We should always bypass the power to ICs in our projects, and there are many related questions here.

Are there times when we should instead (or as well as) bypass the device that creates the noise in the first place? Switching say 10A causes significant spikes or drops. Examples would be power relays, motors or even a high power LED that's controlled via PWM. So we would have capacitors directly across the relay contacts or across the FET switching the LED. I don't see this done often, instead focus seems to be on the control circuitry.

One issue I foresee is the size of the capacitors. We typically stick 0.1uF to 1uF capacitors across most common ICs, perhaps going as low as 10nF for high speed ones. This has been well discussed. It seems to me that it might be more difficult to assess what size to use for power devices...

So to be clear, I'm not asking about bypassing ICs but rather the large metal thing they might be causing to switch on and off.

  • 2
    \$\begingroup\$ Across a power device it's usually called a "snubber". But note that the ICs are themselves noise sources! \$\endgroup\$
    – pjc50
    Oct 10, 2017 at 12:28
  • 2
    \$\begingroup\$ The trick with noise and spikes on supplies is to keep the current loop as short as possible. That way the spurious signals will be less likely to "wander around" on the PCB and disturb other circuits. That's why each IC should have a bypass cap but also any other device generating spikes, noise and spurs. The "crap" generated by high power devices like motors and relays has often a lot more power than the "crap" generated by ICs. So all the more reason to bypass locally for any "crap" generator. That it is not done everywhere doesn't mean it is not good practice to do it anyway. \$\endgroup\$ Oct 10, 2017 at 12:53

2 Answers 2


I'll use this circuit as an example:

enter image description here

First, consider the current loops:

  • FET ON: Supply - Motor - Q1 - Ground - Supply
  • FET OFF: Motor - Diode

This will draw a square wave current from the supply, with relatively high di/dt edges depending on switching speed.

di/dt combines with supply and ground inductance to create noise.

Placing a decoupling capacitor very close to this, between the MOSFET source (at ground) and the load's supply (cathode of FD1) provides a low inductance path for the high di/dt currents. These currents will naturally flow in the low inductance path, ie in the local decoupling cap, thus the supply will be less affected by HF noise. This is an example of treating the noise at the source, it is effective.

I mentioned current loops because you want the area of the loop which carries fast di/dt currents to be small, so it doesn't act as a loop antenna. Placing the cap close also helps with this.

Switching as fast as necessary, but no faster, is always a good idea.

As for the cap, in this example its ripple current is the full motor current, so it should be low-ESR. One or several ceramics (MLCCs) are ideal for this. Since ESR goes down with increasing capacitance, don't skimp on capacitance.


Everything needs to be "bypassed", or, better, "de-coupled". Bypassing an IC works both ways. In modern processors some core rails are a big source of noise/spikes, so "bypassing" actually must employ a spectrum of caps in close proximity to processor pins, and the rails should be isolated (de-coupled) from system supplies with inductor-based filters. Some caps are even placed on chip substrate/carrier, and some caps are internally embedded into silicon.

And yes, every high-power component must have adequate capacity on power rails locally, and ground return paths should be carefully laid out in a system, to keep high-current loops away from other parts of the system.

The size of caps depends on properties of loads, how fast and how strong they consume power. For example, many Wi-Fi radios emit in short bursts while consuming several Watts in peak. To bypass these units, tens of thousand of uF are needed in close proximity to the emitter.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.