Mosfet paralleling reliability in real life - more smaller ones vs fewer bigger

[T3sl4co1l]

Resonant has a surprisingly wide output range, though I'm not sure how effective it is over a given range.  I haven't tested it much.  The maximum gain does depend on Q factor, and Q factor depends on loading, so the current limit must fall as the output voltage goes up (which is really to say, you're tapping a constant fraction of VARs from the tank, at a different impedance Z_L ~ Vout / Iout).  VARs would actually rise with Q, except that there's an inverter current limit (often implemented as a differentiator on tank capacitor voltage; same thing), preventing it from exploding under unloaded conditions.

What's your preoccupation with fast gate drive?  It's not like you can handle, in any meaningful way, the parasitics of a TO-247 at that kind of current and dI/dt.  You can't put a snubber any more than about 10nH away from the switch.

What are you doing, CC/CV supply, full variable range?

Tim

[mzzj]

Could be also worthwhile  to use some sort of two stage design or low/high voltage converters combined if power need is only ~1500W but 30A peak at 650v rail is needed. 
That would imply over 1:10 output voltage range  adjust-ability or sumthing like that. 

[Miyuki]

What's your preoccupation with fast gate drive?  It's not like you can handle, in any meaningful way, the parasitics of a TO-247 at that kind of current and dI/dt.  You can't put a snubber any more than about 10nH away from the switch.

What are you doing, CC/CV supply, full variable range?
Tim

With today available powerful and fast drivers like IXD_614 what will give me Fall time about 10ns with Cl 4nF and Output resistance about 200-300 mOhm

And I need to get Turn off about 20ns @ 30A, gate drive shall not be a big deal with enough negative bias
With some smd diodes to catch voltage rise from transformer and power rails at opposite sides of board to get low inductance path back to capacitors
It might work (at least in my head, it can produce big bang in lab)

Output envelope is more or less limited to constant Power, with limits at about 100A and 80V, with actual output characteristic controlled by software to anything from CC to CV and weird shapes

[T3sl4co1l]

Sure, they exist.  But that's orthogonal to my statement, it's not a counter-argument it's just another fact?

Hmm, come to think of it, as low as 10nH shouldn't be necessary, as long as the peak current really isn't much over 30A.  <60nH should be adequate.  You still have to figure out some kind of clever gate drive -- ideally, using a transistor with Kelvin drive terminals.  Otherwise, source pin ESL degenerates gain at high frequencies, even if the switching loop is tight, and even if the gate drive is extra strong.

I wonder if you'll be able to get away with Si PN diodes for the catch diodes, or need to use 1200V SiC.

If you're obsessing over speed because of efficiency, consider lowering the frequency, or using a resonant converter front-end, powering a synchronous buck (probably multi-phase) output.  The resonant could still provide pre-regulation over a modest range, but I don't know that that would really help you much; the buck still needs to handle the full range.

More bother, certainly, but more likely to meet such requirements I think.

Note that 80V and 100A is a full 8kVA operating range, so you need to dimension the inverter for that much capacity.  Even if you're only using 1.5kW of it at any given condition.  Switching loss goes as a fraction of the total 8kVA range -- all the capacitance and inductance and voltage and current swing are available whether you're using them or not.  That's why resonant is so attractive here, the switch loop reactances are stirred into the supply every cycle, no need for switching loss as such.

(There are resonant buck configurations as well, but they're similarly limited on input or output range.)

Tim

[Miyuki]

Primary current is peaking about that 30A for 100A output
I dont see any Si high voltage transistors with Kelvin lead, just SiC ones and it is not way I want, so just push drive voltage higher

I dont need maximal efficiency, just have coolable semiconductors and have transformer at temperature it wont "melt" (up to 120°C)
And keep reasonably simple control algorithm

I will run some simulations about what waveforms will look with parasitic

[splin]

Switching loss goes as a fraction of the total 8kVA range -- all the capacitance and inductance and voltage and current swing are available whether you're using them or not.

Tim

Not if you have paralled units,  interleaved or not,  powering up only as many as required for the given load. That also allows you to optimize the convertor efficiency as they only need to operate at a fixed output current,  voltage or power (apart from one of them of course). Control could be interesting though - especially for more dynamic loads.

[T3sl4co1l]

Dropping cylinders on a lightly loaded engine, nice.

Good for multi-channel (usually phased), in parallel so variable over a range of current; harder to do over a range of voltage (outputs stack in series), but can be arranged that way as well.

You could even make a grid of both, say 4 or 6 or 9 or however many channels wired back and forth like this, and remove channels that are unused due to the constant power curve!

In the large-NxM limit, each stage could be a fixed V/I converter of whatever sort (probably resonant), and it would be a two-dimensional power DAC.  Just turn channels on and off as needed.  Probably a binary series though; well, maybe not strict binary, but a unary coded n-ary may be cheaper due to parts reuse (that is, use say base 3 instead, where each digit is coded as 00, 01, 11).

And you could still, after all that, have a continuous section, which could even be linear, to fill in the remaining gaps while costing very little overall efficiency.

Probably, the all around ideal N and M are very small, due to sheer parts count increasing cost and size, versus possible efficiency savings, or efficiency losses due to repetition.

Tim

[Miyuki]

So I put it in Spice and when expecting tight layout to keep inductances low
Simulated results are promising, turn on is very soft and at turn off it is still manageable (I know real diodes will be somewhat worse than LTspice ones), secondary side should also be fine with 200V schottkyes
Transistor models are from Infineons library driven with +-15V
Even calculated snubber losses arent something huge

So I thing it worth try to build prototype

[mzzj]

Your sim model doesn't have any shared source inductance?