Failure Modes of MLCCs

In the nineteen seventies I worked for GEC Telecoms and we had a number of failures of Multilayer Ceramic Capacitors (MLCCs) where the capacitors would burst into flame like fireworks. You'd wonder where the combustible material came from! The jet of flame could persist for up to a minute and was hot enough to set fire to a plastic enclosure.

In those days we were using leaded capacitors (wires not SMDs) so the physical stress on them was minimal. (But read on because the problem exists for SMDs too.)

We were able to reproduce the failure mode by connecting a 100nF capacitor across a 5v DC supply then applying a very short duration 2000 volt pulse. The MLCC would burst into flame and flare like a roman candle. We never determined exactly what was happening in the actual equipment or why but it could have been associated with fast risetime pulses from the power supply. The manufacturers introduced a test that was supposed to weed out potentially faulty devices.

Later, I worked for Siemens and we had a number of 19" rack PCBs burn out because 47nF MLCs caught fire. These were connected across 5 volt rails.

As part of the investigation, I made up PCBs with solder pads of various widths. MLCs were soldered across these and the PCBs were subjected to a bowing force that put the MLCs under tension. The force was increased until cracking noises were heard. Measurements indicated a decrease in capacitance and an increase in leakage resistance.

I discovered that failure could be prevented by keeping the width of the PCB solder pads to no more than 60% of the width of the MLC terminations. Basically, there was no need for solder pads. Just solder directly to the copper tracks. No further failures occurred during stress tests. I would assume that the strain was being lost in the tin-lead solder itself, which was undergoing stretching. In the limit, the solder joint would probably crack but I didn't see that happen. In any case, unnecessary bulking of solder on the termination to the PCB should be avoided. By minimising the amount of solder on each joint, you will improve reliability and achieve a significant cost saving by using less solder.

Of course modern "lead-free" solder is less elastic so the problem might be worse and the solution might not be as simple. However, at the time of my investigation, "lead-free" solder was not in use so it was never tested.

As far as existing designs are concerned, there's no need to redesign the solder pads. All that's needed is to modify the solder mask so that the exposed copper pad area is narrowed to 60% maximum of the width of the SM capacitor. It's also wise to mount MLCCs away from points of maximum stress, such as the corner of a cut out in the Printed Circuit Board (PCB), and ensure that the PCB is well supported at pressure points (e.g. push-button switches) to eliminate any possibility of flexing. Where cost is of no consequence or reliability is paramount, use a rigid ceramic substrate.

Finally, I would like to mention my colleague Gresham R. T. Clacy who carried out research at GEC Aycliffe on solder wave problems and discovered that vibrating the solder at approximately 300 Hz minimised air inclusions, thus ensuring complete soldering of all joints. Edit to add that Gresham was responsible for the practical experiments and tests on MLCCs at GEC. I was personally responsible for the work at Siemens but with the assistance of Gerald Scarlett and the support of my then manager, Ivon York.

I am now (2025) retired but can still be contacted via my Satcure website while I remain alive.

Martin T. Pickering B.Eng.