I've been reading and enjoying everyone's posts - lots of experience to share here - so I thought I'd throw mine out there as well.
I have the same 6 FETs driven by 1407 driver like most designs here.
Been blowing a few FETS now and then and even a couple of the driver chips shorted the output high - that was a "long weld pulse'.
Here's what I found / learned so far to share.
1. FETs usually go after they have been weakened and overstressed after a few too many cycles or when one of them gets more than
its share of current. This can happen when there are minor differences in their turn on time - likely caused by slight gate voltage differences or even PCB board layout issues. Since the gate is voltage sensitive using separate (6) 82 ohm resistors off the driver chip is the best bet. Now at least they will all see the same output pulse and have a chance at being turned on at the same time. If you were to have a single drive current resistor then the device that turned on first would 'hold down' the drive voltage further amplifying any device differences.
Conclusion 1. - use separate, closely matched drive resistors. (same as riba layout). For PCB layout try to drive the string with about equal length traces etc. to each FET.
2. Since its not practical to sort FETs with perfectly matched gate turn on thresholds - mitigate differences in their turn on time, to avoid large current sourcing differences between the string at their Drain/ Source. Most of the welders out there I suspect already do this by having a few feet of weld wire to the hand held probes. If you did as I did though - build a dual probe type bus bar fixture with short (low inductance, low resistance) current paths then you may have a problem. The shorter lower inductance leads have very fast current rise times. If one FET (due to its lower gate threshold) turns on first then that first FET to turn on will source a large non shared current (only for a while 'til its smoke). This FET failure usually shows up as a constant 12V across the probe leads - all my FETs have failed shorted. By adding some lead length and or some inductance to the leads - you increase both R and L and put a short (few microsecond) time constant in the current rise time. Now slight differences in the gate thresholds are mitigated during the 'delay' 'til full turn on, as the other FETs catch up, and they are all on (sharing current) by the time the current peaks. We're only talking microseconds here - but it makes a difference.
Conclusion 2. Resistance and added inductance from 5' leads back to Battery+, coiled into 5 to 6 turns approx. 2-3" diameter works fine. This adds about 10usec to the current rise time.
3. Adding inductance to the power switching circuit now creates other issues. The flyback current (google it) is now much larger and creates a large voltage spike across the FET output, trying to fully shut off. This spike of a high frequency is also easily coupled into the entire circuit an may appear on the input power or output driver of the controller. In my case more than once it took out the 1407 chip - short from its output to its 12V Vcc input. With that the FETs stayed on. Not good unless you like to dangerously drill holes in the end of your 18650 cells like a plasma welder.
Conclusion 3. Two solutions here were needed to solve the issue on my setup. A.) as earlier discussed on this thread (see riba et.al. - use of a Schottky diode to dissipate the flyback spike. and B.) Don't use the 12V battery as a source for the controller circuit. I was and am still doing this BUT now I run total separate, twisted leads directly (with a fuse) back to the B+ and B- battery posts. No weld currents are common to the controller power supply current. Also try to route leads away from or perpendicular to any power leads.
One last note - helpful in my analysis and testing was to put a .2ohm (about 70A current peaks) resistor across my probe leads. Now during the various circuit tuning/oscilloscope experiments I had a repeatable, easily measured set up. I looked at both the current waveforms across the load, as well as the power input to the controller (AC coupled to focus on any noise from the pulse wiring). Now it was fairly straight forward to do A/B comparisons on various 'fixes'. During acutally spot welding I measured currents in the range of 1200 to 1400 peak amps using a separate shunt at the battery.
Hope these experiences help.
Cheers,
Geekineer