Is there a problem associated with the microcracking?

TTFN

With load and temperature cycling the cracks tend to grow leading to complete device failure.

Then, you're going to have to trade off some life to weed out the bad parts or beat up the supplier.  They simply should not be supplying parts with microcracks.  It's absurdly expensive to facilitize for weeding out something that shouldn't even be there.

If you're insistent on testing, then you'll need to run an accelerated life/thermal cycle and measure the appropriate indicator, presumably the leakage current.

This would most likely chew up a sizeable number of otherwise good parts to refine the test process.  You'll most likely also cause premature failure on parts that would have otherwise been fine in normal service.

TTFN

I have never tried this, but read about it for finding cracks in PCB traces.

Inject a high frequency signal on the trace and run a scope probe along the trace. The signal supposedly attenuates or alters in some way down stream of the crack.

Maybe some permutation on that will work.

Thanks Kontiki99,
I will keep the idea in mind, but it is not practical in this case since the diode is about 3/8 inch square and sealed inside a metal hermetic housing.

IR stuff,
I know we shouldn't have to do this much sorting, however there are not many manufacturers that make big power diodes in metal cases anymore, and even fewer we have not had problems with.  We are working with the supplyer and attempting to qualify new suppliers to meet our needs.  As a matter of fact the thermal cycle testing you mention is part of our qualification process and is done on a production basis for a couple of products, so far with no decrease in product life.

in the mixer diode quad demage by ESD the noise figure measurement is used to determin if the diodes are in tack of not. not sure that will apply to high power diodes.

kontiki99
Your approch might have sense, hoever I will use reflectometry or reflection in time domain. You shoud see some dicontinuity in the time /distance from the source.
So some network analyzer with time domain could be used for that.
You will used good unit as a standart how the time domain signal should look like and compare that to demage component. The demage component will propably have a lot of more spiks a dips then the good one.