How do you even begin to think about testing hardware like that for correctness?
I remember my days as an intern at Intel. I remember someone using genetic algorithms to try to construct a test suite for some chip. But it was no where near that transistor count.
And for an extra measure of automation there are test suites that are programmed into hardware that can drive chips at line speed on a test bench.
From my understanding: Binary decision diagrams on supercomputers. (https://en.wikipedia.org/wiki/Binary_decision_diagram)
Verification hardware is a major business for all these major CPU players today. Fortunately, solvers on this NP-complete problem have benefited not only from faster computers, but huge advancements in algorithmic improvements in the past 20 years.
Basically chain all the logic together in a special test mode.
Analog blocks get tests written specifically for them and either tested via external test hardware or linked internally.
For example, if you have a DAC and and ADC, provide a link internally and drive the ADC from the DAC to test both. You can also test comparators etc from using a combination of the DAC and ADC, trim bandgaps etc.
If you’re real smart, you do this at probe (wafer), massively in parallel.
Both are very different and very interesting on their own. For the design, verification can be done by computers and it's much more powerful than testing anyway. For the hardware, I imagine there are cores inside the chip with the sole task of testing sub-circuits (fab defects are much less varied than design flaws), but I stopped following that stuff at the era of single billion transistors.
[a]: My knowledge of creating hardware is very limited. So I could be completely wrong.
I'm not super familiar with it, but one day perhaps hardware verification becomes as easy as software testing someday?
Could you expand a bit on how that worked? Seems like an interesting application.