Capacitors explained by James Lewis of Kemet

[apelly]

Just stumbled on this interesting video via the embedded muse:

Regarding capacitors, James Lewis (https://www.baldengineer.com/) gave a presentation at Hardware Developers Didactic Galactic (HDDG) back in February 2016 on the various capacitor technologies (aluminum, ceramic, tantalum).  He brings his experiences from working at Kemet to explain each technology, their unique properties, and evidence against some of the myths around capacitor selection.  His slides are available at https://www.baldengineer.com/slides/hddg11 and a full video of the presentation is at https://www.youtube.com/watch?v=ZAbOHFYRFGg.

[kg4arn]

Enjoyed it. Thanks for sharing.

[ggchab]

Very  interesting. Thank you.

[Electro Detective]

LOL I had blurry mag glasses on and misread the title as "Capacitors explained by Jerry Lewis.." 

Thanks for the video link 

[David Hess]

I kept a copy of this video for the insight into solid tantalum capacitor failures.  It does not cover all of them (something else is going on also) but the connection between thermal cycling and stress fractures is important.

[T3sl4co1l]

Good video, and good products. It's the first time I know ksim, and I would certainly take a look into it. Sometimes I need DC bias derating curve, so my usually place to go is TDK as they have the online characteristics database. I will check out ksim as a second source of a lot of capacitor.
BTW, a question, do you guys know of any other MLCC companies that offer similar curves for all individual part numbers? A generalized curve for a product line won't work for me.

Murata also has data, but I think it's not as complete as TDK's and Kemet's.

Samsung capacitors usually have the curve in the datasheet, but you sometimes have to dig around to find the right datasheet, or look up the part (or a similar) part on their website.

Tim

[RGB255_0_0]

Nice video. And Mike did a presentation also that was played straight after from the same channel

[basinstreetdesign]

LOL I had blurry mag glasses on and misread the title as "Capacitors explained by Jerry Lewis.." 

Yeah, I heard he got tired of comedy - went into engineering... 

[AF6LJ]

Good Stuff Thanks, I enjoyed it.

[MagicSmoker]

Yeah, excellent video and K-SIM is especially enlightening w/r/t capacitance change vs. DC bias. One thing I found especially surprising is that going for a higher voltage rating but keeping the capacitance, dielectric tempco and package the same does not improve the change in capacitance vs. DC bias; in fact, it usually makes it worse!

For example, operating a 10V/0805/X7R/1uF cap at 5V results in a capacitance change of -2.96%, but applying 5V to a 25V cap that is otherwise the same results in a -4.13% change in capacitance, and a -11.83% change for a 50V part! Like a lot of circuit designers I try to keep the number of unique components to a minimum so I tend to always use, say, 50V rated capacitors anywhere the operating voltage is 25V or less, but the data on K-SIM has me rethinking that strategy, and though I don't usually find I need to change the part per se - after all, it rarely matters if the actual value of a bypass/decoupling capacitor is 1uF or 0.47uF - I am, at least, giving it more thought.

[Seekonk]

Capacitors 99,  Not quite 101. 

[T3sl4co1l]

Yeah, excellent video and K-SIM is especially enlightening w/r/t capacitance change vs. DC bias. One thing I found especially surprising is that going for a higher voltage rating but keeping the capacitance, dielectric tempco and package the same does not improve the change in capacitance vs. DC bias; in fact, it usually makes it worse!

Well, why would it not?

There are practical limits to how much energy density they can extract from a given part.

Obviously, they don't need to pursue that limit on smaller values in larger packages (like 0.1uF 50V in 0805), so it doesn't say much about smaller parts.  But for large values, you can guess very quickly what size you will need: 2uF at 50V will be 1210 and up, even if you find a smaller one (1206 or 0805) that claims "50V" rating and 10uF (at zero bias) or 22 or even 47uF.

For smaller values, what matters is, since they're using fewer electrodes in the stack, did they space them out accordingly, utilizing most of the ceramic -- and thus getting you a 100 or even 200V (at -30% C) part, or did they use the same, very thin layers, and just fewer of them (getting you the same "50V" curve, with C simply scaled by ratio)?

Also FYI, for a given dielectric, there is only one C(V) curve.  You can take that curve and scale it horizontally, to match the dielectric thickness (i.e., electric field strength), and vertically to match the number of electrodes and area (i.e., rated capacitance).

The curve can also be offset if a field is "frozen" in place (making an electret).  Some products are available which do this: the C(V) peak is centered at DC bus voltage, 500V say, for optimal performance in VFDs and solar inverters.  They're expensive, and can't be overheated (otherwise a new field is frozen in place -- probably zero, and you lose the benefit).

Tim

[MagicSmoker]

Yeah, excellent video and K-SIM is especially enlightening w/r/t capacitance change vs. DC bias. One thing I found especially surprising is that going for a higher voltage rating but keeping the capacitance, dielectric tempco and package the same does not improve the change in capacitance vs. DC bias; in fact, it usually makes it worse!

Well, why would it not?

How about because a higher voltage capacitor of the same capacitance, footprint and dielectric must either be comprised of more layers of a thicker dielectric - that being the only volumetric degree of freedom left - or else operate the same number and thickness of layers at a higher field strength (which is really just the same as the manufacturer slapping a higher voltage rating on a lower voltage cap, if you think about it)?!

If the capacitor is thicker - which I assumed to be the most common case, apparently somewhat naively, as I'll show below - then the electric field strength (in, say, V/mm thickness) should be the same at a given percent bias. That is to say, applying 25V to a 50V part should result in the same decrease in capacitance as applying 5V to a 10V part (again, assuming the same dielectric, footprint and initial capacitance). Instead, K-SIM reports that capacitance decreases more with voltage rating even when holding the DC bias at the same absolute value (ie - 5V)!

More specifically, the Digikey page listing the exact 3 capacitors I looked at on K-SIM:

Kemet 0805 1uF/10% X7R

Shows that the 25V part is, indeed, thicker than the 10V one, but that the 25V and 50V parts are the same thickness. I would expect the 50V part to exhibit the same decrease in capacitance at the same DC bias as the 25V part, then, but what actually happens - again, according to K-SIM - is the decrease nearly triples, from -4.13% to -11.83%!?!

I can't imagine a plausible explanation for that behavior, really, and so my suspicion at this point is that the data in K-SIM is simply wrong, however I am always open to learning new and exciting things... just don't be an a-hole about it, 'k?

[T3sl4co1l]

How about because a higher voltage capacitor of the same capacitance, footprint and dielectric must either be comprised of more layers of a thicker dielectric - that being the only volumetric degree of freedom left - or else operate the same number and thickness of layers at a higher field strength (which is really just the same as the manufacturer slapping a higher voltage rating on a lower voltage cap, if you think about it)?!

Yeah, large value (for a given footprint) caps are thicker -- although no thicker than width it seems (which is probably a good idea for packaging and assembly reasons).

If we're talking big value caps, then the thickness will already be max, pinning that degree of freedom as well.  Smaller values don't seem to be much thinner, either, so that the degree of freedom in thickness is maybe 2:1 or 3:1.  So, whatever they do inside those chips is still largely up to them, in terms of number of electrodes and spacing.  (That is to say, a 1nF cap isn't 100 times thinner than a 0.1uF cap, they're definitely using less active material by then.  But how much less, isn't clear.)

That is to say, applying 25V to a 50V part should result in the same decrease in capacitance as applying 5V to a 10V part (again, assuming the same dielectric, footprint and initial capacitance). Instead, K-SIM reports that capacitance decreases more with voltage rating even when holding the DC bias at the same absolute value (ie - 5V)!

Right, if they used a proportionate dielectric thickness, you'd expect this.

But it's also clear that some parts are identical, merely rated differently.  (There may still be some value in a higher voltage /spec/, perhaps a higher ultimate breakdown voltage, but no more C(V).)

Anything inbetween is also possible.  It's up to the manufacturer.

Shows that the 25V part is, indeed, thicker than the 10V one, but that the 25V and 50V parts are the same thickness. I would expect the 50V part to exhibit the same decrease in capacitance at the same DC bias as the 25V part, then, but what actually happens - again, according to K-SIM - is the decrease nearly triples, from -4.13% to -11.83%!?!

Yeah, probably they pack the 25V part little tighter, but not so much tighter than the -30% point is fully half the 50V one.

I can't imagine a plausible explanation for that behavior, really, and so my suspicion at this point is that the data in K-SIM is simply wrong, however I am always open to learning new and exciting things... just don't be an a-hole about it, 'k?

Just facts and speculation here.  If you took it personally, I apologize, I guess

Tim

[MagicSmoker]

I can't imagine a plausible explanation for that behavior, really, and so my suspicion at this point is that the data in K-SIM is simply wrong, however I am always open to learning new and exciting things... just don't be an a-hole about it, 'k?

Just facts and speculation here.  If you took it personally, I apologize, I guess  

Nah, no offense taken on my part, I just got the distinct impression that you were being a bit too dismissive of what I thought was quite unexpected behavior as reported by K-SIM, and your subsequent attempts at explaning said behavior seem to have led you to also conclude that this is, indeed, strange stuff. Here are the explanations either posited by you or already considered by me:

1. Electret polarization: that might work, but it costs precious time (seconds? minutes?) on the production line to apply the polarization and you still end up with a polarized capacitor, obviously, which is pretty unusual for an MLCC, to say the least.

2. A higher-k dielectric with wider electrode spacing: this would explain the worse DC bias coefficient, but how are the tempco and the other dielectric-specific parameters the same?

3. The same part but with a different rating stamped on it: this fails because the 50V part exhibits a worse shift in capacitance at the same DC bias voltage as the 25V part, when it should be the same, and that result just screams the 50V part is either operating at a higher field strength or using a higher-k dielectric, but, see above.

It really is quite a puzzle.

[T3sl4co1l]

I don't think they're using different dielectric materials -- tempco and age would be affected then.  Material should be pretty consistent as long as it says "X7R" or what have you.

(There are some quirky effects at very low voltage ratings.  That is, for very thin dielectric layers, where the electric domain size is comparable to dielectric thickness, k goes down.  This, incidentally, limits the feature size of FeRAM devices, because each capacitor has to be some 200nm across, or whatever, for the ferroelectric effect to manifest.  I don't think you can buy MLCCs /this/ low in voltage, though, so we don't have to worry about this.)

It would be interesting if they specified how much of the internal volume is occupied by electrodes and active dielectric.  Alas, they don't, so we'd have to do a lot of cut-and-polish to figure that out...

I don't see it as a puzzle, it's just hidden information: they don't tell you how much of the chip they're using, so it's anyone's guess.

They do seem to land between two "reasonable" limits:

The lower limit assumes the same dielectric thickness, across all parts in a given series and given voltage rating.
 E.g., all 25V rated caps in a series of parts use the same spacing between electrodes.  Filler makes up the rest of the height of the chip (since the finished chip needs a minimum height).

The upper limit is a chip stacked to maximum height, made only of active dielectric (i.e., it's completely full of electrodes).

Parts on the upper limit are the easiest to say something about, because they're pushing the maximum energy density the material can offer.  For these, you can say: when you need some amount of CV or CV^2, you need a part that's so-and-so large.

Between these two cases, the manufacturer has a degree of freedom: for values below the upper limit, they could use thicker dielectric.  This would lead to higher effective voltage ratings (defining the effective voltage rating as, say, the -30% C(V) point), because more internal volume is active (surrounded by electrodes).

I'm assuming the lower limit is actually a limit; there's nothing hard and physical about it, but it would be very disingenuous of them to sell capacitors with even less active dielectric than you expect from the ratings.

Then, what about different voltage ratings?  They could use proportional thickness dielectric, but this seems not to be the case, because higher voltage parts have worse C(V/Vrated).  It's also not the same thickness, because the C(V) curve is better... just not proportionally better.

So the conclusion is, since we don't have any of this internal design information, and we'd have to examine a ton of parts to see what a given manufacturer is actually doing -- we can't go this deep, and must fall back to the one thing the manufacturer does provide (...when they do..), which is the C(V) curves themselves.

Between the C0805C105K3RAC and C0805C105K5RAC, it sure as hell looks like they screwed up big time (violating my supposed lower limit)...
Ya know... they could've transposed the curves.  Got 'em backwards. 

If I compare C0805C475K3PAC and C0805C475K4PAC, I see the '3' (16V) has -30% at 7.3V, while the '4' (25V) has -30% at 8V.  This implies the 25V part has 10% thicker dielectric, but not 56% thicker (despite the 56% higher voltage rating).

Tim

[yada]

Aren't electrolytic polarized by the first time you apply voltage to it? You could actually wire them in backwards as long as the first current and subsequent current are always the same polarity? Or do they charge them up at the factory?

[T3sl4co1l]

Aren't electrolytic polarized by the first time you apply voltage to it? You could actually wire them in backwards as long as the first current and subsequent current are always the same polarity? Or do they charge them up at the factory?

The foils are anodized at the factory.

(Yes indeed, the same process that allows metal to be dyed and tough is also used to create capacitors!)

Tim

[SeanB]

I would guess they stick to a predetermined thickness layer, as this can be controlled in manufacture to get it all right. then for higher values they change the number of layers and the final top and bottom layer thickness ( probably also made up of the same thin sheets, but not metallised, so they can have less issue with cracking and size shift in firing) are adjusted for the values. higher voltages they might alternate active and non active sheets, or stack them to increase the dielectric thickness for the higher voltage, and this will cause the difference.

Just thinking, because once I have a machine that can make a thin slurry that I can control the thickness to under 1% I would be really unwilling to do anything to change that process, rather use those thin sheets in various ways in process, rather than keeping 2 or more sheet thicknesses in stock.  That would be at least reproducable in the plant, even if it means every incoming batch has to be exactly the same as the last, down to particle size, temperature and humidity before going into the final grinding and blending. I bet there are a lot of batches that get placed in a grinder, and are reduced back to powder so the metals can be extracted for recycling, because they failed some process test, or there was a machine fault somewhere ( heater died, controller wire broke so it was too hot or too cold somewhere, gas ran out) that turned it into expensive ceramic scrap.

Wonder if there is somebody working at one of these places that would send Dave a sheet of uncut fired capacitors in the mailbag, or even somebody working at a film capacitor place that would send him a failed reel of film capacitors.

[Seekonk]

This is actually a double post, but I thought it was important that it be seen by as many  people as possible.  It debunks the myth that only X2 capacitors should be used in capacitive dropping supplies. I contacted James Lewis, KEMET, about why the data sheet said not to use X2 capacitors in series with mains applications.  The following is his response:

"The actual issue is related to how typical X2 Capacitors lose capacitance over time. A typical film capacitor loses capacitance, over time, because of two effects: 1) corona and 2) corrosion. The voltage applied to most X2 capacitors tends to be in the 120-277 range. Corona occurs around 300 Vac and higher. So, while some capacitance might be lost because of corona effects, this isn’t the main concern. The real concern is that moisture will eventually egress into the package and oxidize the metallization. Some plastics, like polyester, more readily absorb moisture and accelerate this effect.

In typical X2 applications where the capacitor is across-the-line, the capacitance loss isn’t a concern. First, the capacitance value isn’t critical and second, most people don’t care if the end-product passes EMI tests 4-5 years into its life.

When used in a series-coupling or series supply application, however, the loss can be critical. Even a 10% loss in capacitance can result in not enough current for the supply.

This comes down to (and this applies across the industry), most X2s were not designed with “stable capacitance” in mind. So the R46, our most popular, can easily lose up to 30% of capacitance in a high humidity environment after only a couple of years. Which is why we made series with “better” epoxies, thicker metallization, and series-electrode constructions. Their capacitance loss will be much less than a typical X2, making them more suitable." 

[KD0CAC John]

I missed something , 2X not sure how to use for ID-ing capacitors ?
I did see that the average values were in a range of 120-277 , but is there more to these 2x to help ID ?

[Someone]

Yeah, excellent video and K-SIM is especially enlightening w/r/t capacitance change vs. DC bias. One thing I found especially surprising is that going for a higher voltage rating but keeping the capacitance, dielectric tempco and package the same does not improve the change in capacitance vs. DC bias; in fact, it usually makes it worse!

It can make it better or worse, I wouldn't say its usually worse. Here was a comparison from Muratas online tool of some 10uF 2012 (0805) parts:

But reliability is also increased by adding some derating to the voltage specification, so its not just about getting enough capacitance but also a suitable lifespan. Then you start looking deeper into the specifications for the conditions of the HALT (https://en.wikipedia.org/wiki/Highly_accelerated_life_test) testing done to the specific part numbers etc.