What abuse is required to make bulk 100uF 6.3V MLCCs leaky over 10 years?

[incf]

I've been warned once or twice that using 100uF 6.3V MLCC (edit: 1206 package) for bulk decoupling of 3.3V battery powered products (<10uA idle current) is possibly a mistake due to the long term (10 year) potential for leakage. (in this context they replaced tantalum and other capacitors of questionable spec).

I'd like to know what mistakes can I make to maximize the chance of running into leakage? What are the mechanisms by which these capacitors slowly and silently degrade?

I speculate that mechanical stress maybe could cause cracking, and then moisture gets into the dielectric?

Regards

[1] I have searched for other threads, there are a lot about metrology, fewer about real designs with big MLCCs. Seems simple enough to be worth asking afresh from this angle.

[mjs]

I'll add one more warning. High value MLCCs are problematic in battery powered micropower applications. The dielectric layers get incredibly thin and are prone to cracking. I've seen 10 μF 0402 caps fail in months, starting to consume 0.4 mA @ 3.6V.

NASA has some good stuff on this (download while it lasts): https://ntrs.nasa.gov/api/citations/20240003605/downloads/Teverovsky-20240003605.pdf

I've done the mitigate the issue
- Smaller values in larger packages with higher voltages (my own thinking to reduce effects of cracks - not from literature)
- Flexible termination caps
- Keep caps away from where there might be thermal (manual/selective soldering areas) or mechanical (borders, close to mounting points)

I don't guarantee anything, but I've seen cap failures with those at around 2E9 device-hours (operating hours*number of devices) at close to room temperatures.

[Geoff-AU]

big MLCCs are very vulnerable to cracking.  whether from mechanical (board flex) or thermal cycling (different coefficients of expansion for the substrate, terminals, and PCB).  I wouldn't call a 10uF 0402 physically large though, but that is a lot of capacitance in a tiny package so it's possibly challenging the material's limits.

I know tantalum has its horror stories of ancient times and people who are still licking wounds but it seems like a reasonable option for high capacitance, low voltage, high longevity, and potentially low leakage.  There's also tantalum-polymer.  On the downside, tantalum is a conflict mineral so it's quite justifiable to reject them on ethical grounds.

- Smaller values in larger packages with higher voltages (my own thinking to reduce effects of cracks - not from literature)

A curiosity I just found while looking at some tantalum datasheets is they rate leakage in "CV" units (capacitance in microfarads times rated voltage).  0.1CV and 0.01CV are two values I saw, but you'd have to have a closer look at the test conditions for the leakage spec.

[thm_w]

I know tantalum has its horror stories of ancient times and people who are still licking wounds but it seems like a reasonable option for high capacitance, low voltage, high longevity, and potentially low leakage.  There's also tantalum-polymer.  On the downside, tantalum is a conflict mineral so it's quite justifiable to reject them on ethical grounds.

Agree, modern tantalum are very reliable.

A curiosity I just found while looking at some tantalum datasheets is they rate leakage in "CV" units (capacitance in microfarads times rated voltage).  0.1CV and 0.01CV are two values I saw, but you'd have to have a closer look at the test conditions for the leakage spec.

T498 series has uA leakage listed: https://content.kemet.com/datasheets/TAN_ENG_KIT_19.pdf
6uA for 6.3V 100uF. They are expensive and out of stock though, maybe something similar exists.

I'm not sure if OP wanted solutions or wanted explanations.
There are other things to consider like MLCC degradation, a 100uF 6.3V MLCC might be effectively the same as a 10uF tantalum if its sitting at 5V and aged.

eg this example is down to 33uF at 5V and 25C: https://ds.murata.com/simsurfing/mlcc.html?partnumbers=%5B%22GRM31CR60J107MEA8%22%5D&oripartnumbers=%5B%22GRM31CR60J107MEA8K%22%5D&rgear=suaykx&rgearinfo=com&videoId=5970368749001

[coppercone2]

the nature of the bond and the flexibility of the PCB makes MLCC inherently vulnerable. The two solutions are the rubber elastomer MLCC (that means there is rubber in your circuit as a conductor) and probobly thicker boards. I strongly suspect if you did the same test with a much thicker PCB it would work better.

I wonder if the board resonate with electrostriction on caps and magnetostriction on magnetics to cause damage


BTW people talking about aging ceramic.. I dunno. I think thats pretty clear mechanical damage. How much do you trust damaged parts? To me aging of parts means something like, diffusion in a semiconductor that changes its behaviors, or strain equalization on a resistor, or a reference settling down, tube surface emissivity decreasing. I don't know if its fair to use the same word for cracked up caps

I want to say that maybe the board you are using is not up to par for using that part if you are anticipating cracking as part of the circuit

[wraper]

PCB layout means a lot. Component orientation in particular if it's located where PCB may bend, like near mounting holes and PCB edges. Anyway, making PCB slots in precision voltage reference style to separate MLCC from deformation introduced into PCB should allow using such parts with high reliability.

[Psi]

It's not so much large capacitance but the density of the capacitance they are trying to fit in such as small package.
eg, 100uF in 0805 might be problematic but 100uF in 1210 might be fine.

Use digikey search to see what the max capacitances are for the package size and voltage you want and then ignore the highest value or highest 2 values.  eg, currently for 6.3V in 0805 the biggest is 100uF, then 47uF, then 33uF.   So 100uF is risky, 47uF is probably ok, 33uF is definitely fine.

Placing two in series works great to protect you from one cracking but you then have to use 2x the capacitance since you are putting them in series. But if doing this place one 90deg rotated from the other so cracking/bending force is not on them in the same way. Also consider that if one shorts you circuit now has double the capacitance which is usually ok but not always.

And of course consider where on the PCB you are putting it, you want to avoid the edges of the pcb and avoid anywhere near screw holes.

[jbb]

I’m no expert, but I understand that hand-soldering larger ceramics (1206 and up???) can be a reliability problem because it very rapidly heats up one side of the capacitor when you touch the iron on. The thermal shock (? or gradient ?) can apparently crack the ceramic layers.

I’d also like to point out that - assuming you have soldered the PCB in a panel - the areas near to the PCB breakouts (whether mouse bites or V scoring) can get some extra flexing (and thus be more dangerous to capacitors). Especially if you use the traditional ‘bend by hand until it snaps off’ method.

If you’re doing mass production (or need high reliability) you can get machines to break through the mouse bites or something which looks reminiscent of a pizza cutter to part along the V scores.

Finally, beware large capacitances in small packages. That 10uF might go down to 3uF once the operating DC voltage is applied.

[coppercone2]

What I think the problem is with hand soldering them is lack of reflow. I think if they are crooked they get stress concentrations, so you want to flux them well and then use a hot air gun to let it align itself. The material itself should have terrific thermal shock resistance because its ceramic, but if its not re-melted after hand soldering, you will have it at some slight angle with no self leveling at all, the number one place where SMT seems to break is when its used creatively like for coaxial terminators where its not even on a board


I am some what suspicious of the data, because we see alot of broken ceramics, but we also see things like PCB mounted into the plastic weird molded enclosure with clips, flimsy enclosures, etc. On alot of my 'observations' I can't even guess at the cause of failure because there is so many dodgy design decisions