Ground Plane Splitting: When It Helps and When It Hurts

Separating the analogue ground from the digital ground is one of the most widely repeated pieces of advice in hardware design, and it is also one of the most widely misapplied. The reasoning behind it is sound: digital circuits switch large currents quickly, the ground is not an ideal conductor, and the resulting potential differences will appear in any sensitive circuit that shares the same reference.

The mistake is in the remedy. Cutting the plane into two does not remove the current, it constrains where the current is allowed to return, and a constrained return path is exactly what the rest of the design rules are trying to avoid.

Why the Two Grounds Are Separated in the First Place

Copper has resistance and inductance, so a current flowing through it produces a voltage. For a steady current the resistance dominates; for the fast edges of a digital circuit the inductance dominates. When a returning current passes through copper that a sensitive amplifier also uses as its reference, the amplifier sees the movement as part of its own input signal.

If the two circuits are arranged so that the noisy current and the sensitive reference do not share a path, the interference is largely removed without any change to the netlist. That is the objective. Whether it is achieved by cutting the plane or by arranging the components is a separate question, and on a mixed signal layout the answer is usually the arrangement.

Two Principles Worth Keeping

The two electromagnetic compatibility principles that apply are simple to state. The area enclosed by a current loop should be as small as possible, because the radiated field scales with loop area and with the rate of change of current. And the system should have a single reference plane, because two separate references behave like a dipole antenna, radiating in proportion to their effective length and the current flowing between them.

Both principles point in the same direction: a continuous reference beneath every signal, so that the return current path is as short as the trace above it. Against that background, a plane split is a deliberate exception, and it has to be justified rather than assumed.

What the Split Actually Causes

Consider a signal that crosses the gap between two planes. The return current cannot flow directly beneath the trace, because there is no copper there. It has to detour, and if the two planes are joined at a single point, the detour is the whole distance to that connection. The loop area becomes large, the inductance in the return path rises, and the structure starts to behave like the loop antenna the first principle warns about.

If the two planes are instead connected at several points, the isolation the split was meant to provide disappears, and the designer has achieved the disadvantages of a split with none of the benefit. If they are joined through a long wire, the wire itself becomes the dipole addressed by the second principle.

The most common practical failure follows directly, and it is worth checking in the design review rather than during a measurement campaign. The designer cuts the plane correctly, then routes a signal across the gap because the layout required it, and the emission or the noise coupling appears precisely at that trace. The fault is not that the ground was split; the fault is that a signal was routed without a return path.

When a Split Can Still Be Appropriate

There are cases where separating the planes is reasonable, and they share a common feature: no signal crosses the boundary. Where a board contains a genuinely isolated section such as a mains referenced power stage, an isolated interface with its own reference, or a sensor section that communicates optically or through a transformer, the two references are not connected by copper and no return current is expected to cross.

When the split is used on a mixed signal board, the connection between the two grounds has to be defined. A single point connection bridge beneath the converter, as wide as the device itself, gives the analogue and digital grounds a low impedance path at the one location where they must meet, and it gives the return currents a defined route rather than a network of unintended ones. Placing it correctly is a copper decision as much as a circuit one. The rule that goes with it is absolute: nothing may cross the gap anywhere else on the board.

The Alternative That Solves the Same Problem

The approach that avoids the dilemma entirely is to keep one ground plane and partition the board by placement. The analogue circuits occupy one region, the digital circuits another, and the routing keeps each set of signals inside its own region. The return currents for the digital signals then stay under the digital section because that is where their traces are, and the analogue reference is not asked to carry them.

This works because the partitioning is a property of the layout rather than of the copper. It satisfies both compatibility principles at once, provides every signal with a direct return path, and does not create a boundary that a later revision can accidentally cross. Its weakness is that it depends on discipline: a single trace routed through the analogue region for convenience reintroduces the problem that the split was supposed to prevent, and it does so without the visible warning that a cut plane would give.

For converters, the usual guidance is to connect the analogue and digital ground pins to the same low impedance reference through the shortest possible connections, because the internal structures of the device do not provide that connection by themselves. With a single plane this is achieved by placing the device so that both pins reach the same copper. With split planes it requires a bridge exactly beneath the package, which is a more fragile arrangement.

Deciding Which to Use

Ask whether any signal has to cross between the two domains. If the answer is yes, a split is the wrong choice, because the crossing will defeat it. Ask how many converters are on the board: a single one can be served by a bridge, while ten converters cannot each have their own without producing multiple connections that remove the isolation entirely.

Where the reference must be genuinely separate, such as an isolated supply or an optically coupled interface, the split is appropriate because the boundary is a real one and the currents are already separated by the isolation device. Where the boundary exists only because two kinds of circuit share a board, the placement is the tool to use.

FAQ

Is it ever wrong to use a single ground plane? Only when the two references are genuinely isolated and no current is expected to flow between them.

Can a signal cross the gap if it is a differential pair? The return current of a differential pair flows in its partner conductor, so the pair does not need the plane for its return, but the common mode component does. The practice remains risky.

How wide should the connecting bridge be? As wide as the device whose ground pins it joins, and placed directly beneath it, so the connection is short and low in inductance.

What is the single most common mistake? Splitting the plane correctly and then routing a signal across the boundary, which turns a well intentioned measure into a radiating loop.

Summary

Ground plane splitting addresses a real problem, the sharing of a reference between noisy and sensitive circuits, but it does so by restricting the return path, which creates a different and often worse problem. Partition the board by placement, keep a single reference, and reserve the split for boundaries that are genuinely isolated. If a split is used, connect the two planes at one defined location and let nothing cross it.