SMPS capacitor failure mode [solved]

[mac.6]

Hi all,

I have two fpga dev board with Kintex 480T (quite expensive and not easily found) which have failure on 3.3v rail.
Rail is derived from 5v/5A input, with MPQ8616GL-12 IC.
Yes that's a 12A capable chip for something not exceeding 1A maybe (only jtag and IO buffer are powered by it).
Schematic is pretty much straightforward, with a bunch of 22uF cms capacitors on output and 1mH coil.

I can isolate the output from power rail, I can see that IC is still trying to pulse current but hits OCP, which is what I expect from it, so ouput is probably ok.
Inductor (1mH, current rating unknown) seems ok as there is pulse seen across it (DCR too low to be measured, even on known good inductors)
When probing output there is an almost short situation (<1ohm) which explain that.
The only component tied to ground are voltage divider tap (and is ok) and bank of capacitors.

Primary cause of failure is unknown, can be due to fpga jtag connector which is not keyed and inserted backward, resulting in 3.3v shorted to ground.
Note that other fpga power rails are also using the same IC and schematic, but are unaffected.

Have you seen this kind of failure mode with thin film (probably ceramic) capacitor shorted to ground?
Over voltage transient due to bad power brick on input or over current situation which exceed cap rating?

[Bashstreet]

Yes i have seen it in amplifier circuits it is not as such that uncommon i believe.

[Rasz]

smd caps can also fail mechanically, or due to liquid damage

[mac.6]

I think there is no mechanical damage in this case, both board failed the same way and on the same parts (there is 4 power rails, only 3.3v is killed).
No crack are visible, but since it's 0805 I do not expect to see them if there is any.

I have ordered a bunch of caps to replace them, so now I need to desolder them without killing the board.

If it don't work, I will piggy back a DCDC module and disable the original parts.

[Armadillo]

MLCC cap don't need to literally fails mechanically or visibly, but most of the time it just need to degrade enough to fail the circuit design and kill other components instead.
Degraded to around half capacity is my observation, most of the time.
Well, you can measure those when you took them out.   

[mac.6]

I have remove all the caps, and no luck, still got a ~0.5 ohm path to ground. So it's either the chip itself (probable as ic is slightly discolored) or pcb issue (I doubt).

I will remove the ic and use external DCDC 5->3.3v module to power the board.

[wraper]

Degraded to around half capacity is my observation, most of the time.

If you measure within circuit, it's actually fine in most cases. Type II and III ceramic capacitors drop their capacitance a lot with ageing and restore it with reheating above Curie temperature (while desoldering). Type III (Y5V, Z5U) capacitors also drop their capacitance with heating a lot. For example, just by heating it with fingers or high room temperature, it may drop 20-30% of capacitance.
MLCCs usually fail short or become leaky. Often this happens only above certain voltage or intermittently and they may appear fine when measured with multimeter/LCR meter. Usually this is because of not externally visible microcrack.

[Bashstreet]

Degraded to around half capacity is my observation, most of the time.

If you measure within circuit, it's actually fine in most cases. Type II and III ceramic capacitors drop their capacitance a lot with ageing and restore it with reheating above Curie temperature (while desoldering). Type III (Y5V, Z5U) capacitors also drop their capacitance with heating a lot. For example, just by heating it with fingers or high room temperature, it may drop 20-30% of capacitance.
MLCCs usually fail short or become leaky. Often this happens only above certain voltage or intermittently and they may appear fine when measured with multimeter/LCR meter. Usually this is because of not externally visible microcrack.

Far as i understand that is accurate.

[Armadillo]

I have remove all the caps, and no luck, still got a ~0.5 ohm path to ground. So it's either the chip itself (probable as ic is slightly discolored) or pcb issue (I doubt).

I will remove the ic and use external DCDC 5->3.3v module to power the board.

It would be interesting if you can measure the capacitor connected to Pin 7, its capacitance even after it's been heated out of the circuit and compare to its actual value.

[mac.6]

Will try on one board, but I doubt I have a good enough cap meter on hand (only the ubiquitous transistor/component tester).