Ken Smith said:
But that isn't "worst case". The real worst case is that all the parts
are at the worst extreme that the maker allows to be shipped to the
customer they like the least. ie: the parts I will actually get in
production.
Fair enough, but I think that Monte Carlo analysis (assuming you set
realistic tolerances on each component's parameters) is representative of
what happens in the real world... although, OK, maybe if you're not the
'least liked' guy on the manufacturer's list!
Many simulators have a 'Yield' option that's effectively Monte Carlo
analysis where they also let you set guidelines for what a 'passing' or
'failing' circuit response is and then collect statistics about your yield.
Although obviously it would be nice if you could design circuits that always
had a 100% yield, this tends to become increasingly difficult as you
starting doing higher and higher frequency design and they are so many
variables (such as the relative premeability of your PCB's substrate) that
are difficult or merely expensive to precisely constrain. For some designs,
it's cheaper overall to accept a lower yielder with 'looser' parts than
require tighter tolerance components.
I guess my point here is that while I'd readily admit that the average
design out there probably could obtain a higher yield with no significant
change in cost, there are also times when it's entirely reasonable to accept
a lower yield just so that you can ship the @#$%@# product and get on with
life. I've heard that the IC yields on high-end 3D graphics chips are
abyssmal -- around 10% -- yet clearly there's a demand for them and it'd be
absurd to suggest that they simply shouldn't be manufactured unless the
yield could be increased.
BTW, I suspect that if you simulate any of those chips with the absolute
worst case tolerances on all the components the yield drops to 0%.
---Joel
