I am debugging a failed MOSFET (IRF6662), and in the process of doing so, am having difficulty understanding fully the thermal behaviour of the FETs. Two FETs used for DC switching both failed short independently (2 Ω short over S-D, and lower-than-usual ~1 kΩ resistances S-G and D-G). These failed when I turned the each FET on into a substantial load (34 V into 9 Ω an 1 mF of capacitance). These FETs are driven by a switch controller (LT4363) that has a soft-start feature, so the load voltage ramps up linearly over about 100 ms.

Schematic of relevant area, showing switch controller driving FET and gate network

I believe what happened was thermal damage due to the higher-than-normal power generation into the FET when the switch controller held the FET in ohmic regions before it switched on fully. I estimate that an average power of 20 W was dissipated in that 100 ms period, concentrated mostly into the middle half of that time. Using a thermal resistance junction-to-ambient of 5°C/W (from Fig3, the transient impedance plot for a 100 ms pulse), I get a Tj temperature rise of 10°C, well within acceptable bounds. So, if I have understood this right, it doesn't look like a simple case of overheating. But, does this thermal calculation make sense in this case?

Another possibility is thermal instability. Looking at the SOA of the FET, and estimating the I-V curve across the 100 ms turn-on period, I get this:

FET SOA graph, with estimated on->off I-V trajectory during 100 ms turn-on process

It looks like I am venturing out-of-limits for a period of time, crossing not the max power or max voltage limits, but a "thermal instability limit". But it's unclear if my transient violation of this limit is dangerous, or if this violation is likely to cause a FET failure like I have seen. In general, could such a transient incursion over the thermal instability line of a FET cause it to become damaged like this?

Lastly, these (transient exceeding max junction temperature and thermal instability violation) are the only two thermal damage methods I can see being relevant. I'm not driving the gate anywhere near fast enough for dV/dt problems, nor am I approaching the avalanche breakdown voltage. Are there any other damage mechanisms that might give these same shorting symptoms?

  • 1
    \$\begingroup\$ There's a thermal instability in the sharing of individual parts of the FET die, Vgs decreases as temp increases, resulting in runaway under linear operation. This is the opposite of what happens in hard switching, RDSon increases with temp, resulting in nice sharing, even across devices. You'll notice there's no DC line on that SOA, 10mS being the lowest. If you want long pulse / DC operation in the SOA, you need bipolars/darlingtons. Very few FETs are specified with a DC line in the SOA, and they're stupidly expensive. \$\endgroup\$ – Neil_UK May 21 '18 at 19:40
  • 1
    \$\begingroup\$ Non-trench FETs tend to be more rugged in these types of applications- For example, the following FET from TI (With a DC line in the SOA!) ti.com/lit/ds/symlink/csd19537q3.pdf ($0.44 in 1K quantities) \$\endgroup\$ – John D May 21 '18 at 20:15
  • \$\begingroup\$ What are your feedback resistors? \$\endgroup\$ – John Birckhead May 21 '18 at 21:38
  • \$\begingroup\$ I agree with @Neil_UK that thermal instability could be a problem, but I think you may be right about the transients. Two things you might look into: If you lose power to and have 1000 uf on the output, the output voltage might instantaneously remain above the maximum VGS. This will instantly kill the FET. Also, the charge pump on the device you are using only supplies tens of microamps. the impedances of your additional gate capacitances (1K and 0.1 uF) seem pretty low. \$\endgroup\$ – John Birckhead May 21 '18 at 22:02
  • \$\begingroup\$ @JohnBirckhead: Should the diode in the controller not guarantee that the gate does not deviate too far from the source? I'll calculate the gate drive capacitance - it seemed to work speedily enough, so I didn't worry about it much. Feedback is with a voltage divider that splits the source and runs into the FB pin of the controller. It's a binary on/off control. \$\endgroup\$ – alexandicity May 22 '18 at 10:40

I think your soft start may be your problem. Are the FET's on heatsinks? People usually don't do this because received wisdom is that a FET as a switch is not dissipating heat. But I had an experience (many years ago, also with an IRF series) that was quite educational.

Basically our system had IRF520's (from memory) switching up to 3A or so DC. They were "protected" by a 3.15A fuse and everything had been fine for years. One day (due to an unfortunate accident) we left one of the FET's passing 3A or so (in fact the cable resistance was just enough to stop the fuse blowing). When we came back there was a lot of smoke and only the fact that the containing equipment was flame-retardent prevented a very nasty fire. The board was an ugly charred mess.

I was tasked with the investigation and re-design. It turned out that thermal instability was the problem. Basically, Rds on those FET's rises with temperature. So when they are switching in conditions where they are just OK, thermally - quite soon they are not.

For the redesign, I took several remedial actions - "belt and braces" was in order:

  1. used FET's with even lower Rds
  2. mounted them on a big chunk of copper bar (heatsink)
  3. beefed up the PCB by using a 16/0.2 wire to get the DC off the board.

The result was heavily tested, did well, and is still in use (has been more almost 2 decades).

This is a bit different to your situation, but the fact that you are "soft starting" over 100ms, while the load has an initial current of almost 3.8A, and a time constant of 9ms (5CR is then 45ms) means your FET is getting quite some thermal stress. I don't fully follow your graph, but I wouldn't trust this situation. Switching FET's are not really designed to handle much thermal stress, the general idea is that they slam on and never have much volts across them. But the moment there are even a small number of W getting dissipated - the story can change.

I would consider getting rid of the soft start if your system allows it. (What is it really achieving anyway, if the load is a series CR? What is the initial resistance of the FET relative to 9 ohms?). Also beef up the thermal protection on the FET. Also you might look for any data on the thermal behaviour of Rds. (EDIT - I just looked and saw that indeed it rises quite sharply with T.)

  • \$\begingroup\$ Hi @dmb! Thanks for the comprehensive response. Always good when there's an illustrative story - with a lesson to be taken! :) Yes, it seems we think much the same for the cause. It is a switching FET that I'm not slamming, and the heat/instability is killing it. Adding thermal dissipation is tricky. I have already a PCB with as much copper as I can fit, and cannot add a top heatsink (space constraints!). Does reducing these external thermal resistances matter much in this case, with just a 100ms "pulse"? I would have intuitively thought the thermal capacitance would absorb it all... \$\endgroup\$ – alexandicity May 29 '18 at 10:51
  • \$\begingroup\$ Removing soft start is likely the way to go. It's not needed for these "test" loads (which are just CRs), but having a slowish monotonic rise is good when the load contains ICs and reduces inrush current. But - this can all be achieved in a 10ms or even a 1ms soft-start equally well. And it's an easy change to make. I just wanted to understand my problem first; if the cause is thermal due to lingering in the Ohmic region, then shortening soft-start is the solution. \$\endgroup\$ – alexandicity May 29 '18 at 10:54
  • \$\begingroup\$ Welcome. Yes get rid of soft start. Hard to predict when things are borderline stable (= occasionally unstable), you are in inherently unpredictable territory. Somewhere there is an edge that you mustn't go over, but you are groping a bit. All you can do is have the biggest margin you can. But basically the lower that FET's resistance gets, the quicker, the less heat it has to deal with. I would start there and then cycle it on and off, perhaps at 1Hz, fairly brutally, perhaps in raised ambient temperature . Heatsink is maybe not a big deal for you. (Our problem was steady state.) \$\endgroup\$ – danmcb May 29 '18 at 22:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.