When dealing with capacitances <= 1uF ceramic is your choice. When you need more than 10uF it is economically justified to use electrolytic cap shunted with small ceramic one. But in range 1-10uF it is questionable. What do you suggest?


Your question is naive in that it doesn't consider what type of capacitor best suits the target application. Consider this: -

enter image description here

So, a 1uF ceramic may not be your best choice if wanting an accurate timing circuit (for instance) or some audio application where the microphonic problems associated with a ceramic capacitor cannot be tolerated. Also, these days, it's quite normal to find ceramics up at 100 uF and I'm sure in a few years time 1000 uF will be possible.

What do you suggest?

Choose a capacitor type that suits the application. If one doesn't exist look at alternatives and if those alternatives look problematic, find a different approach.

Pretty picture taken from here.


It all comes down to the frequency of interest. Here is a typical frequency response: frequency response You can see that the electrolytic capacitors have the worst performance. Generally they are used for decoupling at low frequencies (kHz range), i.e. They provide power until the supply can react.

Usually, I try to avoid them, unless you need 100uF or more. As to what is the best combination, that unfortunately depends on your requirements. You'll need to have an impedance spec. You can then select the appropriate capacitor combination. Something like this: impedance spec

Proper power integrity is a broad subject. As a general rule try to avoid using different capacitor values, as that can lead to resonance. See this app note. Pay particular attention to the plot on page 3.

  • \$\begingroup\$ The anti-resonance effects can be easily controlled by reducing Q of ceramic caps from Murata, where a small series ESR eliminates the spike in the spectrum, \$\endgroup\$ Nov 4 '16 at 20:59
  • \$\begingroup\$ @TonyStewart.EEsince'75 That may work for low frequencies, but it becomes increasingly more difficult as you go up. Firstly, decreasing the Q worsens your decoupling. The whole point of decoupling is to minimize the impedance, which you are not doing by decreasing the Q. Secondly, the traces, and the power planes introduce parasitics and resonance. Suppressing those, although not too hard, is not a trivial task. Finally, adding series resistance increases the trace length to the plane a lot. If you use a via in pad you get double the bandwidth than a 0.3mm via 0.1mm(edge) from the pad. \$\endgroup\$
    – user110971
    Nov 4 '16 at 21:15
  • \$\begingroup\$ @TonyStewart.EEsince'75 Also, you need to have access to the tools e.g. ANSYS. Furthermore, we are not talking about very high frequencies here. Around 10s of MHz. As I said in my answer, power integrity is a broad subject. \$\endgroup\$
    – user110971
    Nov 4 '16 at 21:18
  • \$\begingroup\$ Murata and TDK have free tools murata.com/~/media/webrenewal/tool/download/simsurfing/download/… \$\endgroup\$ Nov 4 '16 at 21:31
  • \$\begingroup\$ If you do not understand what I say is true, read electronics.stackexchange.com/questions/264927/… \$\endgroup\$ Nov 4 '16 at 21:40

The OP asks about economical aspects of decoupling caps, with focus on 1uF-10uF range. Electrical aspects aside, inspecting one common supplier of components, Digi-Key, shows that a 10uF 10% 6.3V aluminum cap costs 10c in qty 1000, while a 10uF 6.3V ceramic (size 0805) is only 1.5c/1000ea.

However, there are many aspects of being economical. Economically, considering the smaller size and better reliability/longevity/mecahicals of ceramic caps, the ceramic cap seems to be clear winner in the 10uF category. There is one exception though, when a on-board LDO needs some finite ESR for its stability. Then the low-ESR ceramic is very bad, and it is fairly costly to fit any tantalum or aluminum in the place. Or add an explicit ESR of 1-2 Ohms.

So, OP needs to shift the question into into something like 100uF area, or maybe even higher.


The choice of ripple reduction caps can be critically defined by the impedance spectrum times(x) the current pulse spectrum='voltage ripple spectrum. This requires you define I(f) and V(f) on a spectrum analyzer, then the choice of caps becomes easier with scattering parameter files from OEM... or by trial and error.

The V(f) spectrum results in the Vpp(t) ripple signal.

This applies to 1A or 100A or any value when high ripple current and low ripple voltage are needed.

Careful attention to series (resonant) and parallel(anti-resonant) interactions with parallel caps . or SRF and PRF, result in an optimum solution.

Normally it is easier to work with admittance spectrum Y(f), as this adds rather than Z(f).

Then where necessary measure ripple current in RMS and compare to rating. Neglecting these fundamentals in high power SMPS, leads to early failures in production or in the field.

This is supplementary to the other fine answers.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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