A big, 1200V/400A IGBT transistor, Semikron SKM400GAR12T4, failed in a circuit. It was decided to dissect it if parts of it can be salvaged for another project (large switches are typically assembled from multiple discrete dies), and where the failure happened.
The failed IGBTs shown a dead short between gate and emitter.
Large switches are often composed from discrete dies with relatively accessible interconnects. These interconnects can be severed and undamaged chips used as standalone, or bonding wires to the damaged chips can be removed.
For example, the solid state relays can be composed of two antiparallel thyristors. When they fail short, sometimes only one thyristor fails. After disconnecting it, the part can still be used in less critical duty or an experiment as a single optocoupled thyristor.
The salvaged chips aren't fit for high-reliability duty. They could have suffered damage that degrades their parameters and/or their reliability can be negatively affected. But the salvages are usually for free, which is a cost-benefit ratio that's hard to beat, and enable uses in experiments where the costly parts wouldn't be affordable.
The "brick" of the IGBT was removed from the device and inspected. Its bottom side with the heat spreader was cleaned with turpentine to remove the heat transfer paste.
IGBT brick | IGBT brick | IGBT brick | IGBT brick |
IGBT brick | IGBT brick | Bottom side, cleaned | Bottom side, cleaned |
The brick, made of a thermoplastic polymer, was cut with a heated blade around its perimeter, at a guessed height. The cut was cleaned with a knife, removing the extruded material.
Brick cut | Brick cut | Brick cut | Brick cut, cleaned |
Brick cut, cleaned | Brick cut, cleaned | Brick cut, cleaned | Brick cut, cleaned |
The contact tabs were bent straight, so the two halves of the casing could be separated with a screwdriver and the top removed.
Brick tabs bent | Brick partially separated | Brick partially separated | Brick partially separated |
The inside of the brick casing shows minor signs of the explosion above the destroyed chip.
Brick casing top | Brick casing top, inside, witness marks | Brick casing top, inside, witness marks | Brick casing top, inside, witness marks |
Inside the brick there was a spacer, a plastic frame that keeps things in place. The electronics is potted in a silicone gel that looks and feels and behaves like something fresh from the nose of a god of adhesives. Its behavior is well-visible in the removal pictures. The gel is sticky and very viscous. The spacer was pried apart, then time was given so the gel can flow apart.
Inside, with spacer | Inside, with spacer | Inside, with spacer | Inside, with spacer |
Inside, with spacer | Inside, with spacer | Inside, with spacer |
Spacer removal | Spacer removal | Spacer removal | Spacer removal |
Spacer removal | Spacer removal | Spacer | Spacer |
The IGBT consists of several independent chips, bonded to the heat spreader and to the wiring.
There are four independent IGBT chips with freewheeling diode chips right next to them. There are four more diode chips in the other half of the package.
The transistor-diode pairs and the separate diodes are arranged in pairs on ceramic substrates (alumina? aluminium nitride?) with copper traces, and bonded to the metal spreader that forms the base of the casing.
The gates are connected together with a fairly thin copper busbar, with a wire attached to a spade connector. The transistors' emitters are connected with another wire to the other spade connector. These match connectors on the gate driver board.
The exposed dies show the extent of the damage. One transistor die is clearly destroyed, with char and pieces of die and bonding wires embedded in the gel. This matches the usual appearance of this kind of IGBT post-failure. The index of refraction of the gel makes photographing difficult where its surface is not straight.
Two other transistor dies show signs of thermal damage - spots of surface discoloration.
One transistor die looks undamaged.
The thermally damaged dies and the exploded die show a gate-emitter short. The undamaged-looking die tests okay. The diode dies seem to test okay.
overall | overall | standalone diodes | exploded chip |
overall | overall | overall | overall, closer |
overall | transistor-diode pairs | diodes | diodes |
exploded chip, fragments cleaned | exploded chip, fragments cleaned |
Three of the transistor dies were damaged beyond salvage, with gate-emitter short. Fourth die, the one without signs of thermal damage, tested correct. Therefore one 100-amp/1200V IGBT was recovered, with its freewheeling diode (and possibly with its neighbour's diode).
All four diodes seem to be correct. The diodes on the side of the two damaged transistors will be tested if they can be recovered/reused.
The standalone diodes at the second half seem to be intact. Therefore either a 400-amp diode or (if the large leads are cut to half) a pair of 200-amp diodes can be recovered.
A second IGBT failed at reduced current. Paradoxically, the damage is bigger here.
IGBT brick, cut cleaned | IGBT brick, cut cleaned | IGBT brick, cut cleaned | IGBT brick, cut and tabs bent |
The inside of the housing shows clear witness marks of the explosion above the exploded standalone diode - char marks and traces of the sticky silicone compound.
IGBT housing, explosion marks | IGBT housing, explosion marks |
The silicone layer caused problems with photographing. Water was used to cover the uneven layer and create a smooth surface to mitigate the index-of-refraction deformations. The method is not perfect as the refractive index of the gel and of water don't match, but is a great improvement in comparison with gel-air only.
The IGBT is contacted with massive leads, with the chip pairs connected with halved leads. A Dremel wheel was used to cut the leads to halves, to separate the transistor-diode assembly into more assemblies, at least some of which could be reused.
The grinding produced a copious amount of copper dust, which adhered tightly to the surface of the sticky potting snot, and was difficult to remove.
One 100-amp and one 200-amp diode (a 100-amp die pair) were salvaged. All transistor dies were damaged beyond repair. Attempt to salvage their diodes was not done yet.
IGBT contact cutting | IGBT contact cutting | IGBT contact cutting | IGBT contact cutting |
IGBT contact cutting | IGBT contact cutting |