Control Board PCB Design Principles for Noise Immunity

A control board is a small computer surrounded by the things it controls, and its reliability depends on how well the layout keeps the noisy parts of that arrangement away from the sensitive ones. The rules are not complicated, and most of them are geometric: where a part is placed, how wide a trace is, where a capacitor sits and how a ground is laid out. Applied together they decide the noise immunity of the board, and whether the product behaves the same in a factory as it does on a bench.

Placement: Keep Related Parts Together and Noisy Parts Apart

Components that work together should be placed close together. The clock generator, the crystal and the clock input of the processor belong in one tight group, because the connection between them carries the most sensitive and the most aggressive signal on the board. Spreading them out adds a trace that couples into everything it passes and picks up whatever is around it.

Components that generate disturbance should be placed apart from the circuits that cannot tolerate it. Switching regulators, relay drivers, motor drive stages and any circuit with a large or rapidly changing current are the sources; memory, analog front ends and the logic of the processor are the victims. On a control board PCB where the two must coexist, the separation should be as large as the outline allows, and where the design allows it, the noisy stage is better placed on its own board with a filtered interface than sharing one ground with the sensitive circuit.

Decoupling Capacitors and Where They Belong

Every trace on a board is also an inductor. A supply line to a processor pin carries an inductance that is small in absolute terms and large compared with the impedance the device expects, and the consequence is a voltage spike on the supply rail every time the device switches. The only practical way to prevent that spike is to place a capacitor between the supply pin and ground, close enough that the loop formed by the capacitor, the via and the device is short.

The choice of dielectric matters as much as the value. A ceramic capacitor has a low equivalent series resistance and inductance, so it presents a low impedance at the frequencies that cause the problem, and its capacitance is stable over temperature and time. A tantalum capacitor of the same nominal value has a much higher impedance at high frequency, which is why it is a poor choice for the decoupling position even though its capacitance per unit volume is attractive.

The placement rules that follow from this are the ones worth remembering.

Place a bulk electrolytic capacitor of around 100 µF across the supply at the point where power enters the board, and use a larger value if the space allows, because this capacitor supplies the low-frequency component of the transient current.

Place a ceramic capacitor of about 0.01 µF beside every integrated circuit. Where the layout is too dense to give every device its own capacitor, one capacitor of a few microfarads can serve a group of about ten devices, but this is a compromise rather than an equivalent solution.

Give particular attention to memory devices and to any part whose current changes abruptly when it is disabled. These are the parts whose switching transients are largest relative to their steady current, and they are the ones where a missing capacitor produces an intermittent failure rather than a measurable degradation.

Keep the leads short, and for a high-frequency bypass capacitor, keep them as short as the footprint allows. A capacitor mounted with two long traces to the pin and the ground has an inductance that cancels its benefit at the frequency where it is needed. In surface mount designs, the capacitor is placed directly against the pin it serves, and the connection to the ground plane is made with a via beside the pad rather than with a trace running across the board.

<img src="https://www.gopcba.com/wp-content/uploads/2026/05/供电柜PCBA.png" alt="decoupling capacitor placed beside an IC supply pin” />

Grounding

A control board usually has more than one kind of ground: the system ground, the shield ground, the logic ground and the analog ground. How they are separated and where they are joined decides much of the board’s noise performance.

Logic ground and analog ground should be routed separately rather than shared. Each is connected to the ground of its own supply, and the analog ground should be wider than the digital one and should present the largest practical connection area at the point where analog signals leave the board. Where an analog input or output has to interface with the processor, an optocoupler or an isolated amplifier between the two domains removes the shared path entirely, which is usually cheaper than trying to control it.

The width of the ground is the second decision. A narrow ground has a resistance large enough that the potential along it changes with the current, so a signal that is measured against one end of the ground sees a different reference from the same signal measured at the other end. As a working rule, the main ground should be at least two to three millimetres wide, and the ground connection to a component pin about 1.5 mm. On a double-sided board, filling the unused copper on the bottom layer with ground and connecting it frequently to the top ground reduces the impedance of both.

On a simple two-layer logic board, it is worth connecting the ground as a closed ring rather than as a tree. A ring gives every point on the ground two paths back to the source, which halves the potential difference between any two points compared with a single path, and it is a cheap improvement on a board that has no ground plane.

Single Point or Multiple Point Grounding

The choice between the two classic grounding schemes follows from the frequency of the signals on the board.

Below about 1 MHz, the currents that cause trouble are the low-frequency circulating currents in loops, and the inductance of the wiring is comparatively unimportant. The right arrangement is single point grounding, where every circuit returns to one physical point, so that no loop is formed between two circuits that share a conductor.

Above about 10 MHz, the inductance of the ground path dominates. A long ground trace is no longer a low impedance connection, and the loop currents that made a single point connection attractive are no longer the dominant effect. The right arrangement is multiple point grounding, with a ground plane as the extreme case, so that every circuit has the shortest possible return path.

Between those frequencies, the design depends on which effect is larger in the particular board, and the deciding factor is usually whether a plane can be afforded. A four-layer board with a dedicated ground plane gives the high-frequency behaviour and, if the layout keeps the return paths sensible, it also keeps the low-frequency behaviour acceptable.

Power Routing and Signal Width

Power traces should be sized by the current they carry and routed in the same general direction as the data lines, so that the supply and the return form a low-impedance path along which the currents travel together rather than in opposition.

Data lines should be wide enough to keep their impedance low. A minimum of about 0.3 mm, which is 12 mil, is a reasonable floor for a control board, and a width of 0.46 mm to 0.5 mm, or 18 to 20 mil, is the better target where the space permits. Wider traces also make the board more tolerant of the small geometry variations of the fabrication process.

The last step of the layout, filling the unused copper with ground, helps for the same reason: it shortens the return path for signals that would otherwise have to travel to the nearest ground trace, and it reduces the etching load on the panel. The same rules that make the board quiet also make it easier to build, which is why the layout review and the fabrication review should use one set of drawings; our design and layout team and our PCB manufacturing group review the same data for that reason.

Control boards are usually built in modest volumes and used in equipment that cannot be serviced often, so the noise margin is worth designing in from the start. Where the product is used in an industrial environment, the same principles are applied with more margin, and our industrial PCBA experience covers boards of that class.

FAQ

How many decoupling capacitors does a control board need? At least one ceramic capacitor beside every integrated circuit, plus a bulk capacitor at the supply entry. Grouping ten devices onto one capacitor is a compromise for very dense boards.

Should logic and analog grounds be separated? They should be routed separately and joined at a defined point, or separated completely through an isolator where the analog signals interface with the processor.

When is a single point ground the right choice? Below about 1 MHz, where loop currents dominate. Above about 10 MHz, multiple point grounding or a ground plane is the better arrangement.