Electronics > Projects, Designs, and Technical Stuff
Choosing the best capacitor for switch mode power supply
(1/3) > >>
arildj78:
I'm designing a SMPS and for bulk capacitance I'm considering a bank of 8 to 10 22µF ceramic caps. This is to achieve low ESR which in turn may keep the ripple low.

The switching frequency is 1MHz, and attached is a plot of capacitor Z vs frequency from the cap datasheet. Now, is it good or bad that the part seems resonant at 1MHz?

Arild
tggzzz:
This is a very complex subject to get right.

For starters, make sure you understand the consequences of the C-vs-Vdc spec. Continue by considering failure modes.
MagicSmoker:
It's a good and bad thing, though mostly bad. The capacitor will look like a dead short* over a very narrow frequency centered around its series resonance frequency, and if the switching frequency coincides with such any reflected ripple will be all but eliminated. However, this may cause way more ripple current to flow through the capacitor than what it is rated for.

Ideally you want to use capacitors which have an SRF higher than the switching frequency, but at 1MHz that is going to be hard to do, so a combination of elkos and ceramics or film types is your next best bet. The elkos provide bulk filtering of low frequency ripple (from load changes and the inevitable rolloff of the converter bandwidth well below fsw) and damping for the high-frequency filtering provided by the smaller value/package size ceramic/film caps.

* - actually, the impedance will drop to the ESR, but as that is vanishingly low for an MLCC you can treat it as a dead short.
ejeffrey:
You also can't trust that resonance.  The real capacitor is approximately a series LCR.  L and r don't vary too much but C has a huge variation with temperature and especially voltage which will shift the resonance to higher frequency.
T3sl4co1l:

--- Quote from: MagicSmoker on September 04, 2019, 10:39:19 am ---It's a good and bad thing, though mostly bad. The capacitor will look like a dead short* over a very narrow frequency centered around its series resonance frequency, and if the switching frequency coincides with such any reflected ripple will be all but eliminated. However, this may cause way more ripple current to flow through the capacitor than what it is rated for.
--- End quote ---

Rest assured the AC current flowing in a switching converter is defined by the switching and the inductor, and not the capacitor. :)



--- Quote ---Ideally you want to use capacitors which have an SRF higher than the switching frequency, but at 1MHz that is going to be hard to do, so a combination of elkos and ceramics or film types is your next best bet. The elkos provide bulk filtering of low frequency ripple (from load changes and the inevitable rolloff of the converter bandwidth well below fsw) and damping for the high-frequency filtering provided by the smaller value/package size ceramic/film caps.

* - actually, the impedance will drop to the ESR, but as that is vanishingly low for an MLCC you can treat it as a dead short.

--- End quote ---

The general strategy is this: treat the filter as an RLC network.  Consider the Q factor of each stage, and introduce resistance as needed to avoid peaks in impedance or transmission.

OP: there's nothing wrong with ceramic caps, but using them for bulk tends to be expensive, and you should be aware of their pitfalls: namely, DC bias, tempco, aging, and sensitivity to cracking.  (For these reasons, you would typically choose an X7R dielectric, check the datasheet to make sure you have enough C at DC bias, and use 1210, maybe 1812, or smaller, chips.  Or THT or leadframe parts, but those are even more expensive, so probably not.)

Where ripple reduction is key, an LC filter stage is far more effective than bulk capacitance.  1MHz shouldn't need very much filtering, but we have no sense of scale (is this at 1W or 1kW?).  Obviously you need enough capacitor to handle the ripple current; any more beyond that, may be reasonable to provide with aluminum polymer or electrolytic types.  (Not that polymers are much cheaper, but they are available in much larger values.  Both do suffer from reduced lifetime at high temperature, something to consider.  Tantalum and ceramic are the alternatives, and tantalum may be undesirable because of its poor handling of surge current (tendency to deflagrate) or conflict mineral use.  It's not clear if you've considered other types, so this fills out the list just in case.)

Tim
Navigation
Message Index
Next page
There was an error while thanking
Thanking...

Go to full version
Powered by SMFPacks Advanced Attachments Uploader Mod