UC3843 IC keeps shorting FET

[Dannyx]

Good day folks. I've got this module/sub-board with 3 chunky relays on it out of a Legrand UPS - pretty big and beefy thing. My colleague began repairing it, replaced some components which were blown, but eventually gave up due to lack of time after failing to get it to work after several attempts, so I decided to give it a shot myself.

So what even is this ? Its purpose as part of the UPS itself is of no relevance to us right now - what DOES matter is that we're dealing with a non-isolated power supply driven by U1, a UC3843 (can't figure out how to insert links without pasting the whole URL, sorry... ). Glossing over what it actually does when bolted in its place inside the UPS (since I'm not sure), it's powered straight from the mains AC input of the UPS (measured continuity to confirm this), which is why I just soldered that mains lead where the lugs are to power it on the bench, as it would take far too long to install it back on the UPS for every test.

FET Q1 was blown. U1 was shorted VCC-GND. R5, R13 were open (R13, R5). With no exact schematic available, as expected, we did the next best thing we could and tried analyzing other schematics based around the UC3843 to at least try and suss out the values of these 2 resistors.

R5 is the gate drive - it sits between the gate of the FET and pin 6 of U1. It was fairly simple to work out the value, as most schematics show it as a low-ish 10ohm resistor, or thereabouts which makes sense - you don't put a high resistance between the gate and the driver IC, so we went with 10ohms.
R6 is a 10k resistor sitting between the gate of the FET and GND, which corresponds to the middle AC lug. I'd call it "neutral", but here in the EU that's all academical, because it all depends on how you plug it in at the wall, so I'll just call it "AC". This also makes sense because all the negative legs of the electrolytic capacitors are tied together to this lug.
R13 is not so straightforward: it sits between the large ceramic 0.15ohm current sense resistor and pin 3 of U1. Unfortunately, there doesn't seem to be a clear-cut value for it: it varies from circuit to circuit. I see my colleague left a 1k on there, which indeed seems to be the most prevalent value among other circuits, though I've seen it as high as 5k - never lower than 1k, so I just rolled with it for now.

For Q1, my colleague installed a 6N90C - whether this is a direct replacement of the original, I have no idea, so I just rolled with this too.

There's also the matter of this U2 (LM393) in the bottom-left which to me is a bit of a mystery as to what it does - I know it has something to do with pin 1 (COMP) of the IC, but I don't know how it works: long story short, U2 was also damaged. I removed it and tried it out on its own - the outputs were shot and remained 0v regardless of what I fed on the inputs. I replaced it too. I also checked U3 (TL431) off the board and it's fine (2.5v on the REF pin).

With all this done, I used a 25w bulb in series with the mains and plugged the whole thing it: the filament glowed briefly as the big cap charged, then it began pulsing faintly as if trying to pull the relays in. I took a shot plugging it in without the bulb, thinking perhaps it hasn't got enough current. Sadly, it popped the fuse, shorted D2 (upper-right) and Q1 became shorted D-S (not G though !). Everything else seems to have survived, resistors and all. I'm not sure about U1, other than the fact that it doesn't show any shorts to GND, but given that none of the resistors blew and Q1's gate was ironically NOT shorted, we can assume it's fine for now. Ok, so what now ?

I tried it one more time, thinking perhaps D2 was already shorted the first time around and I simply missed it. Not having another 6N90C on hand, I used a 2SK3532 in its place. Same story though: the filament pulses repeatedly and blows D2 and Q1 when directly on mains. Because of this, I can't take any voltage readings.

Another thing I tried: hooked up my bench supply to those 12v test hooks (AC disconnected, of course). All 3 relays pull in and the board seems stable. Current draw is around 40mA (probably close to 1A at pull-in, with such beefy coils). NOW I can take some partial readings:

- 0.7v on pin 1 (COMP) of U1
- 5v on pin 8 (VREF) of U1
- 12v on pin 1 (OUT) of U2
- 2.5v on pin 1 (REF) of U3

That's it for now....

My money is on R13 not being the right value. I'm aware there IS a way to calculate these values based on calculations and formulas in the datasheet, but it goes way over our heads for now - we're not exactly scientists to be able to work through all that, plus we don't know the specs of the transformer to begin with, so we just have to ballpark it enough to get it going again. Any help as to why may be causing the FET to lock up and short D-S each time would be greatly appreciated.

Pictures were taken in a bit of a hurry I'm afraid (Friday ), so they're not ideal...sorry about that.

[DavidAlfa]

Yeah 25W is to small, try a higher power bulbs, 100-500W will be good.
Anytime the fet blows, I'd replace the UC.

[Dannyx]

The same failure occurred even WITH the 25w bulb, somewhere in the process of me replacing parts and testing it out, so going any higher in watts would make it even worse. It's a very low powered thing anyway, so I'm expecting it to work even with this bulb, as I've tested many other supplies with it (no active PFC ones, as those indeed don't like being current-limited). I was thinking of pulling off the relays to see if it manages to produce ANY output voltage with no load, though that would be tough since they're so big...

I imagine the large pull-in current of all 3 relays trying to turn on at once somehow causes the IC to lock up for some reason. From what I recall, Q1 doesn't immediately become shorted upon connecting to mains - it's not a violent failure either - it happens silently, presumably after the IC's VCC pin ramps up and tries to fire up, only THEN Q1 shorts out...

[magic]

Make sure that there is no short in parallel with the current sense resistor. If possible, try to verify its resistance with a 4-wire ohmmeter or a 1A or similar current source and a voltmeter. Verify again that R13 is alright and that nothing shorts its CS pin end to ground. All these things could result in UC3843 underestimating FET current and keeping it turned on.

Shorting the COMP pin to ground is safe and it should make the PWM output permanently low. The LM393 is most likely meant to turn off the supply under some conditions if its output is connected there. COMP determines peak FET current, higher means more current, beginning at some minimum turn-on threshold.

You could try powering it up with grounded COMP and maybe without the power transistor - nothing should happen, the output must be low. If it isn't, look around for abnormalities. Perhaps there is a zener from VCC to ground and it's blown open circuit, causing overvoltage to the chip and failure. There is probably a hundreds-kΩ resistor pulling up VCC to the positive of the mains rectifier.

Another thing I can think of is that maybe a failure of the snubber could kill the MOSFET, or the MOSFET installed by your colleague has insufficient ratings. I'm too lazy to look up the parts you mentioned, but FYI typical drain voltage rating is at least 600V in those switchers.

Or, wait, is there no snubber here? There is often a diode from the drain to a high voltage capacitor, and a resistor from there to ground, but I don't see it on your pics.

[DavidAlfa]

The snubber seems to be d6 + d4.

[fmashockie]

Per this block diagram below from Legrand's website, looks like the two relays on line and neutral are safety relays.  When the load gets too great, the UPS switches from the inverter to this bypass line (direct from mains) via the 3rd relay. 

I would do the same thing you did before by adding 12V to those pins with your supply.  But probe the RT/CT pin on UC3843 with a scope.  You should be seeing an oscillating waveform there.  The frequency will depend on the values of RT and CT.  But you can calculate an estimation of the frequency based on those values to see if it matches.  Look up the datasheet for the calculation and the pinout to find RT and CT.  That might give you a a good idea as to whether or not its working.  Because the other voltages you measured on the pins look normal.

What I don't understand is how this daughter/sub board is getting its 'info' to open and close those relays.  For example, if the bypass relay is only suppose to kick in when the load reaches a certain point, how does this little board know that?  I would imagine that current/load sensing is taking place on the main board.  I only see connections for L, N, and Bypass (in and out) as well as 12V and GND pins.  But you would think there would be some other connector for signaling from a micro or something right?

[Dannyx]

Perhaps there is a zener from VCC to ground and it's blown open circuit, causing overvoltage to the chip and failure. There is probably a hundreds-kΩ resistor pulling up VCC to the positive of the mains rectifier.

Another thing I can think of is that maybe a failure of the snubber could kill the MOSFET, or the MOSFET installed by your colleague has insufficient ratings. I'm too lazy to look up the parts you mentioned, but FYI typical drain voltage rating is at least 600V in those switchers.

Or, wait, is there no snubber here? There is often a diode from the drain to a high voltage capacitor, and a resistor from there to ground, but I don't see it on your pics.

Couldn't find a zener diode myself indeed, which is unusual, given that I've seen one on the VCC pin on just about every switching IC I've come across so far and it's the first thing I check when the IC's damaged, especially since this one is being powered straight off the HV rail at startup via those 330ohm resistors in the top-left (the ones with the heatshrink tube on them), so a zener makes ABSOLUTE sense there to achieve some sort of regulation ! One of those 330ohm resistors was also blown, by the way. There ARE a couple of diodes just under the first relay, but I can't tell for sure what they are or if they're even part of the VCC circuit. Pulling off the relays would be a nightmare, but I guess I'll have to do it eventually if it comes to that. Even so, the IC seems to have survived so far (no short from VCC to GND on this one YET), though that may not necessarily mean it's functioning as it should.

About the FET: while I don't know for sure what the original FET was, I imagine my colleague did his best to replace it with something suitable, if not identical - he knows that much . An under-spec'd FET was my first thought too, although this is a fairly low power application, so I imagine unless the specs are way off the mark, just about any FET should work. From a factual standpoint, the 6N90C IS indeed rated for 600v and the 2SK3532 I used is even higher, at 900v, with other specs like resistance and capacitance fairly close or better, so I eventually dismissed this as a possibility.

I don't see a snubber either, other than those 2 diodes, which of course measure fine on my DMM.

Per this block diagram below from Legrand's website, looks like the two relays on line and neutral are safety relays.  When the load gets too great, the UPS switches from the inverter to this bypass line (direct from mains) via the 3rd relay. 

I would do the same thing you did before by adding 12V to those pins with your supply.  But probe the RT/CT pin on UC3843 with a scope.  You should be seeing an oscillating waveform there.  The frequency will depend on the values of RT and CT.  But you can calculate an estimation of the frequency based on those values to see if it matches.  Look up the datasheet for the calculation and the pinout to find RT and CT.  That might give you a a good idea as to whether or not its working.  Because the other voltages you measured on the pins look normal.

What I don't understand is how this this daughter/sub board is getting its 'info' to open and close those relays.  For example, if the bypass relay is only suppose to kick in when the load reaches a certain point, how does this little board know that?  I would imagine that current/load sensing is taking place on the main board.  I only see connections for L, N, and Bypass (in and out) as well as 12V and GND pins.  But you would think there would be some other connector for signaling to a micro or something right?

A scope would indeed come in handy here, but I don't have one, so I'll just have to shotgun-approach it or at least limit myself to what I can achieve with a DMM sadly.

True, that was also the first thought that came to my mind when inspecting this thing: if all 3 relays pull in at once and there are no other control pins, what even is the purpose of this board ? I thought there might be other components under the relays that somehow sense the AC input, but there don't seem to be any when peeking through the gap between the board and the relays. One theory is that it's simply separating the AC input from the inverter, preventing the inverter's output from feeding back into the grid, so as long as mains is present, this board is active and the relays are closed, allowing mains to go INTO the unit, but as soon as mains fails, the relays open and the inverter takes over....this is probably completely bonkers and unscientific, especially since I haven't even bothered to look inside the UPS so far. It also doesn't explain why even the bypass relay pulls in, unless there are some other relays somewhere else before this board that handle switching, while these 3 are "dumb" and only do what I suggested.

If this were the case, I was thinking of ditching this power supply entirely and just replace it with a standard 12v power supply. I'd connect the mains input to the lugs, just like in the original setup, and the 12v output to where the board should generate its 12v output. I was just about ready to give it a shot, but then a problem occurred to me: the relays might need to open immediately when the AC input is lost, whereas my off the shelf power supply might have some "hold-up" time as it slowly discharges, keeping the relays closed slightly longer, which may be unacceptable and might lead to something else exploding as a result of two circuits meeting when they shouldn't, otherwise I can't imagine why they wouldn't just use the relays on their own (with 230v coils), instead of designing this whole board....

[fmashockie]

Yea those relays will be a pain to remove.  Look at the size of those pads! likely need a combination of hot air and iron to remove them.  Unless you have a good desolder gun or powerful iron!

[eurgenca]

Just curious, are this 2SK3532 fet's from Ali or similar?

[magic]

Couldn't find a zener diode myself indeed, which is unusual, given that I've seen one on the VCC pin on just about every switching IC I've come across so far and it's the first thing I check when the IC's damaged, especially since this one is being powered straight off the HV rail at startup via those 330ohm resistors in the top-left (the ones with the heatshrink tube on them), so a zener makes ABSOLUTE sense there to achieve some sort of regulation ! One of those 330ohm resistors was also blown, by the way. There ARE a couple of diodes just under the first relay, but I can't tell for sure what they are or if they're even part of the VCC circuit. Pulling off the relays would be a nightmare, but I guess I'll have to do it eventually if it comes to that. Even so, the IC seems to have survived so far (no short from VCC to GND on this one YET), though that may not necessarily mean it's functioning as it should.

Maybe one of the small SMD diodes like D5 or D8?

These resistors should be 330K, not 330Ω. Double check what you have installed and calculate what the resulting current will be.

The chip has a built-in 36V zener supposedly good for up to 30mA, but no one ever uses it at such voltage. Most MOSFETs aren't rated for 36V gate drive.

[Dannyx]

Maybe one of the small SMD diodes like D5 or D8?

These resistors should be 330K, not 330Ω. Double check what you have installed and calculate what the resulting current will be.

The chip has a built-in 36V zener supposedly good for up to 30mA, but no one ever uses it at such voltage. Most MOSFETs aren't rated for 36V gate drive.

About those 2 diodes: they're slightly confusing.
D8 sits between a pin of the transformer and cap C10, which also corresponds to that 12v test hook, so it's likely our schottky diode that rectifies the output of the transformer like it would in any SMPS. I remember I checked the diodes several times and it's fine, on my DMM at least - hope it doesn't break down under load.

D5 has its anode connected to C10 as well (so the 12v "rail") and its cathode is connected to the + leg of cap C2 (if you look close enough you can even see the little stripe in one of the pictures I've taken), which in turn corresponds to the VREF pin (pin 8 of U1). Assuming this IS a zener diode, given that VREF is 5v and VCC is 12v, this would mean it's attempting to clamp VREF at a maximum of 7v ? My zener theory is a little rusty. Even so, I don't see a series resistor there to make this possible, so it's not "actively" regulating - it serves as a one-shot only type of deal, possibly killing the zener if VREF DOES spike above 7v. Still, D5 is fine - no shorts and a drop of 0.5v.

Resistors: actually they're 33k (three orange stripes). My bad.

Just curious, are this 2SK3532 fet's from Ali or similar?

Nope - it's from my parts bin, pulled off junk boards

[Dannyx]

Ok, time for some clarifications and corrections:

I was wrong about the placement of D5: D5 sits between the cathode ("output") of D8 and the VCC pin of U1 (NOT the VREF pin like I wrongly assumed the first time), so it's most definitely not a zener. It also has the positive leg of C2 (small electrolytic on the left) and ceramic cap directly above U1 connected between its cathode ("output") and GND. The VREF pin is also bypassed to GND with ceramic cap C16 (second cap above U1).

I kept overlooking C10 (output electrolytic cap): I removed it today and even after several tests with my (humble) ESR meter, the readings were not satisfactory, so I replaced it right away - the specs seemed way off. When in doubt, replace. C2 is much better, so I left it alone.

I pulled off the first relay that was covering up some stuff underneath - it was surprisingly easy to do with my iron and hot air station. I took some pics there too. No components are out of spec though, so maybe it DOES come down to C10 being all whack....only one way to find out and that is to give it another go.

I also noticed that pin 2 of U1 (VFB) doesn't settle at 2.5v when feeding 12v with my power supply: it drifts up and down with the voltage I apply. Say I feed 10v, pin 2 reads 2.5v, but if I increase the voltage closer to the expected 12v, the pin now reads 3.1v instead ! I was expecting the feedback loop to keep it at a stable 2.5v once the output reaches the desired value. The VFB pin is the inverting input of an op-amp inside the UC3843 itself, so the voltage can only come from somewhere else on the board - will do some more testing. Also notice C23 is missing on the side there...not sure if this has anything to do with our fault, but will try finding a replacement for it as well...

[Vovk_Z]

Accordingly to datasheet Vfb is 2.5 VDC when Vcomp is 2.5 VDC. Otherwise it mustn't be always 2.5 VDC.
Only VREF has to be always constant.
UC3843 is pretty popular, available and cheap so I would just replace it if all other components are tested.

[Dannyx]

Correct. I forgot my op-amp basics. To confirm, I shorted COMP to VFB, creating a closed loop and I got 2.5v on that point, since the N/I input of the internal op-amp is referenced internally to a 2.5v source, so the op-amp is working as it should. Normally, COMP is connected to VFB through R11 (a 30kOhm resistor) and C14 directly below it (upper-left corner of U1) so I just jumped across those with my tweezers.

Also tried powering the board without the transistor installed and it's stable (filament of 25w bulb goes off entirely after the main cap charges). U1 tries to start up but can't because its VCC pin only goes up to around 8v which is too low, since it's being fed via those 33k resistors straight off the rectified DC+ rail. I can see pulses on the VREF pin as it tries to start, but without a scope it's hard to take more precise measurements.

[magic]

D5, D8 look like output rectifiers as, they appear to be connected between transformer secondary windings (I think there is more than one) and capacitors.
Perhaps D12 is the VCC zener?

D4 and D6 are the snubber. They go in parallel with primary winding, D6 blocks D4 conduction when the FET is on and D4 probably is a TVS or zener, absorbing voltage peaks when the FET turns off. D4 going open or D6 going short could damage the FET.

Not sure if you checked all those resistors in feedback circuitry, near TL431 and TL393 yet?

[Dannyx]

Ok, progress: turns out it WAS C10 way out of spec that was causing the FET to blow. Probably because it was letting noise into the VCC pin of U1 causing undesired behavior. I used a 2SK3679 this time to test. Replacing C10 helped a lot: when I powered up the board with AC, still through the 25w lamp of course, the two relays still left on the board latched on this time, unlike last time when nothing happened and the FET eventually died. I could also hear some coil whine coming from the transformer, so now there was switching action going on. I tried leaving it on a little longer but the filament began to glow faintly again until it eventually went out, along with the coil noise, hinting that something had died. I quickly pulled the plug to inspect. Nothing catastrophic fortunately: R5 (the gate drive resistor) had gone from 10ohms to 78ohm, so it's good as dead. If I NOW attempt to power on, the filament just pulses briefly every 0.5s or so, because there's insufficient gate drive to fire the FET. I even applied power after removing R5, essentially disconnecting the FET from U1 entirely, just to confirm nothing else is dead and thankfully it isn't - no reaction from the bulb whatsoever, which makes sense, since the FET is grounded via R6 (10k).

Something sunk too much current from the OUT pin of U1, just not instantly - like something heated up slowly until it eventually gave in. Good thing the resistor was there in the first place to act like a fuse, otherwise I imagine the transistors in U1 would get shorted...

[DavidAlfa]

The resistor drifting to 78R is very strange.
Either a very crappy resistor not able to handle short high current spikes, damaged by soldering, or the fet gate shorted out.

You should make sure the resistor and cap connected to RT pin (R12, C16 ?) are ok, otherwise the switching frequency might be completely off, killing the fet.

[SeanB]

So transistor is a little too beefy, too high capacitance for the chip to drive, causing the resistor to cook up in value. Find a junk transistor with lower gate capacitance and try it, as it really only needs to be rated for 600V plus and 5A peak current anyway, so a lower capacitance will help with lowering gate drive and the charge injection. C23 looks like it is a parallel device with c22 and C21, so likely a do not place part, used only on certain models or power rating to set a RC time constant.

I would replace C2 as well, small electrolytics on SMPS primary side are almost always a failure point, even if they still test acceptable, they often enough were marginal in the first place. I keep a pack of 47uF 50V low ESR ones in stock, just for this case, even if the original is any value from 22uF to 100uF, the new capacitor, even if it needs to be made to fit SMD by bending leads and nipping off a tiny leg to solder on, and a drop of glue to hold it to the transformer invariably next to it, it works.

[Dannyx]

The resistor drifting to 78R is very strange.
Either a very crappy resistor not able to handle short high current spikes, damaged by soldering, or the fet gate shorted out.

You should make sure the resistor and cap connected to RT pin (R12, C16 ?) are ok, otherwise the switching frequency might be completely off, killing the fet.

No, the FET appears to be still fine (at least as far as I can test with my DMM)

Here's a clearer picture of those support components in the RT/CT area. R12 is a 10k (01c SMD code) resistor and it's fine. I went as far as to pull cap C16 off (and lost it several times ) to measure it. I took several readings, of course, and get around 2116-2126pF on my ESR tester, sooooooo that would round out to the nearest value for a ceramic cap which is like what ?....2.2nF ? I tried plugging these numbers into the formula on page 13 of the UC's datasheet (the TI version) and got a frequency of around 81-82kHz...whether that's correct or not I don't know. It doesn't change the fact that neither C16 nor R12 have been removed before, so they're the original values from the factory, so we'll leave those alone for now.

So transistor is a little too beefy, too high capacitance for the chip to drive, causing the resistor to cook up in value. Find a junk transistor with lower gate capacitance and try it, as it really only needs to be rated for 600V plus and 5A peak current anyway, so a lower capacitance will help with lowering gate drive and the charge injection. C23 looks like it is a parallel device with c22 and C21, so likely a do not place part, used only on certain models or power rating to set a RC time constant.

This would mean the 6n90C my colleague used is even worse and it's likely not factory, since its capacitance is even higher, at 1360pF and 1770pF max, whereas the one I put in is ever so slightly better at 1100pF nominal, though still close to the other one at its maximum - 1650pF. I found the 900v rating of the 6n90C rather excessive indeed, but tried sticking with it, thinking perhaps he replaced it with the original part he found on there, assuming it wasn't exploded and illegible. Guess it's worth digging up another transistor, which I thankfully have thousands of...

C23 looked broken off to me, especially since there's some fine scratch marks on the two tracks down and to the left of it, so I thought I should pull off one of the other two to read its value and source a replacement. Upon swabbing those pads with some alcohol though, they appear pretty clean, although everything else on the board appears populated, so I doubt they'd leave just this one part unpopulated. Again, whether this causes the issues I'm having is not certain, as I haven't reverse-engineered the board enough to figure out what those caps do. Very briefly: they're grounded on the inner side ("top" side in the picture) and the outer side ("bottom" of the picture) is connected to the the outer side of R26, anode of D12 and the left pin of Q3. Cathode of D12 goes to 12v AND top side of R25.

I was thinking of replacing the gate resistor with a 22ohm one from my junk bin, but that wouldn't really solve much - it'd waste power and might cause the FET to overheat, so I'll start with the FET first..who thought a simple supply like this would be so specific.... I'd just throw any random FET in dead wall-warts and they'd fire right back up again....but then again, maybe those with a different switching IC or none at all

[DavidAlfa]

The gate resistors is pretty important!
Too low and it'll cause ringing, unexpectedly turning the fet on/off several times on each transition and killing it, but too high and it'll make switching slow, causing lots of heat, also destroying it.

Thus, you want it slighly slower than ideal, but not too much.

As per TI datasheet, Fosc =1.72 / (RRT × CCT), which gives about 78KHz with your values.
At such frequency you'll definitely want to keep the rise/fall times as low as possible.

Calculate 5RC (Full (dis)charge time constant) for 2nF (Exaggerating the gate capacitance a little) and R=22R: 5*44ns,  it's pretty nice.
Even 47R would probably be ok.
Actually the gate capacitance calculation is not that simple due the Miller effect, but gives an approximation.

[Dannyx]

As per TI datasheet, Fosc =1.72 / (RRT × CCT), which gives about 78KHz with your values.
At such frequency you'll definitely want to keep the rise/fall times as low as possible.

Calculate 5RC (Full (dis)charge time constant) for 2nF (Exaggerating the gate capacitance a little) and R=22R: 44ns, it's pretty nice.
Even 47R would probably be ok.

Thanks. Is the calculation you used meant to exemplify the turn on time with a 22ohm resistor for R5 ?

The 10ohm resistor I found on there was most definitely NOT factory, so if could've been higher indeed, hence why it got fried, despite the FET surviving.

[DavidAlfa]

The formula is simple as t = R*C (Ohms, farads).
Then the aproximation of a full (dis)charge cycle is 5*t, thus 5RC.
There're plenty of online calculators for this:
https://www.translatorscafe.com/unit-converter/en-US/calculator/rc-circuit/

In fact I forgot to do the 5x part, so I fixed the previous post.
200ns is not amazingly fast but realistic, the gate normally turns off below ~3V, so it'll be a bit faster than that.
Often the major limitation comes from the gate driver and its ability to provide current.
Fast fet drivers have very strong outputs capable of supplying peaks of 10-20 Amps, some might switch the gate in 10ns!

The resistor shouldn't burn, no matter if it's 1, 10, 22, 33ohms...
Xc (Impedance) of 2nF at 78Khz is ~1Kohm, so the average current will be pretty small.
(Gate voltage is usually 12-15V)
If you want to check yourself:
https://www.omnicalculator.com/physics/capacitive-reactance

[Dannyx]

Yeah, I was sort-of familiar with the time constant (the most memorable thing about it being the letter "TAU" - always found that amusing ), just didn't know how to use it in this context. Guess it's time to try a 20ohm resistor then. Should I keep the 2SK3679 for now ?

[Vovk_Z]

Gate resistors usually die after a mosfet, but not before (though it can be too).

[Dannyx]

This was the complete opposite every time: when the FET shorted, resistor was fine. Resistor burned, FET was fine