That makes a lot of sense and works quite well. Since this causes the "high" rail to be greater than 12V as the resistor discharges the cap, I'm seeing current flowing back in to the source (PSU DUT). Should I be concerned about this?
It's always and necessarily less than was drawn in the first place.
In a truly extreme case, perhaps it could be a concern? Like, if you had some henries to charge up, some monster solenoid or something, or maybe a superconducting magnet, in which case the clamp capacitor needs to also be truly massive, and the supply would be drawn upwards pretty well (and if it has downprogramming, it might well not handle enough current, or power dissipation. But that's still rather contrived, because the sheer scale of energy storage is obvious from the application.
For which, again... the correct answer is to dump it into a load, perhaps a resistor, perhaps a stack of TVS, anything that can dissipate the heat. Or for the superconductor, you might just let it quench (vaporize all the cryo refrigerant).
The point that I find most convincing is that:
1. The MOSFET is the source of the transient, and this is true from the shortest time scales (read: the transmission line between MOSFET and load), to the longest (as above);
2. This is the only node that matters for protective purposes (MOSFET overvoltage causes failure), and
3. It's again scale invariant, in terms of energy: whether you need to dump 1uJ (as in a switching snubber) or 1MJ (as above), this is the correct place to do so.
And, again, and as you already know, or have discovered from the simulation -- it's not enough to clamp the wirewound resistor, there are other inductances in the loop. It's not enough to clamp the bulk of the inductance (e.g. a relay coil) at its terminals, because there may be wiring inbetween as well. And so on.
Clamping a bulk inductive load, may still be
sufficient for a particular application: say your current level is low enough, or switching rate slow enough, or voltage rating high enough (or avalanche rating high enough, or other considerations related to that), to deal with up to some amount of wiring inductance. But it cannot apply in general, for arbitrary and unknown amounts of wiring, or if someone decides to add extra inductance just because (maybe they thought it was noisy and needed filtering in the worst way possible?..), or if someone forgot the clamp device, or its connection loosened over time (or rusted off or something). Exactly zero cases of which, are left uncovered by the clamped switch arrangement.
The main reason you would employ something different, is if the repeat rate is high enough, or energy must be conserved. In that case, your best option is a boost converter motif, for which the switch-diode-capacitor loop must be minimal inductance. The boosted rail can then at least be bled back into the main supply (RCD clamp snubber), which saves on dissipating the extra "+ Vin" that a clamp to GND would otherwise have to dissipate. Or if you want to go to the trouble, a second converter could be used to return that energy back to the original supply entirely, ultimately only costing the converter's efficiency. Or if the energy to be snubbed in the first place, is rather well known (as can be the case for switching converters), there are quasi-resonant / "lossless" snubber networks (using L, C and D) that can return that energy automatically (by the switch's action alone).
I understand that the arguments
I find most convincing, may be rather high level / abstract, so I don't know how well you will appreciate them.
Or worse? still: at least in principle, I prefer that people understand
why, and when, something should be done some way -- not to take a single-case solution as wrote, which will probably lead to misapplication later. But that can also take
years to learn, for deeper topics.
Hmm. Perhaps this is why it takes me so long to do everything...
Out of curiosity, what part of my plan screams "fail" the most?
You seem to be aware that your grounds won't all be. But you're indicating making loops of them anyway, and worse still, making loops across low-level circuitry, inviting all manner of ill behavior: from measurement error to oscillation, perhaps destruction I don't know.
Maybe I'm jumping to conclusions there, but it felt worth adding the differential suggestion at least. Proper grounding takes much more to explain, unfortunately.
For any future readers, I measured the following inductances:
- 0.1R 300W Wire Wound (Tubular) Resistor - 0.836uH
- 1R 300W Wire Wound (Tubular) Resistor - 5.382uH
- 0.25R 1000W Wire Wound (Tubular) Resistor - 16.134uH
- 0.1R 100W Wire Wound (Aluminum Housed, Heatsink required) - 55nH
Measured how? Mind that a random meter at unknown frequency might be reading the resistance more than the inductance (magnitude impedance versus reactance, or parallel vs. series equivalent inductance, or at a mixture of harmonics).
Tim