No announcement yet.

OT: baffling mosfet failures

  • Filter
  • Time
  • Show
Clear All
new posts

  • OT: baffling mosfet failures

    Allright, this is highly off topic, but I don't subscribe to any electronics forums and there's quite a crowd of electronics guys here.
    I've made a small batch of DC fan controllers and some of them suffer from mosfet failure after tapping the powered board on a desk. The circuit looks like this:

    The 60V/20A mosfet NTD5867NL is controlled by a 5V microcontroller, the gate is protected by two schottky's and there's a big schottky freewheeling diode for the inductive kickback. Then there's a relay as safety circuit interrupter in case the mosfet fails shorted, the relay is switched directly by the on/off switch. The nominal fan current is about 4A.

    The following failure is occuring on some of the boards:
    When the board is powered with the mosfet in the off position, when the board is tapped on a desk nothing happens yet. However, when the power is removed and reinstated the mosfet fails shorted.
    I could imagine the mosfet is subjected to quite a large dV/dt when the relay closes on poweron, but I don't understand how the mosfet is damaged by tapping on the desk. The only mechanical part is the relay, but if it opens while the fan is off there shouldn't be any transients. Even is the mosfet is driving the fan I don't see a clear problem.
    The 1000u cap would (in retrospect) be better positioned on the other side of the relay from a dV/dt point of view, but it might burn up the relay contacts after too many power cycles then. (however it's not supposed to be switched on/off that often)

    So my best guess here is that the heavy dV/dt on poweron is sending a current spike through the drain-gate capacitance, and then through the BAT54 to the 5V rail (which is still rising from ground). Still don't understand why this would kill the mosfet.
    As a quick fix I'm going to try to use a protected mosfet NID5003N as it should be a bit more robust but I'd rather understand the real nature of the problem here....

    Does anyone have a better idea about the failure mechanism?

    I'll add a picture of the board to spice up all the boring text :-)

    Thanks for reading,
    Last edited by ikdor; 03-17-2013, 06:05 PM.

  • #2
    The 1000 uf cap can stay where it is, but there should be a .1 uf from x2 pin 1 to ground. You could also put a small value cap from the gate of the mosfet to ground, and/or a resistor in series with the gate lead. To alleviate voltage spikes across the mosfet, use a mov across it.

    I'm having a problem seeing how the mosfet can turn on with the blocking diode d4 in place. Ok, maybe I see it now- I thought the mosfet was triggered by the +5 coming to the d4, but it's getting the trigger directly to the gate. Chances are good that it's oscillating where it shouldn't be because the actual voltage on the gate is not under good control. I'm familiar with mosfets in general, but not that one in particular, so I don't know if the voltage that does reach the gate is enough to turn it on fully. If it doesn't turn on fully, it will dissipate too much VA product.

    There should also be a diode across the relay coil.
    Last edited by darryl; 03-17-2013, 11:43 PM.
    I seldom do anything within the scope of logical reason and calculated cost/benefit, etc- I'm following my passion-


    • #3
      So, physically tapping the board on a hard surface causes the failure?

      If your relay clapper opens during this tapping, it may briefly open the path to +14V for your fly back.

      Try patching the cathode of your fly back (D3) diode directly to +14V.


      • #4
        Put a snubber across the drain-source and a gate resistor to slow down turn on. The relay is opening for split second when shocked and even the inductance of the wires could be enough to cause a large voltage spike via the 1000 uf cap. The Mosfet is only rated to 60 volts so a snubber will limit the rise.

        I forgot to mention that the gate resistor should be as close to the physical gate lead as possible so it doesn't act as an antenna.
        Last edited by Evan; 03-17-2013, 08:50 PM.
        Free software for calculating bolt circles and similar: Click Here


        • #5
          The way the fan is wired, the transient should be snubbed by the diode without any otehr parts..... the fan is directly across the diode. But as jmarkwolf suggests, there should be some capacitance from the top of the diode to ground.

          if you actually connected the diode direct to +14 NOW, that would be worse than what you have, unless you also connect the fan there..

          Evan's snubber is a decent idea, BUT that depends on the switching frequency..... snubbers need to balance dissipation vs snubbed transient voltage, and here a diode + RC snubber may be better than just the RC alone.

          Evan's other point is that you do not appear to have a gate resistor anywhere near the device, looking at your PWB.... I see capacitors but no resistor.. You definitely should have one, to prevent oscillations.

          I have another possible issue..... your uP, the AVR, isn't the strongest drive in the world...... even though you have only 15nC of Qg, it is possible that your uP cannot handle driving that..... OR that it cannot deal with the dv/dt coming back from the mosfet during the board-banging test. Against that, I do not see K2 on the PWB, it on teh back?

          if the mosfet stays linear too long, it will overheat and fail. That can be due to oscillations, bad/weak drive, etc.

          Is the uP still good? If a large dV/dt comes back from the device it might kill the AVR port. (edit... that is even WITH the diode to supply..... i do see that.... but it allows the transient to go at least a little above the AVR 5V supply.....)

          You DO seem to have sufficient voltage drive to stay saturated even with a lot of drain current, so long as the uP can pull up hard and reach 5V. An external pullup for that is good, but conflicts with a safe power-down condition...... I prefer stuff to be "off" until power is stable.
          Last edited by J Tiers; 03-17-2013, 11:45 PM.

          Keep eye on ball.
          Hashim Khan


          • #6
            Is the relay mounted to the backside of the board, or off-board? If it's off-board, then my assertion that mechanical shock affecting the clapper is out the window.

            If it's off-board, then I would look for bad solder joints. "Making and breaking" of bad solder joints, as a result of the mechanical shock, may be the culprit.

            D3 is shown as a Schottky diode, not a zener.

            If it is indeed a Schottky diode it's function would be to direct any flyback energy, and it's connected incorrectly. The cathode should be connected directly to +14V. Otherwise its' connection to +14V is interrupted by the mechanical shock. Which apparently is the main failure mode here.

            If it were a zener, it's function would be transient suppression, and it is connected properly, but it will only protect for positive going transients. And I don't think it would be affected by the mecahnical shock.

            J Tiers point about the drive capacity of the AVR processor may be valid. I don't use those processors, but if the FET were running in linear mode all the time, it could shorten the life.
            Last edited by jmarkwolf; 03-18-2013, 08:53 AM.


            • #7
              Thanks for the ideas guys,

              The missing diode across the relay could be an issue, I forgot to put one in there. I put one across the other (not populated) relay as there I need protection for the drive transistor, but this one could make havoc with the rest of the circuit on power-off.
              I don't fully understand putting 100n from terminal 1 to ground. I imagine putting one across 1 and 2 to limit the voltage rise before the big diode starts conducting, or putting it between 2 and ground to limit the voltage rise rate on the drain. Could someone explain what it would do?

              Regarding the gate resistor I need to put in, on which side of the resistor should I then put the BAT54 diodes? (on JTiers remark on the BAT54's; I think they should protect the AVR as they are schottky's whereas the AVR body diodes are regular ones)

              Relay K2 is indeed on the back behind the mosfet.
              In normal operation the mosfet doesn't get warm, so I don't think it's staying linear too long. I would be very nervous to put a pull up on the gate.

              Just missed jmarkwolf's comment. The relay is on board and the solder joints look rather healthy. I reflowed all the SMDs on a hotplate and did the through holes by hand.
              Diode D3 is indeed a schottky, it's keeping all the inductive energy inside the fan by shorting the loop. The current decay is not very fast but the energy is dissipated in the diode and fan winding.

              Last edited by ikdor; 03-18-2013, 09:12 AM.


              • #8
                Originally posted by ikdor View Post

                Regarding the gate resistor I need to put in, on which side of the resistor should I then put the BAT54 diodes? (on JTiers remark on the BAT54's; I think they should protect the AVR as they are schottky's whereas the AVR body diodes are regular ones)
                Possibly...... it depends on what the limits are for that device. We have used AVRs and had failures unless there was a series resistor..... with no gate resistor, you don't have a series resistor, and the various path impedances will determine the voltage that gets to the AVR.

                of course, the mosfet will be "on" any time the voltage at that point gets to above around 3 V. You can refer to the gate voltage vs drain current curve to see how that will be.

                The question was more to be sure we knew what was failing.... it would be possible for the AVR output to be the problem.

                The gate resistor would go right at the mosfet leads, AFTER the schottky diodes. You can figure the resistance based on the switching frequency, the drive current/voltage available, and the "Qg", the total gate charge to switch the device on or off.

                Originally posted by ikdor View Post
                Relay K2 is indeed on the back behind the mosfet.
                In normal operation the mosfet doesn't get warm, so I don't think it's staying linear too long. I would be very nervous to put a pull up on the gate.

                Diode D3 is indeed a schottky, it's keeping all the inductive energy inside the fan by shorting the loop. The current decay is not very fast but the energy is dissipated in the diode and fan winding.

                I believe you are correct, and that re-connecting D3 will be much WORSE..... the way your schematic is labeled, the fan is directly across D3 as it should be. If re-connected, the fan inductance would no longer have a path to discharge its volt-seconds, since the "fan+" (which would be negative when mosfet turns off) would be open.

                If indeed your connections are all as shown, AND the problem happens ONLY with the board being shocked as you say, then there must be either

                a transient voltage that exceeds one or more of the voltage ratings.

                gate voltage..... can the gate voltage be exceeded in any way? We don't see the whole circuit, so the rest of the drive is unknown.... but the drive should hold the gate voltage down even with transients due to relay.

                Drain volts.... the question is whether there is any transient that exceeds the ratings for drain voltage AND IF there is, whether the inductance contains more energy than the repetitive avalanche energy rating of the device. Even if you DO have a transient, if the device dissipation isn't exceeded, and the avalanche rating at temp etc isn't exceeded, you are STILL OK. I don't *like* relying on that, but you "can" do so.

                it looks like the relay opens everything, and no connection other than parasitic capacitance exists to "fan+" with it open.... but what happens when it closes (chattering, no doubt) is that it provides a ground path again.

                I suppose it is certain that the fan is indeed across X2 pins 1 & 2, and not by chance on 4 or 5, which are direct to +14?

                A condition that allows a linear operation for long enough to destroy the device due to power dissipation. the case type is one with limited heat capacity, and little dissipation capability (although it looks like there is a decent area of foil) so if there is a high current, plus linear operation, that could be an issue.

                A condition that turns on a parasitic "feature" of the mosfet unexpectedly.... There are some parasitic parts in teh device that are supposed to be de-activated by the device design. On Semi is pretty good about that, and the technology is pretty well-known, but it can happen in some cases.

                yes, the pull-up is dicey......I would prefer to use a gate driver instead..... and that may be a good idea anyway.... There are some gate drive types that have a guaranteed hold-down even when not powered..... and some discrete designs which do that as well. But your problem seems to be with the device powered and the mosfet "on", correct?


                What frequency does the fan drive operate at? Just off-on? or is it PWM?

                If it is merely off-on, then you only need a solid gate drive, not one that the drain transient can over-power and drive the device linear. I am not sure the AVR meets that, it probably does not have a high source current to hold up the gate under transients.

                if you have PWM, then you need to be able to drive the gate with the peak charge and discharge currents AND hold it up. If you can drive it well you "probably" can hold it up, but that may not be true of the AVR..... You will also need enough voltage to drive the gate even with the gate resistor.

                The 15 nC Qg is 1/4 of the charge needed with a 600V 40A IGBT type I am using in a design now..... a potentially significant amount, depending on frequency.


                There is a chance that the EMI transient from the relay is sending the AVR temporarily "into the weeds" so that it is not correctly driving the mosfet, or that the switching frequency is suddenly very high due to "executing data".....

                I think you need to put probes on and do the chock test while monitoring the gate.
                Last edited by J Tiers; 03-18-2013, 09:52 AM.

                Keep eye on ball.
                Hashim Khan


                • #9
                  I don't have the experience level of Jtiers but what about static electricity? In one of my former lives, I remember mosfets disliking static. . .


                  • #10
                    Does your AVR reboot due to electrical noise when the 4 amp fan kicks off as the relay shakes as the board is smacked? I bet that makes an amazing amount of electrical noise.
                    Under normal operating conditions does the AVR boot and take the output outta tristate (too lazy to enlarge your pic, read the device, and look up the device and see if that's an output only port).

                    Theory: Normal conditions you power up and the poor fet has a tristate hi-z input on its gate, but that slow series relay isn't going to apply power until long after the uC pin outputs a reasonable low-z drive. Whack the relay on the table and the uC goes bonkers and reboots and for a little while outputs hi-z on the gate while passing 4 amps. Nice noisy signal on the gate (thank you relay!) spends some time in the linear area and kapow. I assume you can reproduce this easily? So cut the trace and force the FET gate "permanent" on with some resistors and then smack it around trying to blow the FET by bouncing the relay. Problem can't be a rebooting uC if the uC isn't even connected anymore. Next step with trace cut for a good laugh momentarily flick the gate a couple times to simulate a rebooting uC going hi-z output aka input during reboot and see if a hi-Z gate under load is survivable by the FET with your protection ckts as is. My guess is the FET will survive experiment #1 and blow up on experiment #2. Fuss with the protection/drive ckts until it survives experiment #2, simulation of a rebooting uC.

                    One way to survive experiment #2 is if you need a low-Z always output there are nifty dedicated FET drivers that don't blow up when unconnected input, or just plain ole transistor ckts, but I like to wedge an optoisolator into the ckt because you can isolate analog and digital ground with those bad boys and nothing you do on the power side (well, less than 2500 volts or so) can possibly get back into the uC. Ironically if you separate the analog and digital grounds using an opto you might find the uC power is now clean enough that the uC stops rebooting when the relay is whacked under power (or maybe not, uC can be finicky beasts). Finally optos are cool.

                    Worst case scenario output many 4-letter words temporarily and slap a COTS SSR in. Those guys have done all the work to make an indestructible solid state switch (well, at least some are indestructible...) and the prices have collapsed since I was a kid, back when those were exotic gadgets. Other than medium to high power RF I expect to live to see the end of widespread electromechanical relays in electronic gear. The trendlines don't have many decades to go before physical relays cost more than SSRs...

                    Look on the bright side, its just a fan and failing on probably isn't immediately health and safety critical. People have blown up car engines trying to make FET switched fuel injectors where they fail shorted and flood the engine (good outcome) or set it on fire (bad outcome). Like the olden days of carb fires in demolition derbys.


                    • #11
                      Thanks again for the extensive opinions.
                      1a There is nothing else on the gate except the uC pin. PWM switching frequency is quite low at 4kHz so the mosfet is in the linear region for a very small percentage of the time.
                      1b the fan is indeed across 1 and 2, 5 and 6 are a charger connection for the battery on 3 and 4

                      The uC IO pin output current varies with output voltage; a drive current of 50mA into 3V tapering of to zero at 5V. So not unthinkable that transients can mess with it a bit. The interesting part is the mosfet fails while the board is tapped if the mosfet is off. It dies shorted not immediately but on the next power-on cycle.
                      More interesting is that only some boards show the effect, others keep functioning just fine...
                      We haven't actually tested it with the mosfets on, this is not my day job so time is a bit limited with a family. I'll see if I can put a scope on it at the office and try to find some weird voltage flying around.
                      I don't think the AVR itself is going into the weeds. The output is driven by a hardware timer, so no software is directly involved with timing and the display keeps on moving as intended before and after the mosfet problem. When the mosfet is replaced, the board functions fine again but the question is of course, how long.

                      After thinking about the starting condition, there might be something to the lack of drive at power-on. There's a 65ms delay until the uC starts up and during that time the gate voltage is determined by the very sharp voltage rise on the drain when the relay closes, distributed across Cdrain-gate and Cgate-source. That might put the gate at an inconvenient voltage for that first 65ms.

                      Thanks for the ideas. The AVR does not seem to be rebooting but I have to double check with the other guy who is killing the boards.
                      Even though the SSR is a more bullet proof solution, the current BOM cost is at 11 euro for small series. Buying a lot of SSRs is going to hurt the wallet.

                      The use of optocouplers and SSRs feels like a failure to me. I just want to be able to understand why my circuit is killing the mosfet and learn something that allows me to make better designs. I shouldn't be that hard to make something that controls a 4A fan :-)

                      I would still be interested to hear what a 100n on connector terminal 1 would do exactly.



                      • #12
                        The idea of the 100n is to provide some sort of "tie-down" when the relay is open.

                        While the fan is across the diode and should not be giving a problem, there are stray capacitances from the fan motor windings to ground, which might possibly cause transient conditions that lead to higher voltages of a "common mode" nature at the mosfet drain, where BOTH terminals of the fan are at a higher voltage than expected. The 100n capacitor is likely to be larger than the stray capacitance, and would tend to "hold down" the voltage in such a case.

                        It is interesting that the shock does not cause immediate failure.... that tends to suggest the transients are not the "direct" issue causing failure.....

                        Just so I understand you.....

                        You said:
                        "The interesting part is the mosfet fails while the board is tapped if the mosfet is off. It dies shorted not immediately but on the next power-on cycle.
                        More interesting is that only some boards show the effect, others keep functioning just fine..."

                        So you tap the board, and shut off the unit... and then when you power it on again, the mosfet dies?

                        or do you mean that the mosfet dies when the turn-on signal appears without taking off power?

                        Because the AVR default is to have the output tri-stated, you may want to put a pull-DOWN on the gate..... that cures the tri-state problem fairly rapidly.

                        If the unit comes on tri-state, AND power is applied to the mosfet before the AVR is awake, the charge-up of drain capacitance can hold it in linear mode for a considerable time until the AVR asserts a definite state on the output, either on or off, no matter which.

                        I am not sure what effect the tapping of the board would have, because if the mosfet is in the OFF state, very little or no current will flow. However, if for some reason (schokky diodes do leak a bit, and they ARE diodes) the gate has a charge on it, AND the power is applied, that charge will act as any legitimate charge, and modulate the channel conductivity, probably to some state you do not want.

                        the effect of the tapping could be merely to charge the gate by rectification... or it could be a co-incidence. Maybe if you do the same things in the same order and timing but do NOT tap the board, the problem will still occur. have you tried that to be certain the tapping is a necessary condition?

                        Whatever the initial cause, I am becoming more convinced that the problem is in a linear mode dissipation failure..... unless you can show that the thing is stone cold after failure and not "on" when it fails.

                        In the present state of the data given, I don't see a way for the various voltages and avalanche energies to be exceeded.

                        I would immediately put on a pull-down, and see if that does the trick. Also a gate resistor, that still could contribute to a linear mode problem.....

                        Ckellog is of course perfectly correct about the static electricity. In this case, however, we hope the schottky diodes will hold it down so the effect is less marked. Static also usually causes an immediate failure, no delay or wait for a future turn-on.

                        But anyone who deals with IGBTs or Mosfets and does not use static protection is being silly...... it will kill the devices.

                        Keep eye on ball.
                        Hashim Khan


                        • #13
                          I would still put a snubber on the FET. The bit of circuit I posted is from my design for the PWM speed controller on my lathe axis drives. They have been working just fine for years now and I often hard switch the outputs from one motor to the other or reverse the motors without turning the drive off. The requirements for my circuit and the posted circuit are very similar. The circuit design is not.
                          Free software for calculating bolt circles and similar: Click Here


                          • #14
                            The fan is in a plastic housing without any ground in the vicinity, would it then still make sense to put the 100n there to tie the fan down?

                            The tapping might indeed be a coincidence, the devices could just as well fail from repeated power-on cycles. We haven't had time to ascertain whether the tapping is indeed the prerequisite for the failure.
                            The shock does not cause the immediate failure, the sequence is like this:
                            -board is powered but fan is off
                            -tap tap
                            -fan is still off
                            -board power is removed
                            -board power reinstated
                            -fan turns on because mosfet source and drain are shorted, gate voltage is 0
                            But of then this only happens some of the time and not on all boards......

                            Anyway, I'll fix the current boards with a pull down on the gate, a diode across the relay and change the mosfet to a protected version. Then we'll see if they stay alive after some abuse. I'll have to skip the gate drive resistor as it's a bit of a nuisance and I don't want to start cutting up tracks for these boards.
                            I will change the 65ms delay time to 0ms as well. The supply voltage is rising fast enough that I don't expect any problems with that and I don't need to wait for a crystal oscillator to stabilise either.

                            Evan: I hear you on the snubber, but it's a bulky fix and I don't want to put any more through hole components on there if I can help it. If I can fix it with a few cents of "bird seed" that would be preferable.

                            Thanks again,


                            • #15
                              It's often recommended to add a small cap across an electrolytic as a further means of keeping instabilities out of an electronic system. Because the electrolytic in your drawing is on the other side of a switch from the actual circuit, it's completely isolated from the circuit when the switch is open. There is good reason to not put an electrolytic after the switch, as you mentioned early on, but the circuit itself may need some device 'hard wired' across its ground and supply points to help prevent instability. That was the basis for my suggestion of the cap from pin 1 to ground.

                              Many circuits recommend a tantalum cap across the electrolytic, or in this case since the electrolytic can be disconnected from the actual circuitry, the cap would be placed on the circuitry side. Values suggested are usually .1 for a bypass cap of some sort, 1 to 10 uf for a tantalum, and I've seen suggestions to use all three for best results. Much is going to depend on what instabilities may exist- devices capable of slewing at high rates of speed and operating from voltages and not current (like mosfets) are very susceptible to instability. The gate resistor is a very common addition to help deal with this.

                              It seems impractical for you to add this resistor at this point, so I might suggest instead that you consider placing a small inductor bead right on the gate lead- use a dab of contact cement or whatever. It won't have nearly as much effect as if the gate lead went through it, but it might make a difference. If it doesn't it's of no consequence.

                              The addition of such low values of capacitance after the switch is not going to have the negative effects that a larger electrolytic might have on the switch or other parts. And just in case there's a possibility of a reverse voltage transient being developed across the mosfet because of an 'open' power connection, you could add an ordinary diode in blocking mode across these caps. An ordinary rectifier such as a 1n4000 series part will have some capacitance as well, and you might consider the rectifier as a substitute for the .1uf cap- though it doesn't have nearly that much capacitance. Under normal circumstances, you would not choose a fast diode for this purpose.

                              I would put the cap in the circuit anyway- just consider it good practice to have it in the circuit. All three parts I'm suggesting would be wired in parallel, then tacked onto the circuit board, with the ground connection being as close to the mosfet source lead as practical.

                              I think you should add the snubber circuit as well, as Evan suggested.

                              One more thing- the layout of your drawing shows a distance between the fan motor pins on X2 and the diode there. That diode should ideally be as close to the motor as practical. I know the drawing doesn't show where the parts actually are physically, but the idea there is to shorten the loop in that circuit. If the diode is snubbing a voltage transient, that means that a fast current is flowing through the motor and the diode through the wiring, which makes it a transmitting antenna. The diode should be right at the source of the transient, where the leads enter the motor basically, just like a diode across a relay coil is usually placed very close to the coil lead pins.
                              Last edited by darryl; 03-19-2013, 07:50 AM.
                              I seldom do anything within the scope of logical reason and calculated cost/benefit, etc- I'm following my passion-