[SOLVED] Mysterious MOSFET failures

I'm building a 2000W sensored brushless DC motor driver. I'm using IR2110 drivers with FDP2532 FETs. The drivers are fed 6 synchronized PWM signals at 72khz. High-side signals have dead-time controlled by the microcontroller. High-side FETs repeatedly fail despite the protections I have implemented. Ive been stuck on this for a week, and it's time to turn to the community for help.

THE SETUP:
An adjustable CC/CV lab power supply feeds juice to the prototype controller. FETs in the controller are mounted to a ridiculously over-sized heatsink. The controller activates the motors winding sequentially with an 80% duty cycle. The motor is under no external load, and draws about .

THE FAILURE:
The motor speeds up as I turn the voltage up on the power supply, and failure typically occurs around 25 volts. Sometimes the motor starts cogging a second before failure, sometimes there is no warning at all, and the motor jolts to a stops, and the power supply goes into current limiting mode.

DIAGNOSTICS:
Each time the failure has occurred, my DMM invariably shows one of the high-side FETs has failed short, and Gate-Drain resistance is a couple ohms. All 5 other FETs are fine, as are the drivers. Each high side FET has failed at least twice so far. Each time a failure occurs, the motor is drawing less than 2A, and the FETs are cool to the touch.

At first, I figured voltage spikes caused by the motor winding were killing the FETs, so I added high-speed diodes in reverse across each FET to clamp the winding voltage. Failure still occurred. Next I figured ringing/noise in the supply rail was killing the FETs. I added polypropylene and electrolytic capacitors, which made the rail voltage very clean. Still no change. Next, I guessed induced voltage spikes, or ringing in the MOSFET gates must be the problem. The gates are rated for +/- 20V, so I added 15v TVS diodes on all 6 gates. Again, no change. Next, I guessed ringing between the motor winding wires was the problem. I added an RC snubber, and 30V TVS diodes between the windings, and still I cannot crank the supply past 25v without getting a failure within a few seconds.

SO LIKE, WTF?
WIth FETs cool to the touch, and a current-limited supply, They cannot be frying from overheat, or overcurrent. WIth TVS diodes on the gates, there cannot be overvoltage on them, and with high-speed diodes clamping the motor terminal, there can be no overvoltage there either. My O-Scope shows clean gate-drive signals with minimal ringing, clean rail voltage, and a slight ringing on the motor terminal in the 1.2mhz range (though this could be internal to the scope lead), but otherwise fine output signal. With my understanding, I see no theoretical way the FETs could fail. I can guess the fact that only high-side FETs fail must have something to do with the problem. I have also read something about "Dv/Dt failure" but don't know what characterizes it.

Thanks for any input!

The schematic shows one of the three identical bridges and FET drivers that make up the device. Sorry if it's a bit messy. I have not used this schematic software before.

Can you put two probes on output with very short gnds in differential mode to get a flat line the move Ch1 to Vg highside to get Vgs. I suspect large negative spike on gate from gate draing feedback 250pf? and motor inductance 250uH?

Test at 20V

Then add 0.05 Ohm drain shunt and measure current using same method.

Remove probe tip and use tip & barrel between 2 short wire test pins for textbook waveforms. Make sure probes are balanced and calibrated for square response and good for 100MHz.

How have you selected the bootstrap capacitor size and type?
Are you getting a large enough voltage across it (i.e. could the charge time be too long for this value)?

I generally use 10u tantalum caps for my MOSFET drivers. I certainly could have used less, but this was what I had on hand. I see no negative impacts, as my prob shows a very good gate drive signal on all six FETs, High and Low. Clean, with steep rise and fall, minimal ringing, and at the proper voltage (11-12v).

- - - Updated - - -

Which outputs do you mean to put the probes on? DO you mean to put one probe on J3, and another on Q2's gate? If there was a spike there, negative or positive, why wouldn't the (D4) TVS clamp it?

My scope only as a 72mhz sample-rate. My probes probably arent close to 100mhz. That's what you get for $100.

Ill put a shunt resistor on the drain when I get back to the lab. I have some .01 Ohm shunts.

Can you show a photo of power stage and driver wiring to give as a feeling of possible layout related problems?

Is there a DC path to 0v thru the motor?

10uF is a bit high for the boostrap caps, try 1uF MLCC close to IC.

There is no dv/dt rating for these parts, if you turn on the bottm fet V fast & hard and the top fet is partially on (or off even with weak pull down on the gate drive) you can turn the top fet on briefly for every turn on of the bottom device, most of the power in this shoot thru will go into the top fet (as the bottom one is hard on) - this may be happening as the applied dv/dt to the top fet goes up with Vbus - solution - buffer the drive to the top fets (or all of them) a simple 2 xtor emitter follower npn on top will do - or a dedicated mosfet drive chip - I'm betting this will cure your ills.

Also check there really is measurable dead time, easy to do with a 100MHz scope on the mid point - when the next fet turns on there will be a noticeable change in slope on the dv/dt of the midpoint voltage, if there isn't and the dv/dt is very high it implies a fet is turning on at the same time its mate is turning off (no deadtime).

My PWM signals from the microcontroller have a decent amount of deadtime, but it is possible that with the addition of turn on/off delays within the FET driver, and the fet it's self, there may be some shoot through. Come to think of it, the problems worsened since I reduced the deadtime thinking it would reduce inductive spikes during the period when both FETs are off. Ill post some scope screenshots of the gate signals, and winding outputs tomorrow.

"if you turn on the bottm fet V fast & hard and the top fet is partially on (or off even with weak pull down on the gate drive) you can turn the top fet on briefly for every turn on of the bottom device, most of the power in this shoot thru will go into the top fet (as the bottom one is hard on) - this may be happening as the applied dv/dt to the top fet goes up with Vbus - solution - buffer the drive to the top fets (or all of them) a simple 2 xtor emitter follower npn on top will do - or a dedicated mosfet drive chip - I'm betting this will cure your ills."

What do you mean by buffer the drive, and use a dedicated chip? I was under the impression that the IR2110 I'm using is already a dedicated FET driver which buffers the outputs. Is the top FET's turn-on due to capacitive coupling between the gate, and drain, as the drain voltage changes quickly when the bottom FET turns on hard? That seems to be a possible failure point.

I'll post photos of the setup when I get in the lab tomorrow. I have reduced the length of wiring between the FETs and drivers. The signal and power wiring to the IR2110s is long enough to make me worry about noise, but I added decoupling caps, and those helped some other problems I had. The power rail caps, TVS diodes, and clamping diodes are all as close as possible to the power rail and FETs. The wiring to the motor is quite long, and makes me worry about ringing between the winding, and wire capacitance, but the scope shows nothing serious.

- - - Updated - - -

Oh, and the motor is wired in a delta configuration. There is no path to 0v, except of course back out through the phase wires when their low FETs are on.

Hi,

you use external diodes UF4007. I don´t think this is a good choice.
The FETs have internal body diodes. They are solid to carry the current switched by the FETs. But they are slow. They have a relatively large recovery time and need a lot of charge to make them high ohmic.

To prevent from cross conducting one uses external fast diodes. The external diodes should prevent from getting the internal diodes conductive.
Therfore the externals need to have a lower forward voltage than the internal diodes. But the UF4007 don´t.
They have 1.7V at 1A while the internals have 1.25V at 33A. So the externals will never carry any current. They are useless.

Use schottky instead. If there is a problem to find one for the high voltage look for silicon diodes with lower forward voltage.


Alternatively/additionlally you could increase deadtime of the FETs.
Use diodes across the gate resistors with kathode to driver. This makes the FET to switch OFF faster.

Or use driver with increased deadtime.

Klaus

Alternatively/additionlally you could increase deadtime of the FETs.

Click to expand...

I think, it's the other way around. The regular way to prevent body diode conduction is to keep deadtime low and perform strictly synchronous switching.

I agree that paralleled silicon diodes are mostly useless, doesn't matter if slow or fast. For low voltage inverters, schottky diodes can be helpful. It should be also noticed that in the low voltage range, MOSFETs with fast body diodes are widely available.

The risky point is substrate diode conduction followed by too fast switch-on of the complementary bridge transistor which can cause latch-up of a parasitic SCR structure inside the FET. Check reverse recovery dV/dt rating in datasheet, if it's given as a small number (e.g. a few V/ns), you should take care to slow down switch-on, or avoid diode conduction, if ever possible.

Hi,

Yes one could imagine to decrease deadtime until the diode is never conductive. But this is going to extremes. If deadtime is too short, then both FETS are conductive and a heavy crosscondictive current flows.
I bit increased deadtime will cause the voltage rise above the diode forward current. (this dV/dt is very fast. It is caused by load inductance, wire inductance and stray inductance. On the other hand dV/dt is slowed down by stray- diode- and FET capacitance.)
Especially when high current is switched off, i expect that only a few ns of deadtime is enough for the voltage to rise and the diode getting conductive. Once conductive it takes time (105ns) and charge (327nC) to get it high ohmic again.
The given dISD/dT of 100A/us (datsheet) is the test condition. I don´t see it as a"limit" but a parameter for the semiconductor test equipment.

But yes. The longer the dead time the longer are the diodes conductive and the more power is dissipated in them.
So deadtime needs to be small, but not too small.

************
Back to the problem:

a scope measurement could show what happens.

At least three channels are necessary:
* Low side gate signel
* High side gate signal (High voltage!!!)
* Bridge output (high voltage)

Both rising and falling edges are of interest.

Klaus

Thanks everyone for all this useful information.

In regards to the parallel diodes... As I understand it, the MOSFET body diode is slow to recover, so switching an inductive load could cause a dangerous negative voltage when switching an inductive load if both FETs are off. My thinking goes that it shouldnt matter that the UF4007 has a higher forward voltage if the diode in the fet hasn't switched on yet. As long as the forward voltage isnt high enough to fry the fet, the UF4007 will take the current for the few nanoseconds required for the FET diode to turn on and take the load with it's lower forward voltage.

In my older Class-D Amp designs I did not utilize deadtime at all, and I noticed when no load is connected, the driver drew significant idle current as the voltage went up, but it never fried as a result. It just got a little warm, and wasted power. In those designs, I was dealing with voltages in the range of 56v, and had no problems aside from excess heat. In this motor controller, I notice no current draw from the power rail, and no noticeable heating of the fets, even without a heatsink. Could potential shoot-through really be quick and powerful enough to fry the fet without making it warm, or drawing measurable current?

WaveTheory said:

I'm building a 2000W sensored brushless DC motor driver. I'm using IR2110 drivers with FDP2532 FETs. The drivers are fed 6 synchronized PWM signals at 72khz. High-side signals have dead-time controlled by the microcontroller.

Click to expand...

i cannot believe i am the first to notice , 72khz for a brushless motor driver at 2000W ! WTF is wrong with you man ?!

did you consider the dynamic loss in the mosfet will dominate the static loss(due to rds) in this kind of setup and why on earth would you need such frequency 16khz to 32khz is more than enough

and as far as your problem , i think it is not caused by the motor or di/dt dv/dt ! i am certian it is simply deadtime , did you calculate how the rise and fall time of mosfet gate , due to gate charge ?? are you sure your dead time is enough

i also propose a test that you can make , remove the motor and instead wire a delta bridge using resistors , say 100ohm or 1kohm (to eliminate inductive spikes and overloads) and check :shock::shock::shock:
please post your calculations and i will be glad to discuss and help as far as i can

good luck

72khz may seem extreme for a motor. Ill soon be trying for 100khz. This is one of the reasons I opted for building a custom controller. There are many reasons I did this.

Firstly, higher frequency means less ripple in the power rail, and that means my electrolytic capacitors run cooler. They get mad hot below 30khz due to increased ripple amplitude. Increasing the frequency even higher opens the possibility of eliminating electrolytic caps all-together, and replacing them with mylar or ceramic. In theory, this would allow a well-designed controller to have an almost indefinite operating life.

Second, it reduces eddy-current in the motor core.

Third, it reduces the size of inductive and capacitive filtering elements on signal and power lines.

The efficiency gains in the capacitors and motor core may not entirely make up for the higher switching losses, but as switches and drivers become more advanced their losses will become less pronounced. In the present, this design saves cost at the expense of increased switching loss. In the near future, it's a free lunch. I see a future with switching speeds in the hundreds of kHz make electrolytic capacitors as obsolete as selenium rectifiers, and make tantalum capacitors overkill. It will reduce the size of inductors in filters, and power converters, and make 300k hour service lives seem common. When that future comes, Ill be ready for it!

Usually you notice some heat, but yes, quite narrow shoot thru effects can kill fets in the blink of an eye...

The 2110 is not that flash and thermally limited above about 70kHz depending on V rail and loading,

a buffer on this part is ALWAYS a good idea... esp to achieve a low Z pull down at high frequencies (i.e. to stop shoot thu induced from dv/dt..)

- - - Updated - - -

For slow body diodes in the mosfets (what you have) really the only cure is to turn the mosfets on slowly (Ron = 220 ohm or higher) with a back diode, usually schottky across this R to give a fast turn off (low turn off losses) this also builds in some guaranteed dead time, and reduces RFI.
If you do it and the failures stop or reduce you at least know where the problem is...

WaveTheory said:

72khz may seem extreme for a motor. Ill soon be trying for 100khz. This is one of the reasons I opted for building a custom controller. There are many reasons I did this.

Firstly, higher frequency means less ripple in the power rail, and that means my electrolytic capacitors run cooler. They get mad hot below 30khz due to increased ripple amplitude. Increasing the frequency even higher opens the possibility of eliminating electrolytic caps all-together, and replacing them with mylar or ceramic. In theory, this would allow a well-designed controller to have an almost indefinite operating life.

Second, it reduces eddy-current in the motor core.

Third, it reduces the size of inductive and capacitive filtering elements on signal and power lines.

The efficiency gains in the capacitors and motor core may not entirely make up for the higher switching losses, but as switches and drivers become more advanced their losses will become less pronounced. In the present, this design saves cost at the expense of increased switching loss. In the near future, it's a free lunch. I see a future with switching speeds in the hundreds of kHz make electrolytic capacitors as obsolete as selenium rectifiers, and make tantalum capacitors overkill. It will reduce the size of inductors in filters, and power converters, and make 300k hour service lives seem common. When that future comes, Ill be ready for it!

Click to expand...

have u considered the possibility of using FILM capacitors , they might solve all the problems of electrolytic without having to go up in frequency .
anyway is your decesion . by the way you didnot say :
did u do the test to find ur problem , did you do the calculations for dead time i told you about ?

Yes. I investigated my PWM signals with the scope. I previously had my dead-time set to about 4 clock cycles (72mhz clk), and the FDP2532 datasheet has a total on-off delay of about 250us, which translates to about 20 clock cycles. I can now spin the motor up to 35v, under load, without failure. It wont go much faster without crashing the MCU due to noise on the hall-effect sensor lines (but that's a different matter). It's not guaranteed, since I only tested for about 30min, but it's a huge improvement. I'd say the problem likely was the deadtime.

I had thought that too-little deadtime would cause enough shoot-through current to make the FETs hot, as this was my experience in the past. I hadn't imagined that it could kill them so suddenly while they were cold, and with a small enough current as not to register on my power supply. I'm now the wiser for it thanks to everyone's help.

As for the capacitors, film capacitors and ceramic capacitors cannot achieve nearly the same density of farads as electrolytics. Using the minimum number of film caps, especially at lower frequencies, would take up more space then the whole rest of the device. Tantalum capacitors are an option, but they are far too expensive to practically replace electrolytics in power levels this high. They also have a danger of thermal runaway.

Wouldnt having a gate resistor of 220R slow down switching enough to get the fets inefficient, and hot? I thought the whole point of MOSFETS is to switch them as fast as possible. I have often seen diodes in reverse across the gate resistor. I don't understand this. I thought the point of the gate resistor is to limit the gate-charge current to something the IR2110 can handle (2A). If you have a diode there, what keeps the IR2110 from frying once it pulls the gate low, and the current bypasses the gate resistor?

- - - Updated - - -

Can someone provide me with an example of a circuit having a buffer on the output of the IR2110? I always see the IR2110 connected straight to the gate of the FET through the gate resistor.

WaveTheory said:

Yes. I investigated my PWM signals with the scope. I previously had my dead-time set to about 4 clock cycles (72mhz clk), and the FDP2532 datasheet has a total on-off delay of about 250us, which translates to about 20 clock cycles. I can now spin the motor up to 35v, under load, without failure. It wont go much faster without crashing the MCU due to noise on the hall-effect sensor lines (but that's a different matter). It's not guaranteed, since I only tested for about 30min, but it's a huge improvement. I'd say the problem likely was the deadtime.

I had thought that too-little deadtime would cause enough shoot-through current to make the FETs hot, as this was my experience in the past. I hadn't imagined that it could kill them so suddenly while they were cold, and with a small enough current as not to register on my power supply. I'm now the wiser for it thanks to everyone's help.

As for the capacitors, film capacitors and ceramic capacitors cannot achieve nearly the same density of farads as electrolytics. Using the minimum number of film caps, especially at lower frequencies, would take up more space then the whole rest of the device. Tantalum capacitors are an option, but they are far too expensive to practically replace electrolytics in power levels this high. They also have a danger of thermal runaway.

Click to expand...

first of all i am happy for you and your sucess with the circuit
second , i must tell you that your point on film is worng !!
film are more expensive and have lower capacitance , but the have higher current capability and higher voltage ratings
so if you need 2000uF electrolytic with 5A ripple current , you can replace by 100uF film with 10A current ripple
did u get the point ?? its not about the micofarads , it about the current ripple

third : the 220ohm resistor is used to slow down the turn on of the mosfet and the antiparallel diode is used to ensure fast turn off . because mosfets usually turn on fater than they are turned off , so this balances the effect
and 220ohm is very high !! unless you are paralleling mosfets then is is divived by this number .
hope this helps
good luck

2A output curent isn't listed under maximum ratings, it's the specified minimal short circuits current, typical current is 2.5 A. It's also noted that it's allowed to flow for 10 µs as a single pulse. This clarifies that the driver has no problems to operate against capacitive loads that are effectively shorting it for a few 10 ns.

The maximum rating of practical interest is the allowed poster dissipation. Adding a gate resistor will beside other effects reduce the driver power dissipation, even in "constant current" region of output characteristic.