12
\$\begingroup\$

Often 100n to 1µF decoupling caps are placed across IC supply lines for digital logic.

For analog circuitry, are decoupling caps necessary when the environment is also shared by microcontrollers and digital logic?

I have never placed them and haven't had any troubles, but I haven't made production stuff yet, so wouldn't have much experience.

\$\endgroup\$

5 Answers 5

12
\$\begingroup\$

In a mixed signal environment for production that has to pass FCC, yes absolutely.

More specifically, what you need to do is look at your current usage, the frequencies that will be present, and determine what your overall power supply capacitance needs to be to minimize those frequencies on the supplies. Otherwise you'll get ringing on the supply planes that can be a huge EMI issue.

You'll get some capacitance from the PCB stack up, assuming you have power and ground planes. You'll then usually come up with a needed capacitance and size of capacitors to achieve your goal.

For instance you can come up with something like:

  • 30 0.1uF 0603 max
  • 30 10nF 0402 to avoid lead inductance
  • 5 10uF tantalum

Then sprinkle these around in a logical manner. 1 0.1uF and 1 10nF per power pin. One tant per major IC or near a section of smaller current/analog ICs.

With mixed signal design you always have to remember that just because a signal is low frequency analog you still have to treat it as an EMI threat. There will be transients from the rest of your system on that signal no matter how amazing your isolation is.

Not only talking about high speed here either. A system with a 25Mhz clock and easily have these issues and fail FCC pretty miserably (trust me :0)

\$\endgroup\$
5
  • \$\begingroup\$ Two things I would add. 1) If you can't fit an 0402 on your PCB, you can skip this level of bypassing. The rule of thumb is generally "use the largest capacitance that you can afford for a given package size" unless you're paranoid about inductance. \$\endgroup\$
    – ajs410
    Oct 18, 2010 at 17:53
  • \$\begingroup\$ doh. 2) be careful with tantalum caps when used as a power input cap. they don't like surge currents. \$\endgroup\$
    – ajs410
    Oct 18, 2010 at 17:54
  • \$\begingroup\$ your first point is not applicable to bypassing to reduce EMI / noise ripple. Rather than sticking the largest cap you can on the power pin you need to look at the frequencies your expecting ripple at and select capacitors that have their impedance minimum at those frequencies. This depends on dialectic, package, construction and capacitance. If your dealing with a 25Mhz system clock your danger frequencies will be 25,75,125,175,225. The 0.1uf is generally chosen to cover the low end and 10nF is generally effective from 80-300mhz or so. \$\endgroup\$
    – Mark
    Oct 19, 2010 at 17:56
  • \$\begingroup\$ The way I approach it is to identify my decoupling needs, both in size and frequency response. So say I know i need coverage from 0 to 250mhz, I choose whatever grouping of caps results in the impedance to ground always being <0.1ohm. Using standard MLCC with X7R dialectic, 0.1uF and 10nF cover me from ~4Mhz to 250Mhz well which is where most concern is. Down into the audio range, <20khz the ESR of ceramics increases quiet a bit so using an electrolytic is beneficial, be it tantalum or aluminium. Thus these are used for low frequency bulk bypass. \$\endgroup\$
    – Mark
    Oct 19, 2010 at 18:10
  • \$\begingroup\$ Then you figure out how much capacitance you need based on the current draw on the various supplies. If you know the current draw you know the represented resistance and can calculate the needed capacitance for the desired ripple reduction. \$\endgroup\$
    – Mark
    Oct 19, 2010 at 18:11
6
\$\begingroup\$

This really depends on the ICs you are using. Generally the higher bandwidth an analogue device has the more important power supply decoupling becomes. Most of the time the data sheet for different devices will give you an indication of what is required. Any high speed amplifier or comparators may be susceptible to oscillation if not correctly bypassed.

\$\endgroup\$
4
\$\begingroup\$

Analog ICs like comparators and op-amps most certainly DO need decoupling, especially if they're used as hysteretic switches. You can see very strange behaviour (bouncing between states being quite common) if there isn't good decoupling and there's some HF noise on the supply.

Coming from an analog switching power background - if I suspect an op-amp or comparator isn't doing what I think it should be, the first two things I always check are: (1) is there a decoupling capacitor, and (2) if there is, is it in a good electrical position in the layout?

\$\endgroup\$
2
\$\begingroup\$

Another point to consider with decoupling capacitors on op-amps is they need to go from rail to ground, not rail to rail. For example an op amp with +/-5 V rails needs to capacitors one from each rail to ground. This will ensure the op amp has properly decoupled power supplies.

You also need to have them on signal paths also, for example a small capacitor across the feedback resistor will help your op amp circuit transition from a simulator to a real PCB with out noise and oscillations.

\$\endgroup\$
2
  • \$\begingroup\$ I've seen circuits with rail to ground caps in addition to one across both rails. Is there any advantage to putting one across the rails? \$\endgroup\$
    – Thomas O
    Oct 19, 2010 at 11:59
  • \$\begingroup\$ No advantage putting them across both rails. Best to keep them rail to ground, as it provides a noise path from the rails back to the ground plane, as that is your 0 V potential source, and everything needs to be decoupled back to this! \$\endgroup\$ Oct 19, 2010 at 12:36
1
\$\begingroup\$

Usually I'd have gone with no. The reason decoupling capacitors are critical for digital circuitry is that they can use high currents when switching states; the capacitor will then reduce the size of that current loop, and even out the draw from the source. For analog circuitry this may be less of an issue, although some of the time the reason it's an issue is because analog circuitry will produce poor results because of the supply noise. Sensitive analog circuitry is therefore separated to its own supply, possibly with capacitors and inductors to smooth things out.

I'm also fairly inexperienced, however, so expect better answers soon.

Edit: Indeed there were better answers. Do decouple opamps, and particularly comparators. Glad I learned something!

\$\endgroup\$
2
  • 2
    \$\begingroup\$ Thanks & +1 for your answer. A device like an LM339 may change quickly though and its current consumption may vary. Would decoupling be necessary here in some situations? \$\endgroup\$
    – Thomas O
    Oct 17, 2010 at 0:17
  • 1
    \$\begingroup\$ Ah, a comparator. I'd be leaning towards includuing capacitors then; they should not hurt. \$\endgroup\$ Oct 17, 2010 at 0:32

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.