I think, like Erik, that there's a massive difference between the original '51, which had a dynamically working logic and the todays '51 with their static logic. You are surely right, that the performance is unpredictable at very low supply voltages arround 1V. But nothing bad will happen, because the internally fabricated flash burning voltage (which needs to be arround 12V, as you told) is much too low to cause any code memory corruption.

Also, after reading this document

http://www.atmel.com/dyn/resou...oc4183.pdf

I tend to believe, that things become fine again, when the supply voltage rises again. I think, that with statically working '51 the reset does more than a simple NMI. It completely resets all the internal circuitry and the micro starts as a fresh one at the moment, when the external reset chip releases the reset line when reaching nominal supply voltage. It should be easy to implement such a fundamental reset when using static logic and the manufacturers must be total idiots when not having realized that with their flash micros.

From my experience I know, that since we use proper reset chips, like MAX1232, we never observed any flash code memory corruption (knock on wood...). We test each product with very slowly rising and falling supply voltage and check the integrity of reset signal and absence of ALE down to Vcc=0V.

Of course, we use multilayer boards and proper supply voltage filtering...

Kai