My thanks to everyone who read or replied.
These EEPROMs are not flash, at the time, 1991, the only choice was either the 93 series or 24 series EEPROMs. I just happened to use the 93C series, looked easier to use when an 8 bit port is used for different things and needed to get the ins, outs and tristates correct!
The EEPROM isn't written except during the calibration sequence, but it is read several times per second to extract lookup table data. When calibrated it is powered from the mains and even if there was a power loss then would have to start the whole procedure anew.
At the time 3V devices weren't around, but when they arrived I did use the dual voltage 93C46 until failures in the field caused a reversion to just the 5V only version. As time progressed these became harder to source, had to buy end of line and other manufacturers parts. The 3V failure was due to the bus lines detecting a spurious write during power on. Various pull up/down resistors have always been used but not using address 0 was quickly implemented.
I have no control over where the units were placed but sure many were in front of a window.
The other big change over the years is the change from transformer based to switch mode wall warts. I have discovered that the switch mode type can put 80V pk-pk signal on the output, only at 1 or 2 uA, but enough to upset a battery powered unit when everything else is off to save battery life. Some used battery, some had a mains power supply as an option, but the software still did the power saving.
Over the 30 odd years I have sold many thousands of units, just that three identical failures in such a short period raises questions.