[theamk]

I think it's important to differentiate "a mismatch between two abstractions" and "hardware bug". Because you can frame any sort of hardware (or even software!) problem like this:

"Capacitor plague of 2000 was a mismatch between two abstractions: the abstraction that capacitor actually provides datasheet-described amount of capacitance vs the reality of the system"

"Toyota unintended acceleration was a mismatch between two abstractions: the abstraction that ECU properly responds to accelerator pedal release vs the reality of the system"

Yes, digital systems are made of analog parts, but that's not a reason to accept systems behaving out of spec. For the last 50 years, the specifications for RAM have been pretty clear: as long as all datasheet requirements are obeyed, the only way to change stored data in one location should be to do a write to that location. If a memory chip does not act according to its own datasheet, it's not a "leaky abstraction", it's a hardware bug.

(Now, can this be fixed economically? I don't know, I could believe the answer is "no". However, the solution in this case is not software workarounds, but rather to make a new spec: "RH-RAM is like regular RAM but cannot tolerate certain access pattern")

[xgk]

I think if you (row)hammer hard enough, every DRAM will eventually flip a bit.

[userbinator]

Read the original Rowhammer paper where they tested various manufacturers and years --- this only started showing up around 2009, and DRAM from before that time was entirely immune to it.

[xgk]

Sorry, I should have said: ... (row)hammer hard enough, every sufficiently dense/modern DRAM ...