MLCC short failure

[Siwastaja]

Hi,

Didn't think it would happen to me!

I have thought that MLCC's are very reliable (against electrical things), but do suffer from stress cracking if the PCB bends.

What came as surprise to me, is that the cracking actually results in a short instead of an open circuit, like discussed here: http://www.edn.com/electronics-blogs/tales-from-the-cube/4406033/Stressed-out-over-capacitor-failure

Also, it surprised me that it happens even with very little or no bending - the PCB is gently mounted from its sides in the grooves of a plastic box, and being a high-quality 4-layer board (without large copper cutouts on any layer), the board is very sturdy.

Please see the attached image. The design is a bidirectional synchronous buck converter running at 70 kHz, cycling and measuring lithium ion cells, transferring energy back to the input ("stationary") battery when discharging. The "input" is 10V, and output 0-4.5V 26A.

The failed capacitor is at the 10V input side, where one module of above mentioned specs share four 1000 uF, 16V low-esr Panasonic caps and ten 1210, 10 uF, 50V, X7R ceramics. I happened to have a supply of these (expensive!) ceramic caps, which is why the voltage is so much overspecified.

My analysis is that the problem is in the wide pads with multiple caps soldered there, and maybe with hand soldering with excess amount of solder. Somehow it might cause bending, or stress since first installed.

There is not much thermal expansion as the PCB is well cooled, but there can be some - the hottest parts run at maybe 30-40 deg C higher than ambient (25 deg C max), but nothing extreme. And the caps are about an inch away from the heat sources, cooled by air intake.

The system has had runtime of 1000 hours during calendar time of two to three months. The capacitor did not fail while running; the system was shut down for four days, the capacitor was biased at 3.5 volts the whole time (from a connected output side battery through a switching diode), and when I started precharging the input side from a current-limited, soft-start 5A, 10.2V power supply, the cap glowed red and smoked. The humidity has been higher than normal during this four-day period.

The cap measures 12 ohms now.

Any ideas or experience? Should I modify the design to use smaller 1206 footprint with separate pads, and is this enough? Any ideas on dealing with the issues with the PCB revision I have at hand, with those wide exposed planes, which is basically just an opening in solder-mask?

[daqq]

Parts like this do NOT like flooding the whole area with a big mass of tin - you are probably outside of the proposed thermal curve (both in time nad temperature) whilst doing this - this may crack the part, or change it somehow. Also, if you are soldering this by hand you should be aware that you are putting it through two different thermal cycles.

[Siwastaja]

http://www.digikey.com/Web%20Export/Supplier%20Content/UnitedChemiCon_565/PDF/UCC_MultilayerCeramicCapacitors.pdf?redirected=1

This document reminds of using proper amount of solder, and using pads that are not wider than the cap.

Can ripple resonance be a problem? Or any other electrical cause? I'm very interested whether this mechanical thing is the only real culprit there, or if there are other potential explanations.

[sacherjj]

I've never had to deal with this, as I make sure to stay at 0805 for smaller.  I remember reading something about soft termination 1206 or 1210 capacitors from Samsung I believe, to try to lessen this issue.  Yeah, here are some on Digikey:  http://www.digikey.com/en/ptm/s/samsung-electromechanics-america/soft-termination-mlcc?site=us&lang=en&ptm=30063&WT.z_ptm_structured=Buy%2520Now%2520Button

We have had many smaller MLCCs fail to short around a part that we needed to remove for programming once or twice on a prototype.  They do not like large solder areas or multiple temperature excusions.  I think you are correct in the hand soldering being the largest factor in early death.  I would use standard isolated pads for the caps.  You can flood nearby if you have good thermal isolation to the caps without issue.

[daqq]

Can ripple resonance be a problem? Or any other electrical cause? I'm very interested whether this mechanical thing is the only real culprit there, or if there are other potential explanations.

Well, I'm guessing that yeah, an overvoltage (by means of resonance, ESD, ripple, or any of the other common culprits) could cause a catastrophical failure like this - but ceramic caps are generally rated well under their breakdown voltage - I'm guessing that if you have a 50V rated cap you could go up quite a lot and still not have it die on you. Take one and measure the break down voltage.

[Dr. Frank]

Hi,
it's a well known effect, that MLCCs fail to short. Failures occur in our series production,  and field.
(We consume billions of MLCC every year)

Root cause is mechanical stress always. That is: PCB bending, bending at connectors or edge of PCB, wrong soldering profile or maybe (as in your case) thermal shock.
 
Electrical breakdown does not occur (so often), and may mainly be caused by faulty ceramic @ supplier.

Therefore, there were several rules:
1) Don't place MLCC near connectors
2) Don't place them at the edges of a PCB (bending on the manufacturing line, or in the housing)
3) Don't use extreme capacity-times-voltage types, as these are on the edge of technology (smallest layer thickness, smallest case size, etc.
4) Avoid too big /long case sizes
5) in critical locations of your schematics, like directly over power input, use either non-shorting types, or place two MLCCs in series, using two different suppliers.
6) The PCB expands differently than the MLCC, over temperature.
Therefore, use the correct PCB material, if using long case sizes, and also, if big temperature differences occur.
Thermal expansion module of PCB has to match the ceramic thermal expansion.
FR4 is ok, but not FR2 or CEM3.

In your case, I would criticise: excess solder, too long and too hot, already causing pre mature cracks during soldering process.
Use proper single (distinct) soldering pads  for each MLCC!
Use much less solder (and in turn: less time) for each junction!

The stress from 4) / 6) = different thermal expansion, bends the PCB, but is also covered  by the solder junction, before this force goes into the MLCC.
Therefore, if you apply too much solder, this junctions gets harder and stiffer, and the expansion force directly acts onto the MLCC.


Frank