Finally I can put some take into this. I’ve worked in memory testing for years and I’ll tell you that it’s actually pretty expected for a memory cell to fail after some time. So much so that what we typically do is build in redundancy into the memory cells. We add more memory cells than we might activate at any given time. When shit goes awry, we can reprogram the memory controller to remap the used memory cells so that the bad cells are mapped out and unused ones are mapped in. We don’t probe memory cells typically unless we’re doing some type of in depth failure analysis. usually we just run a series of algorithms that test each cell and identify which ones aren’t responding correctly, then map those out.
None of this is to diminish the engineering challenges that they faced, just to help give an appreciation for the technical mechanisms we’ve improved over the last few decades