Dirk said:
I'm just worried about reliable starting.
We are going to employ at least 6 in each system and they must start within a mS
of each other after power on. One not starting at all would be a serious flaw.
I have found an overall no-start rate of roughly 1 in 1000 over roughly
20,000 pieces (@20mHz, PIC HS osc, 22pF ceramic caps), and it has always
been either the crystal's fault or a defective PCB. We don't measure
start-up time. Every now and then we get a cluster of maybe 1 bad
crystal in 100, but now we test them first and weed out the bad ones.
Lately (past couple of years) they've all been good. A few years ago FOX
had a bad spell, but then bounced back. More recently Raltron did too,
but less severely. Of course it could have nothing to do with the
manufacturing - maybe we are handling them improperly. Whatever the
cause, they are the weakest link by far. One thing: we have never had a
crystal go bad that passed incoming test.
I have never encountered a bad processor except one we ruined by
disconnecting the programmer mid-cycle. We have never heard of a unit
just dying in the field either (automotive app, industrial temp range
part) - there has always been an external cause such as tampering or
accident.
For your case I'd go with a central oscillator, just for the
synchronized start-up alone. Then add the increased assembly costs for
individual crystals and caps. It's also much easier to trouble-shoot.