still here.
found lots of interesting things and took lots of cool photos only to find that i was measuring the resistive voltage drop across the other oscope probe.
honestly this is not very interesting and the results are 100% predictable.
I was not able to blow the fet up (as in i haven't yet, its an irfz46).
the pwm resistor is twisted pair as yesterday, but this time its 0.4 ohms instead of 0.1
the diode is a 20amp 100v Schottky same as before.
as you may have figured out, the fet source is grounded, the flyback diode is connected across the resistor, and the supply side inductance which is simulated by a 3uH inductor is buffered with 4.2 uF of mylar caps.
duty cycle is 1/16, average current is 1.2 amp right now at 16 volts.
because the resistor is non inductive, the frequency does not matter, it happens to be 10khz right now, at 100 khz the current consumed by the device drops about 10-20% as expected due to switching losses.
Here is where a few variables come in. those two diodes in series allow the supply inductor to charge up the 4.2uF caps to approximately triple the supply voltage under load, double it under no load. this is not a resonant voltage doubling but a simple energy in the inductor becomes voltage in the cap, minus losses. as the board is shown below, within one uS the fet turns on, in the next 5 microseconds the capacitors on the board are discharged to zero from 34 volts. (34v/5us *4uF =27 amps peak).
As the supply inductor current immediately increase these capacitors charge up from zero to 10 volts with about 25% overshoot in the next 10uS or so.
Current and voltage is stable until the switch is turned offm at about 20-24 amps and 10 volts.
At turn off the supply side inductor charges the 4.2uF cap from 10 to 34 volts in 6us, and, and drops back to 32 at 8-9uS after turn off as the in5401 diodes recover.
(linear for the first 3.8us: 20v/3.8us*4.2uF=22a)
Because the duty cycle is so low, a common dmm will read this peak voltage accurately enough to trust. (reading 32 volts, peak is 34, duty cycle is 6.3%, close enough)
the 1n5401 diodes dissipate a lot of heat for some reason, or perhaps 1 watt is actually a lot of heat.
(the fet on the other hand, dissipates about half a watt for conduction, and perhaps a watt for switching losses at 10khz with single digit us switching time)
I attempted to use larger electrolytic's to absorb the current and keep the voltage down.. the problem is they tend to explode. (without the supply side diode)
The high distributed esr and inductance of an electrolytic cap ensures that the voltage at the cap doesn't ring, and it is quite easy to hold down, they just get hot.
If you get rid of the diode feeding the circuit shown in the photo, when the current falls to zero and the capacitor charges, it will oscillate with the supply side capacitance until the entirety of that energy is dissipated in heat, it doesn't just disappear, it is rather easy to calculate the energy dissipated. assuming the current in the resistor is discontinuous (which in my case it is provided the switch is off for at least 6uS) then it is simply one half supply side inductance times I peak squared (times frequency to get watts)
with a real heatsink this fet would be good to 3 amps average current at 1/16th duty cycle. I tested it to 2-3 amps average for a few seconds, but within 30 seconds it would probably have melted the solder
With a bank of electrolytics to buffer the voltage, holding it down to an unmeasured voltage but with nearly no overshoot, i pulsed it at 5 amps average and the same 1/16th duty cycle, it survived just fine, this was without the supply side diode.
End result is just you'll save yourself a lot of headache just building a delay line with a string of cmos inverters and resistors and caps, feeding these signals into (n) comparators and feed the control voltage into all (n) comparators through (n) resistors with a bit of hysteresis requiring (2n) resistors.. you can easily get (n) phases with just a handful of parts, each from 0-100% duty cycle.
with careful layout there would be no need for electrolytic caps, and the added expense is more than made up for the fact that there's no non linearity between amp hours dumped and "average" current, or any other weird stuff.
i'm not going to dick around with an MOV across the fet, but if you keep the frequency low enough that the mov don't heat up, that's entirely acceptable.
also, you should be able to predict how hot the fet gets.. if its hotter, you need to look into what's going on.
if switching is slow, on the order of 1KHZ or so, 10uS switching can be done with 50 cents in parts, and you won't notice any extra heat compared to 100ns switching which requires 2$ gate drivers.
if you have 10uH inductance on the supply side (have you measured it?) at 1 khz and 50 amps that's 12 watts... that heat goes somewhere, if you have 200 volt fets and 130v mov's, it goes into the movs
from what little i've looked into the subject, repetitive avalanche power is "typically" 1/2500 max rated thermal dissipation.... if you insist on just want to "Make a bet that it won't be a problem"