in it, which produces a reflection and a delay. And the current that finds a path elsewhere produces a magnetic loop that radiates, which is why a slot under a fast trace is a common source of an emissions failure that appears at one frequency.

How Much A Slot Costs

The cost depends on the length of the detour and on the frequency. A trace that crosses a slot a few millimetres long inside a closely spaced pair of planes suffers little, because the capacitance between the planes provides a path for the return current. A trace that crosses a slot tens of millimetres long, or that crosses between planes on layers far apart, has no such path and the effect is large.

The rule of thumb is that a discontinuity can be tolerated while its electrical length is a small fraction of the rise time of the signal. A slot that is shorter than about one tenth of the wavelength of the highest significant frequency behaves acceptably; a longer one does not. That is why a low speed board can have split planes everywhere without a problem, and a high speed board cannot have a single crossing.

<img src="https://www.gopcba.com/wp-content/uploads/2026/06/Thermal-Management-1.jpg" alt="Stitching capacitors placed along a split” />

Working With A Split

Where a split is genuinely needed, the crossing is arranged deliberately. A stitching capacitor placed close to the crossing provides a path for the return current between the two planes, and its value and its placement matter: it has to be adjacent to the trace, within a few millimetres, or its own inductance defeats the purpose. Several capacitors are usually placed along a bus that crosses the same slot.

Routing around the slot is the alternative, and it is usually better for a fast trace if the detour is acceptable. The trace is routed to a point where the two planes are joined by a bridge of copper, which is a section of the plane that carries the current across the split without a discontinuity. The rules for arranging the planes and the bridges are described under power plane splitting rules and the general arrangement of the return structure under ground routing and power trace planning.

Design Rules That Avoid The Problem

The most effective rule is to give every high speed signal a continuous reference plane for its whole length. That means choosing the layer assignment so that a signal layer is adjacent to a plane that is not split under it, and keeping the splits away from the region where the fast traces run. A split that has to exist is drawn in the area of the board that carries only slow signals.

The second rule is to keep the reference the same along the route. A trace that changes layers must keep the same reference plane, or the return current has to find a path between planes, which it does only through stitching vias. Vias that join the two planes need to be placed near the signal via, and the general measures for reducing emissions by stackup and layout are described under EMI reduction through stackup and layout.

Verification

Verification at the design stage is a check that every fast net crosses only one reference and that no split lies under it. That check is a geometry query in the layout tool, and it is far more reliable than a manual review of the planes.

Verification on the finished board is a measurement. A time domain reflectometer shows the impedance discontinuity at the crossing, and a near field scan shows the radiation from the slot. Both are used in development, and both are evidence that the design rule was worth following. Where a product has failed an emissions test, the scan is the fastest way to find which slot is responsible.

Analog, Digital And The Common Reference

The usual reason for a split is the belief that a digital ground will contaminate an analog one. In most designs the contamination comes from the placement and the routing rather than from the plane, and a single reference with the converters and their input networks grouped in one area works better than two planes joined at a single point. Where two planes are used, the join has to be a low impedance connection at a defined place, and every signal that has to cross between the domains has to cross at that place with its return current.

The difficulty with two planes is that the return current does not respect the intention of the designer. A signal routed from the digital area into the analog area will return through the join, and if the join is a narrow bridge, the return current develops a voltage across it which appears as noise in the measurement. That is why the arrangement of the join and the routing of the crossing signals are designed together, and why a split that is added late in the layout often makes the measurement worse rather than better.

FAQ

Should grounds be split at all? Usually not on a mixed signal board. A single ground with a careful placement of the converters and the sensitive circuits is normally better than two grounds joined at one point.

Can a signal cross a split if it is slow? It can, and many boards do. The relevant test is whether the discontinuity is a small fraction of the rise time, which for a slow signal it usually is.

How close does a stitching capacitor need to be? Within a few millimetres of the crossing, with a short and wide connection to each plane. A capacitor further away has enough inductance in its own connections to be ineffective at the frequencies that matter.