What the hell happened to friggin' ceramic caps ???

[frogblender]

Is it just me, or have ceramic caps improved through the roof in recent years?  It seems prices have fallen, and microfarads have gone way way WAY up for a given package size.
Example:  47uF 6.3V 0805, costing two lousy pennies in quantity (CL21A476).  Or a 100uF 6.3V 1206 for about double that.  And you can get the above in 0603 and 0805 respectively, if you want to spend nickels instead of pennies.

Man, when I was a boy...

And I don't mean the cheap crap from the HappyDongDragonCorp - we're talking the good sh1t.... Murata, Samsung , Kemet, Yageo, Taiyo Yuden.

Why are the 400-amp power supply designers at nvidia and AMD still futzing with oscons and poscons and wear-eye-protection-tants, when they could switch entirely to ceramics?   Yes, I understand when you operate your ceramics up towards their rated voltage, a good chunk of your microfarads will simply cease to exist.... but ceramics are superior in every other way.

[coppice]

Man, when I was a boy...

How long ago were you a boy? It was long ago when the industry cracked high density ceramic "capacitors", which are good for decoupling, if pretty useless for anything else. I used quotes, as something varying so wildly in character as you excite it is barely a capacitor, and seriously funky in applications like linear filters.

[tszaboo]

Lower voltage ones have been improved, and supply has been much better than ~ 5 years ago. Looks like they have been improving on the number of plates, and reducing the distance between the plates. But don't let the high capacity fool you, they are de-rated a lot with voltage. And I suspect we are seeing the benefits of mass production passed down to us. I mean I don't use several millions of capacitors per year, but the price saving of making that many made them cheaper.

[Bud]

.... but ceramics are superior in every other way.

No. I was designing a switching power supply few years back and used a spectrum analyzer to view and minimize switching noise. The spectrum analyzer clearly indicated ceramic capacitors did not work as good as tantalum ones of a same capacitance value in suppressing kHz range noise. I ended up using tantalums. Ceramics were better at hundred kHz and lower MHz range.

[Someone]

Is it just me, or have ceramic caps improved through the roof in recent years?

Not really:
https://www.eevblog.com/forum/microcontrollers/bypass-caps/msg930381/#msg930381
https://www.eevblog.com/forum/chat/capacitor-roundup-2024/msg5403863/#msg5403863
8 year delta, movement but not radical change.

Why are the 400-amp power supply designers at nvidia and AMD still futzing with oscons and poscons and wear-eye-protection-tants, when they could switch entirely to ceramics?   Yes, I understand when you operate your ceramics up towards their rated voltage, a good chunk of your microfarads will simply cease to exist.... but ceramics are superior in every other way.

See the above, poly caps are highly competitive in both volumetric density and winning in price when you need >100uF.

[Vertamps]

I just had to replace a ceramic cap in a brand new product, it was shorted on the input DC filtering before an buck converter circuit. Who knows why it shorted, voltage was in spec. Could have been physical stress though the reflow oven, micro crack allowed moisture to short it? Manufacturing design would have to be rigorous for ceramics, mayby not at much for resistors. Actually we have had larger size ceramic (thick film) resistors fail open(crack) in boards at work. Bigger ceramics may be bigger risk to physical stress damage.
I am tempted to replace old Tantalum's in gear with ceramics, but i end up buying more tants. 

[ConKbot]

"a good chunk" 95%+ in some cases, lol.

Decouple a bus with 16V or higher, then get back to me.

Bonus points if it's not going to crack and short the bus out with a teeny bit of board flex, so you use leaded SMT parts, or don't go above 2220 size caps.

[Psi]

I will be happy when you can get 50uF at 50V without having to resort of metal frames containing multiple caps, and when the price makes sense and it's not cheaper to just put those multiple caps on the PCB yourself.

eg, a larger size 50V 50uF single SMT cap at the same or less cost as 5x 10uF 50V 0805

[KE5FX]

I just had to replace a ceramic cap in a brand new product, it was shorted on the input DC filtering before an buck converter circuit. Who knows why it shorted, voltage was in spec. Could have been physical stress though the reflow oven, micro crack allowed moisture to short it? Manufacturing design would have to be rigorous for ceramics, mayby not at much for resistors. Actually we have had larger size ceramic (thick film) resistors fail open(crack) in boards at work. Bigger ceramics may be bigger risk to physical stress damage.
I am tempted to replace old Tantalum's in gear with ceramics, but i end up buying more tants.

I've seen similar MLCC failures recently, twice in newly-manufactured units and once in a prototype that had been in use for a couple of years.  I don't expect ceramic caps to be less reliable than tantalums, but that appears to be what is happening

[frogblender]

.... but ceramics are superior in every other way.

No. I was designing a switching power supply few years back and used a spectrum analyzer to view and minimize switching noise. The spectrum analyzer clearly indicated ceramic capacitors did not work as good as tantalum ones of a same capacitance value in suppressing kHz range noise. I ended up using tantalums. Ceramics were better at hundred kHz and lower MHz range.

Yes.   You probably witnessed exactly what we speak of:  your ceramic microfarads start to vanish as you up the voltage, which is why the tant won at low freq.   At higher frequencies though, the ESR of the tant entirely eliminates it from your circuit, but the ceramic, despite losing 50% (or more) of its value, keeps on truckin'.

[Psi]

Yeah, I've also seen a rise in ceramic cap failures.
I suspect the PCBA company was buying cheaper grey market caps.

The problem went away when I changed the cap from 10uF 35V to 10uF 50V. (Not because I needed 50V, it only had 12V on it) Trying to save a few cents on 35V instead of 50V was not worth it.

10uF 50V is a much more common/generic cap which seemed to stop the PCBA company trying to source oddball stuff which caused the problem. IMHO

[coppercone2]

ok so high power density means thinner materials. there is some engineering to be done with materials but the fact is its a balony sandwich made of tera cotta

their fragile. I tried working with all manner of ceramic plates that are fairly thick (Aluminum nitride, aluminum oxide, boron nitride) and they are all really god damn fragile even at 3/32 of an inch. So is gorilla glass for replacing phone screens, super easy to break.


I see a problem with mother boards getting big ceramic capacitors because there is a bunch of heavy shit on them. Like really alot


and even for small caps, the #1 trouble shooting technique for things like phones and PC cards...is you guessed it, look for the cracked ceramic mini capacitor! Literally its the first thing any experienced repair guy goes after when it comes to a phone if they don't have a specific lead.

In order the combat this problem, the rubber decoupling capacitor was invented. They put a piece of conductive polymer into the ceramic capacitor. So now you have a weak link, that is the rubber. How will that stand up to the test of time is anyones guess. Its called something like 'flex mount'. How will the ESR of those change over time? And they are much less cheap, and they would not invent if ceramic capacitors were the real 'silver bullet' solution that is said in this thread.


I suspect they have a good reputation because they have backups (liberal decoupling), meaning a few cracks means it still works. Once they start getting placed frugally you might have a problem...

[Psi]

I remember reading that some ceramic caps made of exotic materials (high uF + low physical size) have an effect where you don't get the full capacity if you run them at low voltage vs their rated voltage.  Like 1V on 25V cap would only get a fraction of rated capacity.

However from searching I've not been able to find any reference to this so I'm probably mis-remembering or getting my wires crossed with the effect where normal ceramic caps read lower uF at higher voltages closer to their rated voltage.

Anyone know what i'm talking about? or am i getting confused?

[uer166]

I remember reading that some ceramic caps made of exotic materials (high uF + low physical size) have an effect where you don't get the full capacity if you run them at low voltage vs their rated voltage.  Like 1V on 25V cap would only get a fraction of rated capacity.

However from searching I've not been able to find any reference to this so I'm probably mis-remembering or getting my wires crossed with the effect where normal ceramic caps read lower uF at higher voltages closer to their rated voltage.

Anyone know what i'm talking about? or am i getting confused?

TDK Ceralink and other high voltage caps are prebiased towards operating voltage, see attached.

[Psi]

Maybe that was it.  Thanks.

[T3sl4co1l]

"Why back in my day..."

I remember even in say 2015, caps of modest values in 1206 and 1210 being over a buck in small quantity, from brand names, mainline distributors.  Of course it's always gone up from there, with larger chips, some leaded parts for some bizarre reason, and leadframe types, topping out the list, but I don't mean sorting to find exotics, I mean the minimum price given those settings.

Not that small quantities as major distributors has ever meant very much, but DK at least seems to track their low-qty price to the volume price.  But how closely varies between manufacturer (I assume based on service agreements and marketing).

Perhaps due to broader market availability of Chinese types (Digikey stocks Walsin and Holy Stone now, for example), everyones' prices have dropped.

The irony is not lost on those of us who "survived" the 2018 ceramic capacitor shortage, which gave reasons including "closing factory due to flat profits" or something to that effect; as we've learned since then, and across many more markets, this probably actually meant "we can't afford to pay our CEO and private investors what they demand so we're killing that line, sorry not sorry".  Presumably, after some much needed restructuring, they're back to their senses and competing still.

Tantalum has also been largely obviated, perhaps in part from pressure as a conflict mineral, but also as ceramics and polymers have filled much of their use, and regulators have largely advanced past the need for the output damping afforded by them.

Why are the 400-amp power supply designers at nvidia and AMD still futzing with oscons and poscons and wear-eye-protection-tants, when they could switch entirely to ceramics?   Yes, I understand when you operate your ceramics up towards their rated voltage, a good chunk of your microfarads will simply cease to exist.... but ceramics are superior in every other way.

Hm, I haven't even seen a branded OSCON in forever.  But as a genericized term for "aluminum polymer", yeah, that's what they use, widely.  Polymers offer much higher value, per area and cost, than ceramics.  Unless you get big chips (smaller market size; perhaps lower yield?) or leadframe types ($$$).  ESR and current handling are, practically speaking, roughly comparable.

Polymers are also available in chip styles, which you may've confused for tantalum.

Tim

[T3sl4co1l]

"a good chunk" 95%+ in some cases, lol.

Decouple a bus with 16V or higher, then get back to me.

Bonus points if it's not going to crack and short the bus out with a teeny bit of board flex, so you use leaded SMT parts, or don't go above 2220 size caps.

Yeah, the C(V) of usable size chips falls off ponderously fast with voltage.  IIRC, IPC recommends stopping at 1210 anyway, but maybe pushing it to 1812 is still fine, and it's true 2220s exist, but they're so expensive to begin with.  So, 1210 is really the largest meaningful choice.  And a, say, 22uF 50V cap used at 12 or 24V (where it'll be maybe 7uF remaining, if that), is more or less equivalent to a 200V cap at 1.5uF (that has maybe 0.4uF remaining at 100V).

You can at least get larger values in higher voltages -- probably, the thicker dielectric reduces pinhole defects, allowing ultimate breakdown to be approached more closely -- but this just means even steeper C(V) reduction, it sucks.

C0G, at least, perform very well -- they have modest density (33nF 630V is abundant in 1210), and the energy density at rating is excellent, better than electrolytic IIRC.  Downside: they're still rather expensive in quantity.  Need 10s of uF at 630V? Eek!

Leadframes at least deal with cracking, but they're also ludicrously expensive, at least what I've seen in small quantities from major distributors.  AFAIK, these are being used in automotive among other applications, so there must be either sufficient advantage in using them (certainly a help in a high vibration, long life application), or someone getting good deals on them -- factory direct, or custom runs even, perhaps.

Tim

[Doctorandus_P]

I guess they got frigged, although I'm not exactly sure what that means.

I'm still waiting for Mega Farad capacitors on a decent voltage and about matchbox sized. It would finally solve the battery problem.

[T3sl4co1l]

Well, that's easy. It's impossible.

A battery works by electrochemical reaction, developing a voltage logarithmic with concentration of reactants.  A battery can be seen as a nonlinear capacitor if you like, but I'm assuming you mean a linear one, or at least a more meaningfully linear one.

A capacitor necessarily must operate with less than ionization/redox energy of the materials it's constructed of, else it would just be a battery again.  But clearly it will have less energy density.

Tim

[coppercone2]

this is one of the solutions for cars

https://www.kyocera-avx.com/products/ceramic-capacitors/surface-mount/automotive-mlcc-with-flexiterm/

but its rubber polymer or something.

similar to TDK



then the next trick is this. its leaded now. No reason you can't make the leads longer and solder it into PCB slots lol

[tooki]

Yeah, the C(V) of usable size chips falls off ponderously fast with voltage.  IIRC, IPC recommends stopping at 1210 anyway, but maybe pushing it to 1812 is still fine, and it's true 2220s exist, but they're so expensive to begin with.  So, 1210 is really the largest meaningful choice.  And a, say, 22uF 50V cap used at 12 or 24V (where it'll be maybe 7uF remaining, if that), is more or less equivalent to a 200V cap at 1.5uF (that has maybe 0.4uF remaining at 100V).

You can at least get larger values in higher voltages -- probably, the thicker dielectric reduces pinhole defects, allowing ultimate breakdown to be approached more closely -- but this just means even steeper C(V) reduction, it sucks.

C0G, at least, perform very well -- they have modest density (33nF 630V is abundant in 1210), and the energy density at rating is excellent, better than electrolytic IIRC.  Downside: they're still rather expensive in quantity.  Need 10s of uF at 630V? Eek!

Do you mean “precipitously fast” with voltage? (“Ponderous” means “slow and clumsy”, which almost certainly isn’t the behavior of the curves in question.) If you do actually mean “slow and clumsy”, can you clarify what you meant?

[frogblender]

Yeah, the C(V) of usable size chips falls off ponderously fast with voltage.  IIRC, IPC recommends stopping at 1210 anyway, but maybe pushing it to 1812 is still fine, and it's true 2220s exist, but they're so expensive to begin with.  So, 1210 is really the largest meaningful choice.  And a, say, 22uF 50V cap used at 12 or 24V (where it'll be maybe 7uF remaining, if that), is more or less equivalent to a 200V cap at 1.5uF (that has maybe 0.4uF remaining at 100V).

You can at least get larger values in higher voltages -- probably, the thicker dielectric reduces pinhole defects, allowing ultimate breakdown to be approached more closely -- but this just means even steeper C(V) reduction, it sucks.

C0G, at least, perform very well -- they have modest density (33nF 630V is abundant in 1210), and the energy density at rating is excellent, better than electrolytic IIRC.  Downside: they're still rather expensive in quantity.  Need 10s of uF at 630V? Eek!

Do you mean “precipitously fast” with voltage? (“Ponderous” means “slow and clumsy”, which almost certainly isn’t the behavior of the curves in question.) If you do actually mean “slow and clumsy”, can you clarify what you meant?

I'm pretty sure he means capacitance drops with increasing voltage.   As in: capacitance goes to hell in a handbasket.   Turns to poop.   Goes up schitt's creek (with or without a paddle).   Takes a direct route to Destination F*cked.    Fast train to Schittsville.  Fubarred.  The wheels fall off.   Total paint on the iceberg.  Off the rails.  Rode hard and put away wet.

Somebody help me out here....

[tooki]

Yeah, the C(V) of usable size chips falls off ponderously fast with voltage.  IIRC, IPC recommends stopping at 1210 anyway, but maybe pushing it to 1812 is still fine, and it's true 2220s exist, but they're so expensive to begin with.  So, 1210 is really the largest meaningful choice.  And a, say, 22uF 50V cap used at 12 or 24V (where it'll be maybe 7uF remaining, if that), is more or less equivalent to a 200V cap at 1.5uF (that has maybe 0.4uF remaining at 100V).

You can at least get larger values in higher voltages -- probably, the thicker dielectric reduces pinhole defects, allowing ultimate breakdown to be approached more closely -- but this just means even steeper C(V) reduction, it sucks.

C0G, at least, perform very well -- they have modest density (33nF 630V is abundant in 1210), and the energy density at rating is excellent, better than electrolytic IIRC.  Downside: they're still rather expensive in quantity.  Need 10s of uF at 630V? Eek!

Do you mean “precipitously fast” with voltage? (“Ponderous” means “slow and clumsy”, which almost certainly isn’t the behavior of the curves in question.) If you do actually mean “slow and clumsy”, can you clarify what you meant?

I'm pretty sure he means capacitance drops with increasing voltage.   As in: capacitance goes to hell in a handbasket.   Turns to poop.   …

Well I assumed that too, which is why I said “Do you mean ‘precipitously fast’ with voltage?”.

…
Goes up schitt's creek (with or without a paddle).   
…
Fast train to Schittsville.

I assume you are a fellow Schitthead?  

[T3sl4co1l]

Do you mean “precipitously fast” with voltage? (“Ponderous” means “slow and clumsy”, which almost certainly isn’t the behavior of the curves in question.) If you do actually mean “slow and clumsy”, can you clarify what you meant?

Heh... in the sense of "to make one ponder", I think.  But that is indeed not the common meaning, and the above would be more correct

Tim

[ConKbot]

Yeah, the C(V) of usable size chips falls off ponderously fast with voltage.  IIRC, IPC recommends stopping at 1210 anyway, but maybe pushing it to 1812 is still fine, and it's true 2220s exist, but they're so expensive to begin with.  So, 1210 is really the largest meaningful choice.  And a, say, 22uF 50V cap used at 12 or 24V (where it'll be maybe 7uF remaining, if that), is more or less equivalent to a 200V cap at 1.5uF (that has maybe 0.4uF remaining at 100V).

You can at least get larger values in higher voltages -- probably, the thicker dielectric reduces pinhole defects, allowing ultimate breakdown to be approached more closely -- but this just means even steeper C(V) reduction, it sucks.

C0G, at least, perform very well -- they have modest density (33nF 630V is abundant in 1210), and the energy density at rating is excellent, better than electrolytic IIRC.  Downside: they're still rather expensive in quantity.  Need 10s of uF at 630V? Eek!

Leadframes at least deal with cracking, but they're also ludicrously expensive, at least what I've seen in small quantities from major distributors.  AFAIK, these are being used in automotive among other applications, so there must be either sufficient advantage in using them (certainly a help in a high vibration, long life application), or someone getting good deals on them -- factory direct, or custom runs even, perhaps.

Tim

I thought it was 1210, but wasn't sure if I was just being overly conservative, didn't want my example to get TOO contrived.

My favorite ceramic capacitor "trick" was for ceramic only decoupling on something that really shouldn't short out with a single failure, was using pairs of caps in series. With two caps in series the nominal voltage was so far up the C(V) curve, capacitance of a pair in series was higher than that of a single cap, which was much further down the curve.