A stuck at fault in the control logic or in the IO interface.
A single "stuck at" fault that would disable a whole chip that is completely 4-way redundand?
You are incompetent.
The above is the proof.
Hey hey, no reason to get rude here!
You should have a look at their R&D slides again. Their die contains of 4 fully self contained "quads" , 48 unrolled hash cores each. Each core must run at about 520 MHz to realize in sum 100 GH/s.
The complete die will have a size of at least 100 mm2. It's 28nm, not 130nm, there will be yield issue for sure for such a huge die. Most likely this will be stuck-at faults in some of the hash cores (which could be easily detected by a BIST or ATPG tests). If these permanent production failures cause permanent non-working hash cores, the remaining cores have to run at a higher frequency to still have 100 GH/s per chip. Because this is what they have to sell.
If 3 of the 4 quads have stuck-at faults in the interface logic, it would be impossible to realize 100 GH/s with the remaining one quad, because in this case it has to run at about 2 GHz.
Of course one could try to implement each quad to be able to run at 2 GHz speed too, but this kind of "margin" will probably double the area you need for one hash core. I doubt that they have added enough margin to compensate even one complete non-working quad to still have 100 GH/s.
The skipped traditional production tests will force them to realize all this compensation on-line in the final miner by firmware procedures. This it not impossible, but the development of that feature is not that straight forward, because they currently do not know how a higher number of chips will behave in terms of yield and how these procedures have to look in detail.