Between the supply and preamp I mean.
1MHz bandwidth seems to imply you can put a filter in there, and not need to clamp things for at least some hundreds of nanoseconds. And the filter will necessarily introduce impedance, making the inrush current well-defined. But it also needs to be terminated and a zero-impedance source doesn't exactly work with that (termination can be on the amp side, but the source impedance still preferably needs equalization -- a constant resistance filter is best).
Yeah good point. The preamp should achieve a noise floor of about 300pV/rtHz and needs to measure noise down to about 2nV/rtHz for power supplies and lower for e.g., transistor noise measurements. That doesn't leave a ton of margin for additional noise sources as I'd like to stay far enough below the DUT noise to maintain accuracy. There are other tricks I'll emply here; mainly, two preamps and cross correlation but that imposes its own challenges and has its own drawbacks so I'd like to get the preamp noise as low as possible first.
The protection circuit will be able to be switched out for certain measurements such as for 0-DC sources where it's really not needed. Such a protection circuit is probably also overkill for biasing arrangements used in transistor noise measurements. Anyway, maybe there's some headroom for adding a filter with a small series resistance if this protection circuit will be exclusively used for power supply and voltage reference measurements. I'll investigate; thanks for the idea.
Also if the source is always zero ohms, a JFET doesn't seem the lowest noise option; a nice thick BJT should be able to do better than a JFET array. Lowest noise may not be the top priority however, so that's fine.
This isn't necessarily true. While the DUT might be low impedance, the blocking cap at the lower frequency range of operation isn't. I specified <1Hz as the cutoff, which I appreciate is a bit vague (this should go down to at least 100mHz, preferably all the way down to 1mHz). But, using 1Hz as the frequency that 10u blocking cap looks like a 16k impedance. The BJT current noise through that cap will swamp the equivalent input noise. JFETs do much better in this regard. And, while you're right that JFETs don't do quite as well as BJTs for true low impedance sources, they can do remarkably well and are really not all that far off the performance of BJTs from what I've seen. I could, of course, increase the blocking capacitance, but this quickly becomes impractical.
Well your first claim was hotplugging so if that's the only thing you're worried about, that could be worked around in many ways. You didn't say it needed to do anything else, so we're free to assume anything and everything until said otherwise...
If it's not the only thing, maybe you could make a hotpluggable interface module and only use it to test supplies with that? More steps, but makes things more flexible / foolproof.
No idea; we're not telepathic here, after all.
I apologize for being vague - that really wasn't my goal. And I do appreciate the suggestions I've received.
As for hotplugging modules, I guess I just don't know enough about those so I'll spend some more time looking into it. The commercial hotplugging modules I've seen have all been for a single polarity supply and for a limited voltage range (makes sense, given the typical application). I suppose the principles underlying these modules could be used to make a custom one. But, looking at the ADM1177 for instance, doesn't make me think that my current topology is wrong. They use a single external MOSFET whereas I need a bidirectional switch. I monitor the voltage and trigger the shutoff with high-speed comparators, whereas they use a sense resistor and ADC. I like my approach better, since it's simpler, should be faster and doesn't require adding a sense resistor.