Handling Power and Ground in a Mixed-Signal PCB Layout

Mixed-signal boards are difficult for reasons that have little to do with the components. The resistors, converters and processors are all ordinary parts; what makes the design hard is that the digital section injects current into the same structures the analog section depends on. Placement decides how much of that current reaches the sensitive circuits, and the way power and ground are handled decides whether the rest of it is filtered out or turned into noise.

Placement Comes Before the Plane

The instinct when a board is described as mixed-signal is to reach for the ground plane and start cutting. The more effective first step is to arrange the components so that the analog circuits and the digital circuits occupy separate regions of the board, with the boundary defined by where the two domains meet rather than by a line drawn on the layout.

That arrangement has three effects at once. It keeps the noisy current loops short and confined to the digital region. It gives the analog circuits a quiet area in which their own return currents can spread without sharing copper with switching currents. And it places the boundary in one identifiable location, so the crossing signals can be dealt with as a small, finite list.

The boundary is where the real work happens. The converter, or whichever device changes domain, should sit across it, with the digital interface pins on one side and the analog pins on the other. That single placement decision determines how much of the filter and how many of the crossing rules will be needed.

Treating the Supplies

A mixed-signal design usually needs more than one supply, and how they are derived matters as much as how clean they are.

The best arrangement is a separate regulator for the analog rail, referenced to the same ground as the digital rail. Two regulators share the ground plane and both return their currents to it, but the switching noise of the digital load stays in the digital regulator loop rather than passing through the analog one.

The cheaper arrangement is a filter between the rails: a ferrite bead in series with the analog supply, followed by a bulk capacitor and the local decoupling. This works, and it is used on a great many boards, but the bead has to be chosen rather than picked. Its impedance at the frequency of interest is what blocks the noise, its DC resistance adds a voltage drop that has to be inside the analog device tolerance, and its current rating must exceed the analog load with margin, because a bead that saturates stops behaving as a bead.

There is a subtlety that is often missed. A bead with a large capacitance behind it forms a resonant circuit, and the resonance can amplify the noise it was meant to remove. Adding a small damping resistance, or selecting a bead with a lossy characteristic rather than a high-Q one, keeps the filter from becoming an amplifier. The position of the filter matters for the same reason: it belongs at the boundary, so that the noise is removed before it enters the analog region rather than after it has spread.

The Return Path Is the Real Constraint

The reason a ground plane is treated so carefully is that every signal current has a return current, and at high frequency the return current does not take the shortest path in distance. It takes the path of least impedance, which is the copper directly beneath the trace.

This is the property that makes a continuous plane valuable and that makes a plane cut dangerous. A trace that crosses a gap in the plane it references has no direct return path underneath it, so its return current has to detour around the end of the gap. The detour creates a large loop, the loop radiates, and the impedance discontinuity at the crossing distorts the edge. On a mixed-signal board, the traces most likely to cross a boundary are exactly the ones that carry the fastest edges.

Where a domain boundary must exist, the arrangement that works is to route the crossing signals directly over the boundary, so that the return path stays close to the trace and passes from one region to the other at the point where the trace does. Series resistors on the crossing lines slow the edges and reduce the current the crossing carries, which reduces the disturbance at the boundary. Filters for the crossing lines are placed on the noisy side of the boundary, so the filtered current does not enter the quiet region.

<img src="https://www.gopcba.com/wp-content/uploads/2024/09/QA_Inspection.webp" alt="ferrite bead filter at a domain boundary” />

Decoupling and Where the Current Goes

Decoupling capacitors are usually described as a store of charge for the device. They are better understood as part of the return path: the loop formed by the capacitor, its vias and the device is the path along which the switching current returns, and the impedance of that loop is what determines how much voltage noise appears on the rail.

Two consequences follow. Every capacitor needs its own short connection to the plane pair, because a capacitor placed at the end of a long trace has a loop dominated by the trace inductance. And the plane pair has to be continuous under the capacitor and under the device, because a break in either one lengthens the loop.

The same logic applies to the analog side with a different emphasis. Analog devices draw small, steady currents, so their decoupling is less about transient supply and more about keeping the reference quiet. The capacitor beside a converter reference pin is protecting a voltage that must not move, and the ground it connects to must be the same ground the analog input signal returns to.

Getting that consistency right across a dense board with several supplies is the part of the design that most often gets reviewed late, and it is the part worth reviewing early. Our design and layout team treats the power and ground structure as a review item on its own, before the routing is closed. Where the design has to be built to a regulatory limit, the same structure determines the EMC result, and our industrial PCBA experience covers boards of that class.

What to Check Before Releasing

The list is short and it is mostly geometric.

Are the analog and digital regions separated by placement, and is the boundary in one place? Does every signal that crosses the boundary have a return path that crosses with it? Is there any trace running over a gap in its reference plane, or any plane cut that no signal needs? Is the filter on the crossing lines on the noisy side? Does every decoupling capacitor reach the plane pair with a short, wide connection?

A negative answer to any of those questions is a change that costs almost nothing before the data is released and a redesign after the first EMC test. The fabrication side of the same decisions, including the layer arrangement and the plane shapes, is where our PCB manufacturing review begins, and it is easier to make those changes while the stack is still being defined.

FAQ

Should the analog and digital grounds be separated? Usually not. Keep one ground plane and separate the circuits by placement. Separate grounds are for specific cases such as an isolated interface, where a single deliberate connection point is defined.

Is a ferrite bead enough to clean an analog supply? Often, if it is chosen for the frequency of interest, has adequate current rating, and is damped so the filter does not resonate.

Why does a plane cut cause problems at high frequency? Because the return current cannot follow the trace beneath it, so it detours around the gap and forms a larger loop that radiates and changes the trace impedance.