EEVblog Electronics Community Forum

Electronics => Projects, Designs, and Technical Stuff => Topic started by: frogblender on April 16, 2021, 04:22:32 am

Title: Ceramic cap self-heating: crap blows up?
Post by: frogblender on April 16, 2021, 04:22:32 am
In the last decade or two, ceramic caps seem to have alot more capacitance for a given size (ie. capacitance per cubic mm has shot up)  ie. 47uF @6.3v in an 0603?  Yup.

So... I'm designing a product that has a 300A @ 800mV 9-phase CPU supply,  and I want to fit it into not much bigger than a cellphone.  Anyhoo, I need lots o farads for that big supply, to get ripple, both input and output, down to reasonable levels.  And cuz of phone sizing I want to change the existing 1210 caps to 0805:  the 1210's and 0805's have identical specs: 22uF 16V 10% x5r (and the smaller 0805's are even cheaper, on digi key, at least).

The supply will cause ripple on the input and output rails, and that ripple flows in and out of the caps, at 1MHz (or whatever, I can't recall the supply's frequency)... and the current flow through the caps' ESR heats them up.   But what the datasheets don't fully spec is the self-heating limits, other than "do not allow self-heating to exceed 20°C".   And actual ESRs are not really spec'd, so I can't accurately calculate expected heating, and the fact that 0805's generally have less than half the power handling of a 1210 has me humming/hawing.

So the question:  anyone have any experience using physically tiny ceramic caps on large-current (and large ripple) supplies?  Or can shed light in general on the above scenario in regards to ceramic cap esr heating?  Help is appreciated.



Title: Re: Ceramic cap self-heating: crap blows up?
Post by: frogblender on April 16, 2021, 04:31:34 am
TDK has a FAQ on this subject:   https://product.tdk.com/en/contact/faq/capacitors-0029.html

Reading through this reveals this self-heating business is somewhat uncharted, poorly-quantified territory :(


Title: Re: Ceramic cap self-heating: crap blows up?
Post by: Berni on April 16, 2021, 05:28:32 am
Usually ceramic capacitors have such a low ESR that nobody worries about maximum ripple current or self heating. Its more of a problem with electrolytics and cheap tantalums.

But you seam to have a rather special application of them. At your point you will likely have to consider the resistance of your PCB tracks and flood fills, along with what sort of heating that also causes.

What you can do when the manufacturers fails or forgets to characterize what you need is simply do it yourself. You can buy a few different capacitors and test them yourself for what you need. For example LC resonant circuits are very good at pushing massive ripple currents trough capacitors.  Build such a cirucit with the beefyest inductor you can find and use something like a ZVS oscillator to easily pump power into it. What you end up is potentially many 10s of amps of sinewave shaped current running trough your capacitor. Then you can just measure how hot the capacitor gets or place some extra voltage probes on it to measure the ESR at 1MHz via ohms law. Should give you a good idea of what the difference between the big and small caps is.
Title: Re: Ceramic cap self-heating: crap blows up?
Post by: james_s on April 16, 2021, 06:02:54 am
There's something to keep in mind about ceramic capacitors that is less well known than it ought to be and can really bite you. Capacitance drops when there is a DC voltage across them, quite sharply in some cases, and this effect tends to be more pronounced the smaller you go in physical size. That 22uF 0603 capacitor might be only 3uF once you have 12V across it. Study the datasheet carefully.
Title: Re: Ceramic cap self-heating: crap blows up?
Post by: Miyuki on April 16, 2021, 06:10:26 am
For CPU voltage are special capacitors with huge capacities and packages more looking like a qfn optimized for voltages about that 1V and huge currents
Like this fun
https://www.bulcomp-eng.com/datasheet/Tokin%200E907.pdf (https://www.bulcomp-eng.com/datasheet/Tokin%200E907.pdf)
Title: Re: Ceramic cap self-heating: crap blows up?
Post by: Kleinstein on April 16, 2021, 03:56:37 pm
300 A as a 9 phase regulator is still quite a lot of current per phase. More phases reduce the ripple current - with the low supply votlage one may have to group them properly, so that the ripple current actually flows between the phases and not all to the caps.

With the ESR, there is actual resistance and also dielectric loss. There can also be Eddy currents and the magnetic flield pushing the current to the outside - so the actal loss may be higher than the simple addition of the single components.

Chances are the capacitors would also be thermally coupled to the rest of the circuit - thick conductors for the current also provide good heat transfer in both directions. So it is not only the loss of the caps, but more like a coupled system. 20 K temperature rise due to self heating still sounds moderate.

Capacitor loss can be significant. I have seen water cooled mica caps in an induction heater. Not sure if really needed, as water cooling was there anyway, so it was low effort.
Title: Re: Ceramic cap self-heating: crap blows up?
Post by: T3sl4co1l on April 16, 2021, 05:51:11 pm
It's not so unusual; I've seen amperage ratings in datasheets before.

Hm, I forget offhand which brands I've seen using them.

Another rough guess is this: find ESR at the required frequency.  (If they don't have impedance and C(V) curves, keep shopping.)  Assume power dissipation is half that of an equivalent sized resistor -- for 0805, that would be about 1/16W.  The justification for this is, the ceramic material is lower thermal conductivity than the alumina substrate of the resistor, so the internal temperature rise will be proportionally higher.  The temp rise also shouldn't be too high, as that affects capacitance, and probably has implications for mechanical stress (and eventual cracking).  Put these two together, and that shall be your current limit: I = sqrt(P/ESR).

One or two amperes should be about right for 0805.  With 9 phases, you'll have ballpark 1/9th ripple or 33A total, which should fit in the required area -- heh, but you may want to stack some things to get them in there and the inductors and the switches (2-sided load at least, but even multiple boards with board-to-board headers may be an option).


With the ESR, there is actual resistance and also dielectric loss. There can also be Eddy currents and the magnetic flield pushing the current to the outside - so the actal loss may be higher than the simple addition of the single components.

Yes, current capacity drops with rising frequency.  This is a good reason to stick to smaller parts.  I'm not sure where the cutoff is exactly -- you'd need a figure for the effective resistance of the capacitor (metallization and dielectric*) to calculate that.  I wouldn't recommend anything bigger than 1210 anyway, for mechanical reasons if nothing else; whatever is probably fine at 1MHz.

*Yes, dielectric exhibits skin effect as well -- since it supports more propagation than loss, it's generally better to model it as, well, a lossy dielectric; but it works this way, too.  The cool thing about physics is, it doesn't matter how the loss arises, just that it's there -- anywhere there's an equivalent bulk resistance, you will find skin effect!

Analogously, this also matters in ferrite, but as the loss is modest, it takes quite a thickness to have much skin effect -- irrelevant in most electronics, but actually an important consideration in very large transformers, such as used in power transmission converters, induction heating, etc.  Think 100s kW at 10s kHz, with cores the size of your leg. :)  (Which aren't standard parts, and good luck ordering them in such small quantities as these -- they're usually built from ferrite bricks glued together, which also conveniently provides air gap between blocks, permitting magnetic field to surround them, much as laminated iron serves for mains frequency transformers.)


Quote
Capacitor loss can be significant. I have seen water cooled mica caps in an induction heater. Not sure if really needed, as water cooling was there anyway, so it was low effort.

This one is easier to calculate -- simply take the total ESR and the total ripple currents (input and output), after all.  The nice thing about bypass is, you can just use more capacitance, and the reactive power keeps going down.  Reactive power being ripple voltage times ripple current.  We can reduce voltage arbitrarily, and thus we can also afford to use quite lossy capacitors (type 2 dielectric).  This is of great importance to resonant supplies (the active components need quite high Q), and other resonant applications of course!

Induction heaters usually have the double whammy of needing high power density as well as high reactive power, so it's almost impossible to get anything done without water cooling, at least above a few kW.

Tim
Title: Re: Ceramic cap self-heating: crap blows up?
Post by: Berni on April 16, 2021, 06:01:27 pm
Capacitor loss can be significant. I have seen water cooled mica caps in an induction heater. Not sure if really needed, as water cooling was there anyway, so it was low effort.

With big serious induction heaters the water cooling is indeed required because the ripple currents they experience are insane. Tho its not how you might picture a capacitor with pipe fittings on it. They actually just construct the capacitors as flat pucks with the whole top and bottom being a terminal, then sandwich these between bus bars that have water flowing trough them, providing the beefy electrical connection and cooling simultaneously.

Even small induction heaters ran using single regular troughhole transistors need some well speced caps and offten multiple in parallel, using just a regular film cap makes them overheat and blow up. I seen the troughhole pin on capacitors start glowing like a light bulb before.

But with these low voltages most of the losses tend to be the actual resistive losses from simply ESR. But that being said (as others pointed out) the more capacitance dense ceramic cap materials can be really crappy sometimes. Its not unusual for them to lose over half of there capacitance near the rated voltage. The capacitance varies by 10s of % just by changing temperature and they are so non linear that it can make the startup waveform look weirdly wobbly. So definitely do your homework on these.
Title: Re: Ceramic cap self-heating: crap blows up?
Post by: T3sl4co1l on April 16, 2021, 06:06:59 pm
Even small induction heaters ran using single regular troughhole transistors need some well speced caps and offten multiple in parallel, using just a regular film cap makes them overheat and blow up. I seen the troughhole pin on capacitors start glowing like a light bulb before.

Hah, I've done that before with just an electrolytic -- turns out one capacitor simply wasn't enough for the ripple current the thing made.  Ran out of space to put any more. :P https://www.seventransistorlabs.com/Images/CukHack4.jpg (https://www.seventransistorlabs.com/Images/CukHack4.jpg)

Tim
Title: Re: Ceramic cap self-heating: crap blows up?
Post by: floobydust on April 16, 2021, 06:07:18 pm
For a given physical size, smaller packaged MLCC parts have a higher voltage coefficient of capacitance and thus lower effective capacitance when you include the DC bias. So 1210 and 0805 with identical specs have much different net capacitance. Also, a smaller part can dissipate less heat.

Taiyo Yuden (https://www.yuden.co.jp/ut/product/#) MLCC datasheets are very good, they include a ripple current and DC bias spec.
Worth looking at those and comparing value w/DC, voltage, size tradeoffs e.g. 100uF 2.5V 0805 and 47uF 6.3V 0603 parts, sometimes the best performing part is not what you'd think.
Title: Re: Ceramic cap self-heating: crap blows up?
Post by: wraper on April 16, 2021, 06:07:47 pm
For CPU voltage are special capacitors with huge capacities and packages more looking like a qfn optimized for voltages about that 1V and huge currents
Like this fun
https://www.bulcomp-eng.com/datasheet/Tokin%200E907.pdf (https://www.bulcomp-eng.com/datasheet/Tokin%200E907.pdf)
Crap which caused failures in millions of laptops. BTW not all MLCC are equal. They come in regular and ultra low ESR types.
Title: Re: Ceramic cap self-heating: crap blows up?
Post by: Bud on April 16, 2021, 06:45:07 pm
In general, for a given voltage rating and nominal capacitance value, smaller physical size capacitor will have smaller actual capacitance under DC bias. Charts for dependency of capacitance drop vs physical size can be found on inet.

Edit: that is in addition to the drop caused by the dielectric type.
Title: Re: Ceramic cap self-heating: crap blows up?
Post by: tszaboo on April 16, 2021, 11:19:54 pm
That is not as much power and current density as you think. But you also need bulk capacitance. Forget MLCC for that, you probably never reach the required bulk capacitance with it. Use POSCAP or similar. ESL could easily matter more than ESR, and you probably end up with a hundred capacitors in your design.
Title: Re: Ceramic cap self-heating: crap blows up?
Post by: David Hess on April 17, 2021, 02:10:02 am
I would be looking at metal frame MLCC capacitors which are intended for power applications where larger values are needed.  Their datasheets give detailed ESR versus frequency data and some idea of ripple current capability.  Given their specifications, I think you are severely overestimating the capability of 0805 and 1210 parts.
Title: Re: Ceramic cap self-heating: crap blows up?
Post by: amyk on April 17, 2021, 04:40:40 am
For CPU voltage are special capacitors with huge capacities and packages more looking like a qfn optimized for voltages about that 1V and huge currents
Like this fun
https://www.bulcomp-eng.com/datasheet/Tokin%200E907.pdf (https://www.bulcomp-eng.com/datasheet/Tokin%200E907.pdf)
Crap which caused failures in millions of laptops.
That was my first thought on seeing that name... a 1000h@105C cap right under a hot CPU, what else can you call it but planned obolescence. :palm:
Title: Re: Ceramic cap self-heating: crap blows up?
Post by: james_s on April 17, 2021, 06:11:10 am
Crap which caused failures in millions of laptops.
That was my first thought on seeing that name... a 1000h@105C cap right under a hot CPU, what else can you call it but planned obolescence. :palm:
[/quote]

An oversight? Incompetence?

Design errors are not rare. They have killed otherwise decent products and even sunk entire companies when products fail in large numbers well before the expected lifespan. I'm not familiar with the particular laptop in question but until relatively recent times there was no need to consider any kind of deliberate obsolescence with a laptop, technological progress would obsolete them within a few years. A decoupling capacitor failing and causing the machine to get flaky and unstable also seems like a really foolish way to engineer in obsolescence. Nobody in their right mind is going to go "Wow this thing is a piece of shit, it's 6 months old and crashing constantly. I'm gonna go out and buy the latest model from this company!"