I imagine that research teams encountered lots of write errors, while they tried to track down anomalies that cause them, then construct workarounds (example, routines to verify that reads and writes are correct). Somewhere there ought to be a few real-life stories.
Thinking about typical electronic problems in any fast-acting circuit...
* There might be glitches on internal power lines due to switching between banks of memory, etc.
* Accidental interactions during a refresh routine, etc.
* Magnetic and inductive interference from nearby devices, speakers, hard-drives, etc.
Another possibility: cosmic rays have been speculated to be an occasional cause of glitches in micro-circuitry. It's the sort of thing that's hard to verify if or when it happens. It can't absolutely be ruled out.