My opinion on capacitors:
- Definitely avoid electrolytics
- Tantalums are, at *best*, dubious.
- Ceramics don't die, but they do vary with temp, voltage and age.
I have some old equipment with them and none are toast (my Tek 475, whereas the electrolytics are all dried up), but that's just one case.
The main hazards are excessive voltage, and current spikes; they will be more reliable under conditions where this is controlled. For instance, using 16 to 25V rated caps on a 5V rail, and supplying each subcircuit from a current limiter (since a small bypass cap looks like a grain of sand against a heavy source, even if that source is current limited). Tantalums have also been made with internal fuses: of course, you need to accommodate failure in your design, which begs the question, why did you bother installing it in the first place?
- Ceramics are more-or-less forever, but they do change. A typical X7R (and I won't even consider any worse grade) is +/-10% at room temperature, but can be -20% at rated temperature (I think??) and -50% at rated voltage. The derating is therefore similar to tantalums: pick 3x more voltage rating than you need. High-Q ceramics also age, where the value simply drops over time, while under polarization I think. I forget if this is included in the rated tolerance, or if it's additional as well. (Drops of -50% aren't uncommon, but I don't recall if that's Z5U or what.) So the challenge is, desigining your circuit to accommodate a wide range of capacitance while meeting guaranteed performance.
Gold standard would be using C0G where possible (essentially an ideal capacitor at most frequencies, and AFAIK, stable under all conditions), of course, these are bulky and expensive. Definitely worth considering for the smaller timing and signal filtering components (say, anything up to
10n, maybe even 100n).
- Resistors: carbon composition can age a bit, especially under heavy load, but with those pretty much history these days, that's not a problem. :) I don't know of any issues otherwise. General advice applies, don't overheat them (as much for their own sake as for the sake of stuff nearby).
- Generic silicon thoughts: I don't know that modern molding materials (i.e., since, say, the 60s or 70s?) are a problem (at least over here? :) ). Plenty of equipment survives from those days, including power amplifiers, for instance, that see wide temperature swings.
- BGAs: I don't think these are actually a big problem. If leaded solder is used (reball if necessary?), cracking balls isn't a problem. The chip can also be underfilled with resin, encapsulating the balls and gluing the chip down. Pains could also be taken to minimize flexural stress on the chip and board (a good idea around any large or brittle device), say with stiffening frames or strategic routes through the board.
With the amount of consumer stuff since RoHS, you'd think we'd have seen a lot more examples of tin whiskers -- apparently it was all hot air, and the processes turned out much more reliable than anyone expected. Yes there have been notorious cases of cracked balls: Xbox's Red Ring of Death for one, but that's a thermal issue at its root. It's characteristic of the process, but not one that is commonly seen under normal operation (of suitable vibration and temperature conditions).
I also heard QFPs can be more failure prone, I guess because they have so damn many leads and not much solder to hold them down? Only example I have is a computer from 1987, which contains PLCC and TQFP gate arrays, but old logic like that never sees strain or temperature cycling, so it's a bad example.
- Wiring -- do what the aerospace people do: use teflon wire, and lots of ties. I suppose I wouldn't mind PVC wire myself, but it probably will go brittle after long enough. Of course, you can't clamp or pull teflon too tightly either, because it cold-flows! Vibration and stress is the killer on connections, so keeping all that neatly secured will go a long way.
Now, all of that said -- plenty of consumer electronics have demonstrated a life time over half a century, at not much above ambient temperature -- but guaranteeing an MTBF of the same isn't so easy. The best approach is going to be finding mil-spec components, ceramic body packages where possible, and doing what the avionics people do, massive loads of ceramics to replace electrolytics. Gold plated everything is quite typical. Mil spec doesn't seem to have any problems with FR4 (heck, even old consumer phenolic from the toob days remains mostly intact, and there's no kidding about temperature cycling there). Controlling temperature is the other killer; excessive heatsinking isn't a bad thing!
Tim