Transient positive voltage spikes killing MOSFET drivers

[MarkM]

I see what you mean.  I finally had a chance to test myself.  The spike isn't random at all.  It happens every time the MOSFETs turn off.  I tested how the board was before any modifications and I saw a spike of 18.5v every pulse.  I added a 1k series gate resistor and the spike went down to about 13.5vabout 1v over battery voltage or so.  I used the board for awhile over and over and the MOSFETs didn't heat up at all.  I also added a 470uf electrolytic capacitor and saw no change at all, which was odd.

I know the proper thing to do is use a gate resistor for each MOSFET and to calculate the correct value.  Impossibe to add two resistors on this board revision, but I'll fix in the next one.  I'll also test a lower value resistor after calculating, but even limiting it to 10-12ma is working just fine @ 60hz.  No issues.  Side note, I'm trying to figure out how people are using low resistance for the series gate resistor in an acceptable sized package.  I'd like to go 0805 max.  Maybe 1206 would work, though.

I attached some pics showing the spikes after adding the gate resistor.  My scope sucks and is reading a bit higher than it should has well, so the spikes are probably only about 13.3v.

If you have any other ideas, please let me know.  The results so far are acceptable with the gate resistor to me.  If I get a chance, I'll show you guys the board in action.

[PSR B1257]

I also added a 470uf electrolytic capacitor and saw no change at all, which was odd.

The battery has a very low ESR and the wires to the board are also quite short, therefore the voltage drop at the input terminals of the board is quite low too.

Try this with wires ten times this long (or even longer), with and without a capacitor and you'll most likely see a difference.

I finally had a chance to test myself.  The spike isn't random at all.

There you go... 

Impossibe to add two resistors on this board revision

I wouldn't be too concerned about a single resistor for both MOSFETs, not to mention the resistor-diode combination for different charge and discharge times. For this application your one resistor is just fine.

[MarkM]

Thanks!  I'm trying to see if I'm calulating this correctly.  From what I calculated, the 1k resistor with this setup should be just fine.

10v worst case Vgs with 3s input using a driver with dead battery
1k ohm series gate resistor
2 MOSFETs in parallel with 25nC gate charge(rounded up to 30nC each)
Frequency = 100hz
Period = 10ms

10v/1k= 10ma of current

60nC/10ma = 6µS(0.006ms) rise/fall time.

Time the MOSFETs spend in linear region each period:
0.006ms * 2 / 10ms=0.12%

^rise + fall time divided by period. Is that correct? Even limiting it to 10ma seems perfectly OK to me.

[MarkM]

Well, the driver is still dying, but for some reason I didn't realize that it's not while it's running.  I've only killed a couple so far myself.  The drivers are actually being killed by power on transients. 

[krish2487]

The calculations look good.


You have started another thread on the PSU transient spikes on the driver.
I ll just post the reply to it here, because I have already suggested a solution for it on this thread.


Diode decoupling and putting a LC filter inbetween the battery and the driver Power pins.
If you use a slow diode like 1N400x then you can more effectively isolate the fast acting transients from propagating to the driver chip.
 

As I have mentioned earlier, the spikes occur during the turn off. The same property of an inductor (no instantaneous rise / fall of current) allows you to place a inductor(ferrite bead) in series with the battery and reduce the spike. 


If you are able to see the spikes on a oscilloscope, then you can see the difference it makes with and without a ferrite bead.

[T3sl4co1l]

Diode decoupling and putting a LC filter inbetween the battery and the driver Power pins.
If you use a slow diode like 1N400x then you can more effectively isolate the fast acting transients from propagating to the driver chip.

Uh...

[MarkM]

Hi, thanks again.  I was thinking my question was different enough to create another post, but you're right, it comes back to the same thing.  I'm researching now.

EDIT:I forgot to post.  My buddy was intentionally sending voltage spikes of 25v into the MOSFET driver and it wouldn't die.  I think the issue could be inrush current. 

[MarkM]

Unfortunately, I can't capture the power on spike with my crappy wannabe scope.  I did run some tests with my board, though.  I added in a rocker switch on the input from the battery pack and a 100µf electrolytic capacitor on the input from the battery as well.  I sat there and switched on and off over and over(even very quickly) for literally 20+ minutes.  The driver never died.  Keep in mind this is still the TC4427 on this board.

I took the big cap off the board and the driver smoked the first time I turned it on!  I put another driver on and forgot to add the cap for testing.  The driver smoked the second time I turned it on.  It's a bit of a bummer, as I don't have anymore, but I'm going to order more and really test it for a long time.  It's almost confirmed.  Needs more testing and to be scoped properly.

The other thing I did is I rigged the TC4422a to test since I have some.  At first I added it to a breadboard with the same value ceramic caps soldered to 2.54mm pin headers with the same rocker switch setup.  I ran the test over and over for awhile with the bulk cap.  It didn't die....  I took the big cap off and it still didn't die.  Bummer...  I didn't test as long as the first one, but I tested for a good 5 minutes over and over.

I ended up soldering the TC4422a with jumpers on my board, as the pinout is different from the TC4427 and test like that.  I still couldn't kill it without the cap for awhile, but I eventually did after 5 minutes or so.  I don't think the 4422a is quite as susceptible to the inrush current/power on spike.  There's no reason for me to use this driver, though.

I'll update this thread when I confirm this and hopefully mark it solved.  Thanks again for your help!   

[Zero999]

Unfortunately, I can't capture the power on spike with my crappy wannabe scope

I notice you're using Rigol 'scope. What model? Rigol is generally a good brand, especially for the price. Have you hacked it to achieve the full bandwidth?

[MarkM]

Those are pics from my buddy's scope that's helping me test. I'm not sure which model it is, but he said something about hacking it.  He captured the power up spike on his.  I have one of those little Hantek USB scopes. 

[Zero999]

Those are pics from my buddy's scope that's helping me test. I'm not sure which model it is, but he said something about hacking it.  He captured the power up spike on his.  I have one of those little Hantek USB scopes.

Those USB 'scopes are shite.

[MarkM]

Haha!  For sure.  I found a guy local with a Hantek DSO5102P for $200.  I might just go for it.  I need to be able to test myself. 

[Zarhi]

I'm switching two of these N channel MOSFETs in parallel with a TC4422a driver @ 60-100hz using an Atmega328 PWM pin on a custom PCB on battery power(2s-3s lipo pack).  I started with the TC4427 driver and they died randomly.  Now the TC4422a is dying randomly.

Same situation two years ago. LED pwm controller, PIC16F1509, 2xTC4427A, 4xIRF1010NS, 78L05. Mosfet drivers randomly dying at power on. Powered from 12V/7Ah lead-acid battery, small 12V 'wallmart' type adapter, ATX PSU, professional 12V/500W constant voltage LED power supply... no difference: tc4427 randomly dies at power on ( start heating itself ). I replace about 20 tc4427a, none of them survives more that 10-15 power on. Does not matter LED's are connected to mosfets, or not. Interesting was that if tc4427 survives power on, they work infinite time ( over an week ), including multiple cpu programming.

Series resistor between tc4427 and irf1010ns does not help. Decoupling also does not help. Just for test I made another pcb with TC4468. And no problems at all. I never was able to burn TC4468, even with inductive load.

Final pcb was made without mosfet drivers at all. PIC was able to drive irf1010ns at 976Hz and 3906Hz just fine, power loss was acceptable.

[MarkM]

Thanks for the tip! Glad to see I'm not the only one.

[T3sl4co1l]

I did mention earlier that the supply voltage ratings of these drivers are quite low.

If you're going to keep blowing chips, and never going to take the time to figure out what's actually going on, I can't be of much help.    If you're interested in doing some scoping, we can probably find why they're dying.

Specifically, power-on transients sounds like it might be related to lead inductance ringing with bypass capacitance.

Tim

[MarkM]

Hi, the thread is a bit rough to follow.. My buddy has a Rigol that he's using.  I've been using a POS hantek 6022be, which I can't capture the spike with.  I took a drive and scored a hantek DSO5102P last night for $185, so I can actually test myself now.  I didn't feel like I was just 'burning chips', because there was a point to it and I couldn't do anything else myself.

This issue is the ceramic capacitors charging when the battery is plugged in. That's 100% confirmed.  My buddy tested the tc4427 on a bare board with and without the capacitors.  Without the capacitors the tc4427 didn't die and there was no huge spike when turned on.  Add caps and there's a huge spike and the tc4427 is dead.  Actually, you can test this with caps by themselves and see the same thing.  https://www.pololu.com/docs/0J16/all

Zarhi said he/she had the same issue with the tc4427 dying like that and it's most likely the same issue.  The TC4422a was very hard to kill for me even with the caps.  I'm for sure ditching the TC4xxx either way, though.  I'm looking at cheaper options like the ZGD3003, which should be fine.  For some reason this driver isn't very popular, and I don't know why.  Any thoughts?


As far as actually fixing the issue, I've been thinking about it and reading quite a bit.  I think the best option would be to use a MOSFET to limit the inrush current. What are your thoughts on that?

http://www.eet-china.com/ARTICLES/2000DEC/2000DEC13_AMD_AN5.PDF?SOURCES=DOWNLOAD

[MarkM]

Yes, that's definitely the issue. 

[steverino]

Just curious, the referenced LT application note concluded  "To obtain optimum transient characteristic, the input circuit
has to be damped." and then showed three different simple damping solutions.  Will these not work in your circuit?

[MarkM]

The Motorola app note?  I'm going to have to read both a few more times to let all of the calculations sink in...  I'm honestly not even sure if this is really needed for my board.  The little arduino boards use a similar LDO to power them with similar input caps, but I've never seen one with any kind of inrush current limiter. I've plugged and unplugged battery packs over and over while testing with zero issues even after a long time.  My board is basically an arduino UNO with MOSFETs, a MOSFET driver and an OLED display on the board.

The other MOSFET driver is good up to 40v(ZXGD3003).  The max spikes that was seen was 25v or so.  Would you add some kind of inrush current limiter in this situation?   

[MarkM]

Thanks.  The 10x AL cap is actually what I tested so far, but not with the scope that I just aquired.  The driver didn't die, so it definitely helps.  There may be room on the board to add the big cap and use the other driver.

I think adding the big cap with the other driver would be a reasonable solution.
 

[MarkM]

Man, you lost me on those pics.  Care to elaborate on that? 

I got a chance to capture some of the spikes myself and test some things.  My setup is simple.  I have a 10µf ceramic cap soldered to 2.54mm pin headers, a rocker switch, XT60 connecter, 3s battery pack and a load resistor on a breadboard.  Check out the first pic.  The next pic is with a 47µf electrolytic capacitor.  After that is a 220µf.  I lost my 100µf cap somehow...  It looks like a 100µf cap should do just fine for my purposes from the results.  220µf would work as well if I could find one small enough.  Any thoughts?

I need to sit down and learn how to use the scope.  It wouldn't show the max values for each trigger.  I'll figure that out.  Loving it so far.

 

[nuno]

If you twist your power cable plus and minus, or put them more apart, does the ringing frequency of the 1st test (only the 10uF ceramic cap) change?

Looks like the wiring inductance is resonating with the input cap. The resonance frequency of such arrangement (series LC) is given by: Fr = 1 / (2 x pi x SQRT(L x C))

A rough estimate from your 1st scope shot could be 5us period, with C = 10uF and solving for L we have 5us = 2 x pi x SQRT(L x 10uF) => L = (5us / (2 x Pi))^2 / 10uF ~ 63.4nH which sounds realistic for a "piece" of wiring.

In AcHmed99's circuit, with 4.7uH inductance and 39uF cap, the formula above gives ~11.7KHz for the resonance frequency, which is spot on with his simulations.

[T3sl4co1l]

If you twist your power cable plus and minus, or put them more apart, does the ringing frequency of the 1st test (only the 10uF ceramic cap) change?

Should be.  Mind that, if you're thinking of a solution along these lines -- this isn't a help, because who's going to give a crap about where they are?  They're just wires...  Hey, I hooked it up and now it doesn't work...

Better to embrace the effect (free filtering..?), and damp it to an expected worst-case level (say, several meters of loose wiring -- a few uH, probably).

You can also use a TVS to clamp the spike.

If you're curious, this is what a short circuit at 30V looks like, when you switch it into 150mm of twisted pair:



20A/div.  The leading edge is about 0.2uH inductance, and whatever resistance was around (it's curved a bit).  The hump and cutoff are due to the current limiting and fault protecting action of the electronic fuse being tested.

What's not seen during this transient is the voltage: it dips down (putting the full 30V across the wiring), then the current limiter pulls out of saturation and the voltage overshoots.  A TVS diode protects the current limiter, clamping the flyback at 50-60V.

You can do a similar thing for supply protection -- and should -- by adding a TVS across the supply, at the connector or at the first large capacitor.  This also affords surge and reverse protection.

Tim

[nuno]

If you twist your power cable plus and minus, or put them more apart, does the ringing frequency of the 1st test (only the 10uF ceramic cap) change?

Should be.  Mind that, if you're thinking of a solution along these lines -- this isn't a help, because who's going to give a crap about where they are?  They're just wires...  Hey, I hooked it up and now it doesn't work...

I'm not . It's just an easy way for the OP to confirm that the effect is indeed this wiring-cap resonance.

Personally, and since I've met this effect while simulating "power systems" ^*, I would just add input capacitance, also making sure the resonance frequency is far away from any "power" oscillations (such as PWM signals) used in the system.


^* Ideally we want caps with as low ESR (and ESL) as possible here. Caps are charged/discharged in a controlled way.

[T3sl4co1l]

Personally, and since I've met this effect while simulating "power systems" ^*, I would just add input capacitance, also making sure the resonance frequency is far away from any "power" oscillations (such as PWM signals) used in the system.


^* Ideally we want caps with as low ESR (and ESL) as possible here. Caps are charged/discharged in a controlled way.

Depends on what "power systems" include, but this can be undesirable as well.  Startup and turn-on transients can very easily excite unintended resonances, so it's always good design to make sure all resonances are well damped.  If nothing else, you can't always guarantee that the only inputs to, or outputs from, the system will be fully under control of the control circuitry.  E.g., say you slew-rate limit the circuit, so it can't experience startup transients under normal conditions.  But someone can always short-circuit the output, forcing a transient there.  This can extend to, yes, rather unlikely stimuli (e.g., a diode fails shorted, discharging the filter network?), but the "best practice" conclusion is that, such situations should be avoided, by design, in the first place.

For an analogy, it's the difference between a car that drives absolutely smoothly, versus one that has some stupid rattle or vibration when the engine is at a particular RPM and load.  Or suspension that rides over bumps just fine, but sways just a little nauseatingly (a common mode - differential mode problem).

What I like to do with DC link filters is, put on enough electrolytic bulk as needed for filtering / ride-through, and enough film capacitors to handle the high frequency switching ripple and to get a nice low loop impedance going into the inverter (which may have additional inductance added intentionally).  The electrolytics don't have vanishingly small ESR, so they act to dampen the film caps (which, depending on how they are connected by traces or pours, may act essentially like one solid bulk capacitor, or may exhibit some lumped-equivalent transmission line behavior; in either case, ESR plays a role in damping it).

You generally want to avoid multiple levels of bypasses, because you don't need it, and you don't want it.  Placing small ceramics right at the inverter results in an additional LC loop, which probably isn't well damped because sqrt(L/C) will be higher than the surrounding impedances, and a suitable amount of series inductance
is desirable in the inverter.

Tim