ZBasic Forum

I've looked at the Atmel datasheets, but haven't found an MTBF specification. Is this something that isn't an issue (i.e. they simply don't go bad), or perhaps it's considered an industry no-no to publish. More likely, I missed it in the datasheet.

At this time I have one Mega32 running for 2000 hours continuously with no problems, although I'd like to see 10000 before I consider the device (and the code) thoroughly tested. Of course the EEPROM write cycle limit needs to be accounted for, but for the device as a whole, should I even be worrying about it?

-Don

Perhaps there is an Industrial or Manufacturing engineer that can shed some light on this topic for us, but a few thoughts:
You can obviously greatly accelerate the testing / ageing process by putting the device through thermal cycles, power cycles, vibration, UV light, etc., compressing years into weeks(?).
If your device is going where it can not be repaired, (eg satellites, implantables, spy gear, etc), or are doing life support (ventilators, defibrillators, flight guidance, reactor control, etc.), a lengthy and well defined MTBF is critical.
For the generic consumer market it is a less important issue, both with short warranties, and with rapidly evolving technologies making old gear quickly obsolete, even if it still works, (e.g. mp3 players, cell phones).
You may well have heard the old story of Henry Ford sending people out across the country to rummage through the early junk yards, inspecting his vehicles, to see just what had failed, and what was still in great shape. The issue wasn't to see what had failed. The issue was that the other parts were over engineered for the product's useful life, and money could be saved through re-designing a LESS durable component. An interesting twist on the subject.
JC

most projects I've worked on show that the system MTBF is dominated by other than semiconductors, i.e., power supply, PCBs, connectors, battery, electromechanical things.

There is a MIL STD for prediction of MTBF but it is very pessimistic. There's another method for extrapolating. But these small microprocessors operated away from their extremes of voltage and temperature, are rarely a concern.