Heatsinks can be a particular cause of EMC difficulties. The concern is the effect of stray capacitance and RF radiation from heatsinks either on digital devices carrying RF voltages, such as microprocessors and ASICs, or on switching power components such as MOSFETs, IGBTs or triacs.

There are often direct contradictions between what is good for thermal design and what is good for EMC design. This is at least in part because thermal design attempts to remove heat from hot parts and exhaust it to the outside world; while EMC design attempts to keep emissions from noisy parts inside the enclosure and prevent them from reaching the outside. Usually, hot parts are also noisy parts (VLSI ICs and power switching devices). And, efficient thermal paths are also efficient at noise conduction and radiation.

Good practice therefore treats the thermal and EMC design constraints together, and looks to find the optimum compromise between the two. Metal structures in the mechanical design, and their connections to circuit 0V or chassis ground, are usually a key part in this compromise.

HF radiation from a digital device

To analyse the problem, we can create simplified equivalent circuit of a PCB with a 0V plane, a processor, a heatsink on the processor, and an enclosure around the whole assembly. The enclosure may be screened with the heatsink within it, or it may be unscreened and the heatsink couples directly with the exterior. The processor is the source of the RF noise: voltages V_N on the silicon at clock frequencies or due to data operations appear with respect to the 0V plane and are capacitively coupled directly to the metal of the heatsink. The heatsink must have good thermal contact to the chip, and a by-product of this is a high coupling capacitance C1. In its turn, the heatsink is a large chunk of metal, and it has a high capacitance C2 to its surroundings, either the enclosure or the external environment.

This energizes the entire enclosure (if it's screened) with circulating currents at the noise frequency and its harmonics. Any weakness in the enclosure will allow these frequencies to radiate, and with an unscreened enclosure, the heatsink will radiate directly. To control this, the most effective measure is to prevent the heatsink from carrying high levels of digital noise. This means it must be referenced to the 0V plane on the PCB, and cannot be allowed to float. Capacitance C1 and noise source V_N cannot be removed, but if the heatsink is connected to 0V then the circulating noise currents remain in its locality and the voltage on the heatsink is minimized. The enclosure itself is not stressed, since C2 is not fed from a noise source.

To avoid high frequency resonances with the capacitance of the heatsink, the inductance of the 0V connection needs to be well below 1nH to be effective, and this means that multiple contacts around the outside of the heatsink must be provided - a single one, such as a mounting screw, is almost never enough. The most effective designs use a continuous conductive gasket around the periphery of the heatsink (which must be conductively finished) to a ground plane contact strip on the surface of the PCB.

LF coupling from a power switching device

A similar circuit can be devised for conducted emissions from switching converters (see the section on power switching circuits for further discussion). Here, the source is the dV/dt on the drain terminal of the switching device (MOSFET or IGBT, typically). For non-isolated devices, the drain is usually the mounting tab. The tab is closely coupled to the device heatsink; it may actually be directly connected to the heatsink, or isolated from it by a thermally conductive but electrically insulating washer. In the latter case, the capacitance C1 is of the order of 20-40pF for typical small devices (TO220 or TO247). ISO-TAB devices don't eliminate this capacitance, although they will reduce it somewhat. The heatsink is again coupled to its environment, as described above for a digital device, by C2, which depends on the heatsink's surface area.

In the mains (or DC) conducted emissions test the measurement reference is the test ground plane, which may be directly connected to the chassis of the switching converter and therefore C2 couples directly back to this. Depending on whether and to where the heatsink is connected, either C1 or C2 may be shorted out. To minimise the effect of coupling from the switching device through the heatsink to the environment, the best connection is to return the heatsink directly to the local switching device 0V.

What this equivalent circuit shows is that the one place not to connect the heatsink to, is the enclosure. If this is done, then the capacitance C2 has been bypassed and all the noise coupled through C1 flows in the enclosure. This is bound to worsen emissions. The only source-level cure in this case is to reduce C1, or eliminate it by incorporating a screen to 0V between the device and its heatsink. Far better to design from the start with the aim of keeping heatsink coupling to a minimum. Don't use the chassis as a heatsink with a switching device directly mounted to it; or if you must, and after all for good thermal reasons a lot of designs do, then expect to have to take substantial extra measures in filtering.