| ??? 01/17/07 20:47 Read: times |
#131003 - do you want to do the manufacturer's job? Responding to: ???'s previous message |
Richard Erlacher said:
What??? Do you mean that a part with internal malfunctions are OK so long as, initially, they don't impact your application? Well, you know, I've never been in other than "100 in a month or less" situation - the economy of this doesn't allow me to think otherwise. It's not the zillions of parts on the market that I asked about, though. It's just the 805x system. What's 805x system. I still can imagine - OK, maybe just half a zillion - types of parts connected to 805x. Let's start with resistors, capacitors... Even for a very simple circuit, just the basics - power supply, '51, reset, MAX232, and "those little grains" holding it together - I can imagine of thousands of modes of failure, some of them fatal, some not. Just one example for many. Our PCB house populated a set of boards with electrolythics with swapped polarity. Would you think of this when designing your "bulletproof" test? How would you screen for it? It influenced the board performance dramatically - real smoke came out, but not immediately, after several tens of hours of operation at elevated temperature (inside a box). And what about an improperly manufactured DIP switch, which developed an intermittent contact during burn-in? I think you can tell me much more of these stories. "Smoke test" is for devices that were built on the kitchen table and are going no further. OK. It's time to define smoke test. For me, it's what happens when you power the device up the first time. There is plenty of smoke sources. There are a few devices where the power can be switched on "gradually" in some way, but that's not the typical case I think. Real testing involves thinking about failure modes and stressing the system under test, within prespecified limits. This may as well read as "try to wiggle with every button on it". Why do you think this is not an adequate test for a given application? Why do you think the ICs are more prone to initial failure than the "buttons"? Why do you think this might not be a full input vector of stimuli for a given device? Why would you expect a uniform approach to manufacturing testing across the whole spectrum of possible electronic devices? I'm including in my question, all the boards that are being built 1000 per day, for the past ten years, into which established components are being used, and not "new" systems never-before-tested. Oh, this changes the economics a bit, against the "less than 100 in a month", don't you think? One thing I'm curious about is how people determine whether their ATMEL or other mfg's parts actually work as specified.
My own experience indicates that, at least with ATMEL, there's room for doubt. This one is simple. You have your doubts, you don't use ATMEL's. I rely on the manufacturer's procedures. That's simply all. |
