MLCC X5R capacitor degradation

[BravoV]

If these are related ?

Click images to enlarge.



From Kemet.

[T3sl4co1l]

If these are related ?

Click images to enlarge.

This is familiar. I'm not aware of anything that drops precipitously like the earlier figure, which very much looks made up, but maybe some manufacturers made something that changes phase very slowly but also rather suddenly (days being slow compared to ~instant changes, and different from the log(time) aging of traditional materials).

A lot of damage was made a few years ago on blog posts, youtube videos and this very forum by "helpful" people teaching around that the issue is mainly with Y5V capacitors and X7R is fine. Some rules of thumbs very pulled out of thin air, such as "Y5V can drop below -80% under full DC bias but X7R only -30%" or "derating voltage helps". These are all totally false.

The best rule of thumb, by far, is simply: if the volumetric energy density is too good to be true, then it isn't. As ejeffrey writes above, if footprint size is a fixed parameter for you, then the height is relevant to look at. And you should really look at joules per volume. And calculate the joules at your DC bias voltage; the rating is meaningless. E / V = CU^2 / (width * length * thickness).

Yeah, while Y5V and Z5U (there's even a Z5V if you want to go looking for true awfulness) may be the worst, there are no exceptions, all type 2's exhibit the characteristic.

I occasionally need to use X5R for the greater density; a modest compromise while temp ratings are respected.  I haven't had to use anything worse, at least in quantity.  I have heard tale of Z5x being used -- responsibly engineered -- in extreme cost reduction applications.

The volumetric point extends with respect to voltage.  Like the 2220 I mentioned earlier, you can get a solid 10 or 22nF in X7R at 250V, and while you can buy 220nF or even more with such a rating, none of them offer much more capacitance at voltage.  The C(V)/C(0) ratio can get much worse than for low-voltage types, because the insulation quality is better -- that is, thicker layers are more reliable so can be pushed closer to limit.  (Hm, I haven't tested this, but I'm betting breakdown is closer to ratings -- whereas a 16V part might fail anywhere from 40 to 150V, I'd bet a 250V part might fail 400-600V?)  So, not even being able to get a proper 100nF at 250V+ even in large chips like 2220 is disappointing, but works out when you do the numbers -- they're bigger, sure, but they aren't many times larger than say a 1210.

Y1/Y2 ceramics are also commonly Y5V.  IIRC, these tend to drop off by a couple hundred volts, so are functional at mains voltage, but not under surge conditions -- which might well be a feature rather than a bug, the saturation reducing current flow through the capacitor.  The temperature characteristics seem the more important property here, something to keep in mind if you're doing EMI filters for wide temperature ranges.  (Or conversely, use then if you don't need to pass tests at temp extremes -- shitty, but may happen that that's what a project needs done.)

Tim

[iMo]

Hmm, back to the tantalum capacitors then??

[T3sl4co1l]

Nah, no need for conflict minerals plus thermite reaction potential when aluminum polymers exist.

Tim

[mawyatt]

As one who's lived thru the development of the modern ceramic and polymer capacitors, it's interesting to see the diversity of such. Long ago the scarcity/political control/cost of tantalum material became the driver for development of higher density ceramic caps to replace tantalum, later the same for the Polymer types which also help replace some electrolytics.

AVX, TDK, Kemet and a few others were the usual suppliers of quality ceramic caps, now you have Samsung, Vishay, Wurth, UCC, and a host of others, including many from China, it's difficult to know which supplier and ceramic capacitor type to use for a given application and why we invested in the means to measure the usual Frequency sweeps, but also Voltage sweeps and even some crude Temperature sweeps.

If one is designing a high quantity or mission critical product, as mentioned by others its wise to learn about the various nuances of ceramic capacitors!

Best,

[Conrad Hoffman]

Thought this was a pretty good summary:
https://blog.samtec.com/wp-content/uploads/2020/07/07_16_2020_geek_speek_MLCC_losses.pdf

[TimFox]

In my opinion, if the temperature, voltage, and time variation of capacitor parameters is important to your application, you should only buy from manufacturers (e.g., Kemet, AVX, Murata, etc.) that provide actual datasheets that show how these parameters vary with package size and dielectric.
MLCC capacitors from quality vendors need not be expensive.

[floobydust]

Thought this was a pretty good summary:
https://blog.samtec.com/wp-content/uploads/2020/07/07_16_2020_geek_speek_MLCC_losses.pdf

It is a very good summary. First part that upsets me:
"It was also shown that in addition to an immediate change of capacitance with the applied bias, there is also a longer exponential relaxation, which can change capacitance further by as much as 25% over the course of a few minutes."

Second, is that some MLCC cap manufacturers are -25% and others -70% with bias. Like I need to waste more time searching for and looking at curves for each specific cap and comparing, then purchasing wants to make a substitution and I have to repeat the whole ordeal.

[TimFox]

Thought this was a pretty good summary:
https://blog.samtec.com/wp-content/uploads/2020/07/07_16_2020_geek_speek_MLCC_losses.pdf

It is a very good summary. First part that upsets me:
"It was also shown that in addition to an immediate change of capacitance with the applied bias, there is also a longer exponential relaxation, which can change capacitance further by as much as 25% over the course of a few minutes."

Second, is that some MLCC cap manufacturers are -25% and others -70% with bias. Like I need to waste more time searching for and looking at curves for each specific cap and comparing, then purchasing wants to make a substitution and I have to repeat the whole ordeal.

Before I retired, some of my engineer co-workers would design in a particular component that had good performance in the circuit (such as a tantalum capacitor with better-than-normal ESR), but forget to put such specifications into the BOM.  Naturally, purchasing would order a cheaper part that met the specifications (usually tolerance and voltage rating) listed in the BOM.
For MLCC capacitor dielectrics, the physics of the situation depends on the voltage gradient in the dielectric, which naturally varies with construction and layer thickness.

[AnalogTodd]

I'd have thought you saw this already; you've been around a while!

I knew capacity rating was reduced by bias, but not that it was so extreme like -80% lol.
I'm no engineer neither I work at hw design, so yep there's still a lot to learn!

A lot of damage was made a few years ago on blog posts, youtube videos and this very forum by "helpful" people teaching around that the issue is mainly with Y5V capacitors and X7R is fine. Some rules of thumbs pulled out of thin air, such as "Y5V can drop below -80% under full DC bias but X7R only -30%" or "derating voltage helps". These are all totally false.

The best rule of thumb, by far, is simply: if the volumetric energy density is too good to be true, then it isn't. As ejeffrey writes above, if footprint size is a fixed parameter for you, then the height is relevant to look at. And you should really look at joules per volume. And calculate the joules at your DC bias voltage; the rating is meaningless. E / V = ½CU^2 / (width * length * thickness).

I know that while working at a major semiconductor manufacturer, we started looking into ceramic capacitors in the mid- to late-1990's. At the time we started doing the due diligence on MLCC's, we found that Y5V and Z5U were horrid for temperature and voltage characteristics, but X5R and X7R actually held up quite well. What came about was in the early 2000's, more and more people were switching to ceramics and there was demand for higher capacitance values in smaller case sizes (after all, look at the push on how much smaller products have gotten).  At first, very few people noticed that there was an issue, but we started getting phone calls about circuits not operating right and our investigations came to the conclusions on voltage coefficient fairly quickly. In 1998, we were including verbiage in the applications sections of our data sheets warning about Z5U and Y5V having issues. By 2003 or so, we were also discussing how it could happen with X5R and X7R and that care should be exercised when using same. We would update older data sheets from time to time and while doing so often updated the relevant text. I of course laughed when competitor data sheets came out a year after we included this info and copied our text word for word.

The relaxation in capacitance over time is one of the more recent concerns (again, push the laws of physics and they push back). It doesn't surprise me, manufacturers are trying to get more stuff into smaller packages but are now pushing against the limits of physics. It's like one of those stress balloons filled with goo--squeeze it down and one part will smush out from between your fingers in some location, try and tuck that back in and it squeezes out somewhere else.

[ejeffrey]

Second, is that some MLCC cap manufacturers are -25% and others -70% with bias. Like I need to waste more time searching for and looking at curves for each specific cap and comparing, then purchasing wants to make a substitution and I have to repeat the whole ordeal.

It's not per manufacturer.  They are all using the same materials after all.  Pretty much all manufacturers make capacitors that lose 50% of capacity at voltage and those that lose 85%.  It's all in the device thickness.  If you put a minimum thickness on your BOM and can convince your purchasing people to pay attention to it you will filter out 99% of the inappropriate substitutions.  You still have to check a given part, but you will drastically cut down on the number you have to reject.

The other option is to just use larger footprint in a low profile such that you are OK with a 80% drop.  This is the lazy approach but works if you can afford the PCB space.

[floobydust]

Huge variations between manufacturers, and also within their product lineup as well.

[T3sl4co1l]

About the only thing you can say, is there's a certain maximum value, in terms of somewhere between C*V and C*V^2, and that's for C under bias, in a given chip size.  You're guaranteed not to reach that maximum at, say, 10nF 50V 0805 -- which is still not to say you're guaranteed a flattish C(V) curve, they can always make things worse (i.e. use few, very thin layers, and add more as Cnom goes up) -- but it does mean, as you look at the largest values (say 10uF 6.3V 0805), you're more likely to be pushing that limit because there's just not much wiggle room left: they have to use the thinnest layers, as many as possible, to get the total value, and at a very low defect rate too (which puts pressure towards thicker i.e. more reliable layers).

And that doesn't mean much in terms of values -- you aren't going to avoid checking C(V) characteristics in any range of values, and indeed it's all the more important to do so when finding maximal values where you have to feel out where their C(V)*V^x (1<x<2?) limit lies.

The one upshot: I haven't seen any small-value parts that aren't widely spaced layers, i.e. they're more or less using the full volume of the chip and spacing out layers rather than using few closely-spaced layers.  (Which means a 10nF 0805 might as well be 200V or more, regardless of posted rating?)  That doesn't mean they can't, just that, of the ones I've seen characteristics of, they haven't.

Tim

[CosteC]

Thank you all for support.

Original capacitor is: KEMET C0805C475M3PACTU, so not bottom of the barrel.
Resoldering this capacitor restores vast majority of initial capacitance, all down to measurement accuracy.

I think this document is also revealing how againg accelerates.
https://www.vishay.com/docs/45263/timedepcapdrix7rmlccexptoconstdcbiasvolt.pdf

[PCB.Wiz]

Yes.  And it isn't solved by derating the voltage.  A 35 V and 16 V capacitor of the same dielectric and volume will usually have the same effective capacitance at 12V.  If a MLCC manufacturer doesn't provide capacitance vs voltage plots you should probably just ignore them. 

That's not quite true
This curve shows you can de-rate by both voltage and package-size to reduce the change at a given voltage, and X7R is better than X5R

https://www.eevblog.com/forum/projects/is-ceramic-capacitor-package-relevant/msg4480996/#msg4480996

[exe]

X7R is better than X5R

Not a capacitor expert, but my conclusion is that the distinction between X7R and X5R is somewhat... arbitrary. At the end, it's just a temparture rating, not a performance indicator. So, an X5R from one manufacturer can be better than X7R from another one. I think an MLCC with large capacitance in a small package sucks no matter what class it is.

I'd say the only thing that matters is curves from the datasheet. All that X5R/X7R ratings don't really tell the story. I'd also wouldn't trust the plot that you linked. It's only relevant to the parts that were evaluated. Different parts will behave differently, even if it's all X7R 0805 1u caps.

[tooki]

X7R is better than X5R

Not a capacitor expert, but my conclusion is that the distinction between X7R and X5R is somewhat... arbitrary. At the end, it's just a temparture rating, not a performance indicator. So, an X5R from one manufacturer can be better than X7R from another one.

No, it’s not just a temperature rating. Yes, the 5 or 7 means the temperature range. But the R means ±15% across the full temperature range, so that means that an X5R at 85C (its upper limit) will show a significantly larger tempco effect than the X7R, because 85C is far away from the X7R’s upper limit of 125C.

If your product has the caps operating at anything but ambient temperature, then the real world difference is insignificant. But if your device warms up, then for any application where the capacitance value is important, you want a higher maximum temperature, even if you won’t actually go anywhere near that maximum.

[T3sl4co1l]

Yep.  Like I said -- I don't mind using X5R.  Just don't use them beyond their ratings, go for X7R or better in that case.

The compositions are probably very similar -- you could imagine a continuously variable formula between them (and beyond), where k, breakdown, saturation, and temp range all vary together, more or less in proportion.  So you get a less dense but wider range X7R, or vice versa X5R, etc.  Those two cases merely happen to be the most popular, so it looks like a discrete (two valued) system, but that's a consequence of defined standards plus commercial preference, not a limitation of materials science.

I don't actually have any clue what the chemistry and variation are like; I would just assume they're variable in a more-or-less continuous manner over some range.  The characteristics of X5R and X7R are certainly similar enough to suspect as much.  And an extra 10% (in terms of absolute temperature: 85 vs. 125°C is about 40 out of 400K) is hardly anything in the grand scheme of things.  Really, these are quite precisely formulated!


The comparative plot doesn't really mean anything, because you can look at a dozen "10uF 16V X5R 1206" and get a dozen different curves.  Only the highest C(V) among them is representative of maximum possible density.  All the rest are either... being lazy in some way, or use an older (less precise? or potentially less cheat-ey..?) formulation.  And since there's no part numbers on there, the data are basically useless -- consider it reference material, not hard fact.

Tim

[Siwastaja]

Yes.  And it isn't solved by derating the voltage.  A 35 V and 16 V capacitor of the same dielectric and volume will usually have the same effective capacitance at 12V.  If a MLCC manufacturer doesn't provide capacitance vs voltage plots you should probably just ignore them. 

That's not quite true
This curve shows you can de-rate by both voltage and package-size to reduce the change at a given voltage, and X7R is better than X5R

https://www.eevblog.com/forum/projects/is-ceramic-capacitor-package-relevant/msg4480996/#msg4480996

Did you look what you replied to and did you look what you posted? The graph demonstrates exactly what ejeffrey said. For example, curves for 0805-X7R-16V and 0805-X7R-10V are completely similar. Same between 0805-X5R-16V and 0805-X5R-10V. Picking a 16V part and derating it to 10V yields exactly the same result as using the 10V part at 10V - probably because they are actually the same part.

[Kokoriantz]

I will be using 2//1uf50v 1206 X7R, as decoupling 36v. How I can determine the value decrease in time?
Should I replace by though hole polypropylene?

[CosteC]

I will be using 2//1uf50v 1206 X7R, as decoupling 36v. How I can determine the value decrease in time?
Should I replace by though hole polypropylene?

Without datasheet it is hard to tell. I would take ageing (3 decades) + DC bias + temperature worst case.

[Kokoriantz]

Decades on the table is in hours. For X7R it is 2%/10 hours. My 1uF will be 0.5uF in 250hrs ?

[exe]

Decades on the table is in hours. For X7R it is 2%/10 hours. My 1uF will be 0.5uF in 250hrs ?

I think decade means it loses 2% first 10 hours, then another 2% in 100 hours, then another 2% in 1000 hours, etc. I'd check in murata's simsurf for similar capacitors to see which curves they have. Or, better, just buy from vendor that provides such simulators

[TimFox]

I will be using 2//1uf50v 1206 X7R, as decoupling 36v. How I can determine the value decrease in time?
Should I replace by though hole polypropylene?

Even polyester (Mylar) TH caps will have no decrease in time, with roughly the same loss as X7R, in a smaller package than polypropylene.