We experienced occasional problems - usually corruption of stored parameters in FRAM (bit-banged I2C interface) - but never 8052 appl ROM corruption (Winbond with 7705 or MAX supervisor). It was hopeless trying to do objective testing by flicking the power switch (or with a pair of timer relays, as devised by our test department wizards). In the end, I lashed up a 5v supply via a beefy emitter follower - base driven from my Wavetek function generator. Hence I could supply triangular waveforms of varying rise and fall times and see where the critical timing occured.

Eventually found the problem, but it was very obscure, and relied on a quite narrow dV/dt band. I can't remember the details now. We'd have never found it by flicking power on and off.

Quite a useful and easily rigged test setup.

Dave