Homemade spot welder keeps blowing MOSFETs

[Refrigerator]

Seems i can't make a welder of any type that doesn't just immediately vaporize itself. 
Long story short: i got some 18650 cells from my internship and i want to make a pack but i don't have a welder. So i made one and it exploded over and over again.
Long story long:
Ok, so the first few failures were my fault because i was basically using whatever scrap MOSFETs i could salvage and was just trying my luck basically.
But recently i was ordering parts from mouser and decided to throw in some proper high current MOSFETs and got 10 MDT4003SCT's.
Made my MOSFET assembly with plenty of copper and added a 10 ohm resistor in series with each gate, more to protect my gate driver than anything really.
Below included is a rough schematic for ease of understanding.
And all was well until today. Previously i used an old battery from my car but it would discharge too quickly, just not enough juice left in it.
So i took my 95Ah 850CCA 12V battery that had a shorted cell and cut that cell open to add a copper bar across the electrodes to bypass it so it's a 10V battery now.
And it took only about three tries before my MOSFET assembly let out the smoke. But while it worked anything i tried to weld would turn into vapor, the peak current was massive.
I actually didn't expect the battery to have this much oomph, like yeah, i know it's new, but it caught me completely offguard just how violent it is. 
I'm using a rather short 16mm² welding wire to tie everything together, which funnily enough was the thinnest welding wire i could find at the time.
I still have 9 IRF1324PbF's that i got from aliexpress. Had 10 but decided to crack one open to see it it's genuine and it was 
Well at least the silion die inside was massive and the bondwires were chunky as well, i'll include a pic.
According to the datasheet IRF1324 has about half the R_DSON of the MDT4003 so i have high hopes for it.

So i still want to make a welder but i think it would be best to put some more effort forth before i blow it up again.
Some thoughts in hindsight:
1. This welder is kind of sort of a boost converter if you squint hard enough so there will be a high voltage spike once my MOSFET's turn off. I knew this already before but wishful thinking lead me to not use a snubber of any kind = big mistake. Now i'm thinking about using a MOSFET as a voltage clamp. I've been experimenting in LTspice already with some source follower MOSFET voltage clamps. I chose MOSFETs because BJTs seem to not handle that well with high peak current. Also i doubt that a zener or a transorb will do. The IRF1324 has a 24V_DSS so i need the snubber/clamp to be fast and reliable.
2. Were my wires too thick?  Perhaps i should have used thinner wire to limit the current. Almost every DIY battery tab spot welder out there uses 6mm² wire, which is much thinner than what i use. I kind of went with the idea that more copper = more better but how good is too good?
3. Is my gate driver not good enough? I tried to make it reasonably beefy and used transistors that are capable of 3A sustained current. Or perhaps is the slope too steep and is causing ringing or other unvanted behaviour? I've seen some DIY welder projects where MOSFETs were driven straight from an arduino pin and they worked fine. Surely my driver is not worse than that.

I haven't spent that much money on this project (yet ) but i'm already feeling like i should calm down before i bury myself in a pile of blown up MOSFETs.
Also i took some cool looking macro shots of the blown up MDT4003 MOSFETs. Notice how small the silicon die is, also are those bondwires aluminum or silver?
BTW does Dave still do the magic smoke pic thingy? I have plenty of those 

What do you guys think? I guess it's time to place bets whether it explodes again or not. 
 
Pic list:
1. MDT4003 die shot from the side
2. MDT4003 die shot from above
3. MDT4003 blown right out
4. IRF1324 from aliexpress cracked open
5. Gate driver schematic, output stage x10 MOSFETs
6. Gate drive signal with 10nF load cap. Red trace - arduino pin, yellow trace - gate driver output.
7. MOSFET assembly, 10 MOSFETs in total, 5 per side. Double sided FR4 with copper bus bars soldered on. Under the black tape is a strip of protoboard that holds the gate resistors.

[f4eru]

1) shorting a battery is for sure an uncontrolled high current event, can give out over thousand amperes. Your mosfets are probably still too small for that.
2) Yes, the inductive spike, with the same 1500 amperes has to be dealt with for a few 100ns.
3) why not getting a working double pulse welder from Alibaba for 150-200$ ?

[Pulsepowerguy]

The current is uncontrolled and the stored energy of inductance in the circuit are bad for both the voltage and current rating of the FET. Consider using a limited current from the battery to charge a large capacitor instead ofr using the battery directly. This will limit the total energy of the pulse.

Definitely add a properly sized diode from the drain of the FET to the battery. This will clamp the drain to way less than its 40V rating at turn-off.

 

[Refrigerator]

3) why not getting a working double pulse welder from Alibaba for 150-200$ ?

No fun in that 

[ledtester]

3) why not getting a working double pulse welder from Alibaba for 150-200$ ?

No fun in that 

You can get workable spot welders from aliexpress/ebay now for < $20. They have flaws, but their designs are very simple and the flaws can be mitigated quite simply.

This video goes over their schematics, problems and workarounds:

All "Mini Spot Welder" versions: circuit Analysis, differences and faults
https://youtu.be/RSrlXqFxhp8

Even if you don't buy one of them it could serve as a good resource what the common problems are and how to build your own out of the parts you have on hand.

[jmelson]

Well, you need to put in a little bit of resistance.  Not a lot, we're talking about maybe a foot of #12 solid copper wire
(sorry about US measurements) in series.  How long is your weld pulse?  It should only be 100 us to 1 ms long, I'd think.

Also, check the rise and fall times of the gate drive.  You want the voltage at the gate to transition in a 100 ns or less, never leaving the FET in the linear region very long.  The Miller capacitance could make this hard to do, that's why commercial gate drivers can source/sink several amps.

Jon

[Refrigerator]

Well, you need to put in a little bit of resistance.  Not a lot, we're talking about maybe a foot of #12 solid copper wire
(sorry about US measurements) in series.  How long is your weld pulse?  It should only be 100 us to 1 ms long, I'd think.

Also, check the rise and fall times of the gate drive.  You want the voltage at the gate to transition in a 100 ns or less, never leaving the FET in the linear region very long.  The Miller capacitance could make this hard to do, that's why commercial gate drivers can source/sink several amps.

Jon

100ns is pretty fast. I think my gate drive rise/fall time is well below 100µs right now, i'll have to whip out the scope to confirm.

[NiHaoMike]

Perhaps it would make more sense to use a transformer so that you'll instead be generating higher voltage AC at a lower current? 48V would be a good primary voltage, reasonably safe to work with and large power supplies are fairly easy to find for cheap on the surplus market.

[Refrigerator]

3) why not getting a working double pulse welder from Alibaba for 150-200$ ?

No fun in that 

You can get workable spot welders from aliexpress/ebay now for < $20. They have flaws, but their designs are very simple and the flaws can be mitigated quite simply.

This video goes over their schematics, problems and workarounds:

All "Mini Spot Welder" versions: circuit Analysis, differences and faults
https://youtu.be/RSrlXqFxhp8

Even if you don't buy one of them it could serve as a good resource what the common problems are and how to build your own out of the parts you have on hand.

I've already seen these videos. It surprised me that tiny SMD transistors were enough to drive those MOSFETs.
I didn't see any snubber on these but i'll have to check.

[Refrigerator]

Perhaps it would make more sense to use a transformer so that you'll instead be generating higher voltage AC at a lower current? 48V would be a good primary voltage, reasonably safe to work with and large power supplies are fairly easy to find for cheap on the surplus market.

I actually was thinking about making a switch mode power supply with a high current secondary. When you think about it i could build an inverter to run off the 10V battery with maybe a 1:5 ratio on the transformer, which would mean 5x less current on the primary. But finding a suitable core is hard and magnetic field coupling would be poor with few turns of thick wire. If i used copper strip maybe i could make it better, but finding copper strip is not that easy also. I can get some aluminum bar but it might be too thick and stiff to wind into a coil.
I wanted to experiment with using regular laminated transformer cores for low frequency (sub 5kHz) SMPS so i might do just that with this.
A simple ZVS circuit would be enough to drive the transformer at full tilt, since that's all you really need. With a more sophisticated controller you could add all kinds of control and PWM but i'm kind of getting carried away here.

I have a ~15kg transformer with a primary wound sitting around ready to be used so if all else fails i'll throw a couple turns of thick wire on it and add an SSR with an arduino for zero crossing detection to control the thing.   
The transformer was meant to be a spot welder actually but for thick sheet metal and not battery tabs, hence why it's so big. IIRC the secondary was supposed to operate at a little over 2kA. But i can lower that with thinner wire and fewer turns.

[Refrigerator]

Played around with LTspice and came up with a way to make the MOSFETs eat their own back EMF.
If i place a 10V zener from gate to drain the MOSFET will clamp the drain voltage at around 17-ish volts.
Problem is that the peak current is the same as welding current so whatever snubber i use needs to be able to handle that.
If i place a shottky on the drain and charge a cap with it a 4700µF cap charges up to over 20V in a single pulse, so that's a no-go.
Also if i replaced my 16mm² wire with 6mm² wire the peak current should drop below 1kA.
Problem is that the inductance rises as wire diameter decreases.
I need about 10m to drop the current below 1kA, battery ISR is about 5-ish m .

Bonus pic is my battery mod. I actually managed to un-short the cell but it was so badly sulphated that it would probably barely pass enough current to start a car.
So since the bad cell was a massive chokepoint i bypassed it  and cooked my MOSFETs shortly after 

PS: it's kind of funny talking about kiloAmps, like oh yeah just add a couple milliOhms to drop it below 1k, she'll be right. 

[f4eru]

100ns is pretty fast. I think my gate drive rise/fall time is well below 100µs right now, i'll have to whip out the scope to confirm.

Too slow. You have to speed it up. Put darlingtons, with final tranistor multiple in paralell, lower the resistors, etc...

If i place a 10V zener from gate to drain the MOSFET will clamp the drain voltage at around 17-ish volts.

It is used for some lower power stuff. Not a  good idea here at all.
2000A * 17V -> 34 kW instant power.

Use big fat fast diodes over the load, and add also the biggest TVS you can find onto the mos, both are needed.

I have a ~15kg transformer with a primary wound sitting around ready to be used so if all else fails i'll throw a couple turns of thick wir

Not nice. It my weld, but not well, because it's much too slow in rise/fall time for those small and fast spot welds, there's too much inductance in the transformer

[Mechatrommer]

Made my MOSFET assembly with plenty of copper and added a 10 ohm resistor in series with each gate, more to protect my gate driver than anything really.
Below included is a rough schematic for ease of understanding.

just to emphasize what other have said that you seem to ignore... what you show in OP, without current limiting resistance at the drain and source, is a smoke guaranteed circuit, regardless of how fast the gate is. you cant respect only the gate pin, you need to respect all 3 pins, nuff said. hint: lower the voltage, or/and increase wire length, and mind the inductance spark (high voltage)... protection diodes?

[jmelson]

100ns is pretty fast. I think my gate drive rise/fall time is well below 100µs right now, i'll have to whip out the scope to confirm.

100 us??!!??  That will blow the FETs right there!  Yes, 100 ns is a bit fast, you can probably survive full turn on and off (including Miller delay) of 500 ns or a little more.  Anything over 1 us, however, in this extreme condition will guarantee failure of the FETs.  Make sure to measure the gate charge and discharge with a voltage swing on the drain.  As the drain voltage drops during turn-on, the gate-drain capacitance fights the driver trying to charge the gate.  And, the opposite at turn-off.  That is the Miller effect.

You can put a 100 Ohm resistor in place of the welding electrodes to observe the Miller effect on the bench.  You will be surprised how bad it gets with a weak gate driver.

Jon

[Refrigerator]

100ns is pretty fast. I think my gate drive rise/fall time is well below 100µs right now, i'll have to whip out the scope to confirm.

100 us??!!??  That will blow the FETs right there!

Well i did say well below 100µs. I haven't yet checked the rise/fall time with my scope.

[Refrigerator]

Checked the rise and fall times with my scope using a 100nF load on the gate driver and might have found my problem.
Rise time comes in at 79ns but fall time comes in at a much bigger 18µs.
So it looks like my gate driver needs improving.
Hopefully this slow fall time will increase the rise time on my learning curve, huh. 

Ps: i thought that maybe my low side transistor was blown but LTspice shows similar behaviour.

[Refrigerator]

Played around with LTspice and managed lower the rise time down to about 280ns. I wonder if this is good enough.
All it took was another shottky and an NPN under the low side PNP to form an SCR. The shottky it there to limit base current.
There aren't any transistors equivalent to what i have so i think i'll just upgrade my gate driver and test it out to see what the new fall time is for real, brb.

[Refrigerator]

If i place a 10V zener from gate to drain the MOSFET will clamp the drain voltage at around 17-ish volts.

It is used for some lower power stuff. Not a  good idea here at all.
2000A * 17V -> 34 kW instant power.

But only a 30mJ increase in energy dissipation per MOSFET according to LTspice.
I have some massive diodes that i'll put across the load to shunt as much of the overvoltage as possible back to the battery.

[Refrigerator]

So i modded the circuit and nothing seemed to change, weird right?
But then i took a look at my breadboard and noticed that my load cap was inserted one pin over and was for the previous test as well so that one's completely false.
I can't go back to the previous configuration but after the changes now i get around 1.1µs rise and 572ns fall times.
Since the high side transistor wasn't changed in any way i can assume that the rise time was the same previously also.
So it seems my gate drive circuit needs some improvement still.

[jmelson]

Checked the rise and fall times with my scope using a 100nF load on the gate driver and might have found my problem.

Well, these are not real.  You must connect the power FET and have representative voltage on the drain to see the Miller effect.  On many FETs, the Miller effect is BIGGER than the plain gate capacitance.  (Maybe not so with only 12 V supply, though.)

Jon

[jmelson]

Played around with LTspice and managed lower the rise time down to about 280ns. I wonder if this is good enough.
All it took was another shottky and an NPN under the low side PNP to form an SCR. The shottky it there to limit base current.
There aren't any transistors equivalent to what i have so i think i'll just upgrade my gate driver and test it out to see what the new fall time is for real, brb.

It looks like R5 is what is draining the capacitance, not your 2 transistors.  You could try reducing R2, not sure it would help much.

Jon

[Refrigerator]

Played around with LTspice and managed lower the rise time down to about 280ns. I wonder if this is good enough.
All it took was another shottky and an NPN under the low side PNP to form an SCR. The shottky it there to limit base current.
There aren't any transistors equivalent to what i have so i think i'll just upgrade my gate driver and test it out to see what the new fall time is for real, brb.

It looks like R5 is what is draining the capacitance, not your 2 transistors.  You could try reducing R2, not sure it would help much.

Jon

R5 does barely anything actually, if i change it to 200 Ohms i can reduce the fall time by about 40-ish ns.
It's purpose is to keep the gates discharged whenever the gate driver is unplugged from the MOSFET assembly.

[Pulsepowerguy]

Without a freewheel diode connected from the drains to the battery there is great risk of over stressing the FETs during turn-off.

I prefer to use a gate driver chip for my designs (though discrete is fine if done right). My go-to parts are from IXYS/Littelfuse, e.g., IXD_604 or IXD_614. These are very rugged and inexpensive.
https://ixapps.ixys.com/DataSheet/IXD_604.pdf

[Refrigerator]

I don't really have that many components to choose from but i did recently receive some AP4506GEH that i had ordered from aliexpress.
These are a PMOS + NMOS combo type thingies, joined at the drain.
LTspice didn't have these but it's easy enough to just pop in some other MOSFETs that are close enough.
And these MOSFETs also needed a gate driver each of their own so now my MOSFET gate driver MOSFETs needed a gate driver.
But i couldn't drive the gate driver MOSFET gate drivers straight from the microcontroller pin.
So i added a driver for the gate drivers of the gate driver MOSFETs. 
And after a bunch of tinkering  i managed to get rise and fall times below 4ns.
But that's before the gate resistor. After the gate resistor the fall time alone increases up to almost 200ns.
So i'd say it's pretty much impossible to get rise and fall times on the gate any lower than 100ns since it's limited by the gate resistors. 

I included the LTspice simulation in case anyone wants to check out.

Without a freewheel diode connected from the drains to the battery there is great risk of over stressing the FETs during turn-off.

I prefer to use a gate driver chip for my designs (though discrete is fine if done right). My go-to parts are from IXYS/Littelfuse, e.g., IXD_604 or IXD_614. These are very rugged and inexpensive.
https://ixapps.ixys.com/DataSheet/IXD_604.pdf

That actually looks like a really nice part to have i'll have to check it out.

[Manul]

But that's before the gate resistor. After the gate resistor the fall time alone increases up to almost 200ns.
So i'd say it's pretty much impossible to get rise and fall times on the gate any lower than 100ns since it's limited by the gate resistors. 

To go faster, you need lower value resistor, more current. The problem is that most likely it will start ringing. If you want to push high current into gate, you can, but you need very low parasitic inductances to prevent ringing. That means a good layout, very short traces. Also, not all mosfets are equal, you might be able to find one with lower gate capacitance. Still, this is not the whole story, because there is miller effect, which can prolong the turn-on. Also there is parasitic source inductance, so high drain-source current spikes cause the actual source potential to rise above ground and can cause slow down of turn-on. That is why you can find mosfets in 4 pin packages with kelvin source pin used only by gate driver, but not carrying drain-source current. All in all, high speed, high current mosfet circuits can be a pain to do good.