what I don't get is why anyone so experienced would build such a massive chip for something like btc mining. With big chips you get worse yields, usually lower clock potential and more thermal issues. I can only really think of getting slightly better performance/area from going big but then you might not gain anything due to smaller chips normally clocking higher/better yield.
A bitcoin chip does not suffer from nearly the same level of yield issues that regular ICs do. With regular ICs if there is the slightest imperfection anywhere on the chip, then the whole chip becomes useless. E.G. A CPU that executes one particular instruction wrong is pretty much useless to anyone. A bitcoin chip is composed of many replicas of a single mining core or engine. A tiny defect is very likely to only affect a single core or engine, so reduce the chips performance by only a tiny fraction.
In a normal chip it's necessary to build a clock tree to maintain clock synchronization over much of the chip. This is because there are lots of dependencies and connections between logic in one part of the chip and logic somewhere else. In a bitcoin chip each engine or core is mostly independent from the others, so the clock only needs to be synchronous within one engine - much smaller than the whole chip. Effectively it's lots of small chips in a big chip.
What's important in designing and optimizing a bitcoin chip to produce this sort of result is a very different set of skills from what's important in the normal process of SOC design (system on a chip design), and takes a highly experienced team.
+1
. Might help to quell suspicions and gather more interest closer to product launch.
