For sufficiently low duty cycle, can you get away with driving SCR above rated current as long as A^2 * s are maintained?

Question

I'm wondering if the A^2-s rating of an SCR is what ultimately decides whether a given current for a given time is okay, even if the data sheet does not list such a number.

For instance, right now I am looking at a device that accepts an RMS on state current of 30 A. This device can obviously accept (30 A)^2 * (1 s) = 900 A^2-s. So, can I most likely get away with higher currents so long as pulse width < (1 / (current^2))? In this case, if pulse width T = 1 us, that suggests that 8 kA is okay because (8 kA)^2 * 1E-6 = 64 A^2-s < 90. Is that how that works?

Answer 1

There are approximately three limits you need to stay within:

1. Thermal -- junction temperature must remain within ratings. Note you can't measure Tj while operating (except as inferred from terminal voltage and current, which aren't well enough specified to do this, you'd need to calibrate per part first), and Tj is NOT case temperature so you can't measure it externally -- there is temperature drop and delay between them.
2. Physical -- you must avoid effects like electromigration, hotspot formation, etc. I'm not sure what all mechanisms apply to SCRs in pulsed operation, but likely there are some things like this. These may be one-shot (exceed it and it blows up then and there) or repetitive in nature (e.g. electromigration can be pushed if low lifetime is acceptable). These are rarely if ever specified, unfortunately.
3. Rate -- if nothing else, current can only increase at less than the maximum dI/dt. If other limits pull in during that rising edge, well, the edge needs to stop then and there. The ultimate ratings (defined by the other limits) could perhaps allow a much taller square pulse, but this limit prevents a pulse from being too "square". And at the time scales you're asking about, an exponential decay (with or without ringing) is about all you're going to do.

The first limit is more or less described by surge or fusing ratings, and directly by thermal impedance if given. Surge × number of cycles is more or less equivalent, as well (for the >8.3ms range). Note that these ratings only provide discrete points on the thermal response -- i.e. 8.3ms half-sine pulse, which doesn't tell you anything about what's happening on the say 10µs time scale.

The second limit is sometimes described by a short-term pulsed rating. I don't recall if I've ever seen this on an SCR, but it is occasionally provided on regular rectifiers (say for a 1ms square pulse, or 10µs, or whatever). Again, doesn't tell you anything about time scales far away from this, but is at least a point on a continuum, so tells you a little about nearby time scales.

Regarding scales: heat spreads by diffusion, giving a P ∼ 1/sqrt(t) and E ∼ sqrt(t) characteristic. Time scales of 10s µs involve heat flow through the die itself; up to some ms, heat flow from die into metal tab / substrate; and 10s ms to 1s+, heat flow into heatsink, PCB and eventually the ambient. Within each regime, it's reasonable to expect the diffusion scaling, but between regimes, the thermal impedance can be higher or lower than expected, which is to say, don't try too hard to extrapolate from any given point.

One might rightfully wonder whether the fusing rating should actually be more like I²√t. Well, fuses are closer to I²t in behavior, so they're specified that way; and they open within some ms, so the scaling error can be tolerated. Again, I wouldn't try to extrapolate from this rating, at least not very far.

Of course, say if you only need one pulse ever, you may well be satisfied that the device fails shorted (or maybe it arc-flashes internally, but is still conductive enough to do the job), having switched that one pulse exactly as commanded.

As mentioned in the comments, the datasheet only specifies what it specifies; anything beyond that, you're on your own.

To be clear, you're more or less on your own anyway; datasheets aren't legal documents, and at best you could make a case for, I suppose, something like implied sales contract, or maybe false advertising, and basically refund the cost of parts you paid for (or a bit more as a class action perhaps, but that would be really out there) -- not counting legal fees of course, which would make such a plan utterly ridiculous.

Your best recourse is simply to contact the manufacturer and ask them about what you want to do. They may have documentation available, whether public that you didn't locate yourself, or private (perhaps subject to an NDA?).

Or you can do qualification testing in-house, but this is labor intensive, and needs to be done at least on a statistical sampling basis, as the manufacturer may change their design from time to time and they are under no expectation / obligation to preserve parameters not on the datasheet.

Which is really what it comes down to: the further you stray from the datasheet, the less confidence you have in the ratings you've extrapolated, and the less reliability may be. Various effects have different scaling factors: some proportional, some quadratic, some exponential!