Does using two caps with half capacity in parallel lowers the ESR?

[sb1370]

Does ESR reduces if instead of one cap, we use two caps with half the capacity in parallel, in practice? Because according to ESR=tan δ / (C*f*2π), there would be no change if we do it.
Also what's the effect on ripple current? does it add up if we put caps in parallel?

[ogden]

Does ESR reduces if instead of one cap, we use two caps with half the capacity in parallel, in practice? Because according to ESR=tan δ / (C*f*2π), there would be no change if we do it.
Also what's the effect on ripple current? does it add up if we put caps in parallel?

It depends. What happens in practice is different to theory & formulas. Aswer is: refer to datasheet, do the math. As I already said in other thread - it is bad idea to put multiple capacitors where it is designad to have one, especially if you care about low ESR, impedance. Reason: capacitor leads have not only resistance but inductance as well. Impedance of inductor is frequency-dependent. Bunch of multiple capacitors with nonzero length leads may be good at low frequencies but completely fail their job filtering high frequency ripple in the SMPS application.

[Gyro]

This seems to be the same subject as  https://www.eevblog.com/forum/repair/cant-find-low-esr-cap-alternative-solution/

Please don't double post, you're already getting plenty of help in that thread and wasted duplication of replies annoys people.

[T3sl4co1l]

Note that, if the capacitors are perfectly identical in formulation (same C*ESR product, same V), the ESR is probably around the same but the total ripple current or power dissipation rating may be higher.  There's more surface area, and less distance to the middle, for two smaller caps versus one larger.

I don't know that this is actually generally applicable.  It doesn't matter what you might suspect, ripple ratings are given in the datasheet and the manufacturer knows best!

Tim

[sb1370]

This seems to be the same subject as  https://www.eevblog.com/forum/repair/cant-find-low-esr-cap-alternative-solution/

Please don't double post, you're already getting plenty of help in that thread and wasted duplication of replies annoys people.

Sorry, I just wanted to know the answer from the basic and general POV rather than specific application.

Note that, if the capacitors are perfectly identical in formulation (same C*ESR product, same V), the ESR is probably around the same but the total ripple current or power dissipation rating may be higher.  There's more surface area, and less distance to the middle, for two smaller caps versus one larger.

I don't know that this is actually generally applicable.  It doesn't matter what you might suspect, ripple ratings are given in the datasheet and the manufacturer knows best!

Tim

Maybe I phrased my question badly, I meant if we put two caps in parallel what will happed with the ripple current? it'd be sum of individuals?

[T3sl4co1l]

Yes, ripple currents add in parallel.

[Siwastaja]

Note that the series inductances affect ripple current sharing, so if you have two capacitors to bypass a hard-switching loop, the closer one will take a higher portion of the ripple current. It will also see more heating from the switching semiconductors being closer to them.

For low-voltage converters, the "silver bullet" is to use high enough f_sw to allow low enough C to be used for a MLCC solution. Their ESR is small enough so that heating usually isn't a concern. Electrolytics may still be used for damping but in that case they take a fairly small part of the ripple current.

You can simulate all of this quite well in SPICE. Play around with online wire/loop inductance calculators to get a rough idea how much L your layout adds to each capacitor.

[Vovk_Z]

Does ESR reduces if instead of one cap, we use two caps with half the capacity in parallel, in practice?

In general you don't have to think about ESR when you parallel caps but you have to take into consideration a ripple current instead of. Because ESR is a parameter rather of a single cap but not a group. And when you parallel them the parameters change in a not straight way. I mean a large amount of a high ESR caps doesn't fold together an one good low-ESR cap.

[David Hess]

Does ESR reduces if instead of one cap, we use two caps with half the capacity in parallel, in practice? Because according to ESR=tan δ / (C*f*2π), there would be no change if we do it.

Assuming no difficulties in layout, ESR and ESL are decreased by the paralleled leads and decreased average distance to the plate surfaces, but see below about layout.

Also what's the effect on ripple current? does it add up if we put caps in parallel?

The ripple current rating increases because of lower ESR and because of the increase in surface area to volume which aids cooling.

Note that, if the capacitors are perfectly identical in formulation (same C*ESR product, same V), the ESR is probably around the same but the total ripple current or power dissipation rating may be higher.  There's more surface area, and less distance to the middle, for two smaller caps versus one larger.

I don't know that this is actually generally applicable.  It doesn't matter what you might suspect, ripple ratings are given in the datasheet and the manufacturer knows best!

When operating at close to the ripple current limit, it makes a significant difference.  Some designs are not even feasible with practical parts unless smaller capacitors are used in parallel although this is much less of a problem now with better capacitors.

Note that the series inductances affect ripple current sharing, so if you have two capacitors to bypass a hard-switching loop, the closer one will take a higher portion of the ripple current. It will also see more heating from the switching semiconductors being closer to them.

Careful layout is necessary for good current sharing, but usually large flood fills are sufficient.

For low-voltage converters, the "silver bullet" is to use high enough f_sw to allow low enough C to be used for a MLCC solution. Their ESR is small enough so that heating usually isn't a concern. Electrolytics may still be used for damping but in that case they take a fairly small part of the ripple current.

Better capacitors and multi-phase designs made a huge difference.  I have only seen MLCCs used in military, aerospace, and harsh environment designs and that has been done going back decades.

[bdunham7]

Does ESR reduces if instead of one cap, we use two caps with half the capacity in parallel, in practice? Because according to ESR=tan δ / (C*f*2π), there would be no change if we do it.
Also what's the effect on ripple current? does it add up if we put caps in parallel?

It is said that 'you might be an engineer' if you spend three hours debating something easily resolved with a two minute experiment.

[ssashton]

Only using identical caps though - if you parallel different caps you will broaden (hopefully) the lowest range of impedance curve. Obviously adding 10uF and 100nF in parallel will not lower ESR as such shown above. Dave did a brilliant video showing the impedance curves of some caps. Anyone able to link it in?

Also worth knowing that too low ESR can cause circuit instability. You can make even a LM317 regulator unstable with super-dooper low ESR on the output.

[David Hess]

It is said that 'you might be an engineer' if you spend three hours debating something easily resolved with a two minute experiment.

I do not know why anybody would debate that two of the same capacitor in parallel would have half the ESR.  Now compare parallel capacitors in the same series but half the capacitance each for the same total capacitance.

The basic idea though is that the same capacitance divided among more capacitors has lower ESR and ESL.

[bdunham7]

It is said that 'you might be an engineer' if you spend three hours debating something easily resolved with a two minute experiment.

I do not know why anybody would debate that two of the same capacitor in parallel would have half the ESR.  Now compare parallel capacitors in the same series but half the capacitance each for the same total capacitance.

The basic idea though is that the same capacitance divided among more capacitors has lower ESR and ESL.

I don't have a 1000uF of the same type to compare.  I suppose it could go either way.  The claim was made that the extra lead length required to install 2 caps in the place where one was originally would induce enough extra ESR and ESL to negatively affect the performance of a switching supply.  Since the OP's device probably has a switching frequency on the order of 50kHz, I doubt it.  The caps I'm measuring were not selected for an SMPS, but rather the linear supply in a Fluke DMM, so they aren't super-low ESR or anything.  But here is Z 10hz to 500kHz:

[T3sl4co1l]

It's flat, not even a pixel off, despite adding 2cm or more of lead length?  (Nevermind what variation in the clip leads might be.)

Plot looks to be log, so a pixel is about 10%.  10% of 60mohm is 6mohm.  6mohm at 500kHz is ~2nH inductance.  The pictured connection easily adds ten times more, therefore the measurement is erroneous.

Nice to know it can't measure inductances that small, I guess?

Only using identical caps though - if you parallel different caps you will broaden (hopefully) the lowest range of impedance curve. Obviously adding 10uF and 100nF in parallel will not lower ESR as such shown above.

Obviously it will -- but it depends where you're measuring it, and what you're adding.  If those are ceramic caps, their ESR valleys will be around 10 and 100mohms respectively.  They will dominate over Fc = 1 / (2 pi (100mohm) (10uF)) = 160kHz.  If you're working below that frequency, the 100mohm electrolytic will dominate, but above, the ceramic(s) will dominate.  (Also, if there's say 5nH between the two ceramics, they'll be parallel resonant around 7.1MHz with Zo = 0.22 ohm, which will be reasonably damped by the ESR of the 0.1 or the electrolytic.  Best placement would be with the 0.1 towards the electrolytic.)

This is also a handy method to keep harmonics, if not the most powerful fundamental, away from electrolytics that aren't quite rated for enough ripple current otherwise.  Use ceramics at low voltage, or film at high voltage.  (Or also, at low voltage, just using polymers, which are basically films if they could be made in proportionally low voltages and high capacities -- their ESR, ripple rating, and energy density are quite similar.)

Tim

[bdunham7]

It's flat, not even a pixel off, despite adding 2cm or more of lead length?  (Nevermind what variation in the clip leads might be.)

Plot looks to be log, so a pixel is about 10%.  10% of 60mohm is 6mohm.  6mohm at 500kHz is ~2nH inductance.  The pictured connection easily adds ten times more, therefore the measurement is erroneous.

Nice to know it can't measure inductances that small, I guess?

I calculate about 30nH, give or take.  As for the measurement accuracy, I've no idea really. My only other comparison tops out at 100kHz.  The machine appears OK in basic tests, but I've not tried sweeps before and there may be settings I'm overlooking.  Or, perhaps my test capacitor is more complex than simple elements can describe.

If you look at the scale, you can see that the long-lead version has about 8mohm more Z.  The lead resistance should be less than 2mohm (20mm x 0.8mm x2), so there is some additional Z, just not as much as expected by a simple model.  The flat tail is not what you were expecting--you expect Z to turn up at some point, right?  I was expecting that too, but there it is.  So what would you expect a plot of Θ to do with the same frequency sweep?

[bdunham7]

So the plot isn't log, it is....something else.  I did this again but from 10kHz to 500kHz, just to get the scale down to where it makes sense.  And I'm not actually doing the measurements as pictured, I have the cap down on a table with the clips hooked up nice and straight each time, then I hold it up to the screen so I can keep track of which measurement is which.  I think the kelvin clip setup is probably good to 100kHz, but 500kHz is a stretch.  They're pretty touchy and I've gotten several obviously erroneous readings that I've had to do over.  I don't have any fancy fixtures for the Quadtech.

Now if I had a 1000uF from the same series, we could compare.  But I'm not sure what the point was anymore.  Nothing here is going to prevent two of these from working in the OP's power supply.

[ogden]

Nothing here is going to prevent two of these from working in the OP's power supply.

Yes they will work as capacitors indeed

Switching supplies w/o proper output filters outputs broad harmonics spectrum up-to tens of MHz. Design of that supply do not expect 15nH wire between power rail and output filter capacitor. When you add capacitor with long leads - hi-frequency ripple will remain unfiltered less filtered. Downstream circuit may be fine, but your neighbors listening radio may not. 500KHz RLC is fine tool indeed, yet impedance analyzer going at least up-to 10MHz will better demonstrate capacitor impedance increase at high frequencies due to parasitic inductance of it's leads. Other option: >= 20MHz scope, soldering iron, SMPS supply, capacitor with long leads at the beginning and short in the end.

[bdunham7]

I don't think that capacitor alone is going to effectively filter out "10s of MHz" in any case, regardless of lead length.  I won't post another photo, but when I measure the cap as an inductor at 500kHz, it is about 60nH with a Q of 2.8.  Or, 90nH with the ridiculously long leads and a Q of 3.5.   So Z @ 10MHz would be about 3.8 ohms with short leads or 5.6 with long.  I'm just not seeing that as a critical factor when you consider all of the other places you have inductance, like the PCB. 

The OP's power supply does have an additional filtering.  He mentioned an inductor and another capacitor, which is common in this specific application.  There's probably a small bypass cap or two hiding somewhere too.  Nobody uses 1000uF el-caps to filter out 10Mhz+ noise.

[T3sl4co1l]

Ah, I see, it is linear after all.  Only the lowest tick mark has a decimal, doesn't it.  Yeah, so you can't read bupkis for ESL at that scale.

Tim