Design for long term reliability?

Question

I need to design a small board that will go into a large piece of public infrastructure intended to last many decades. I am looking for papers and the like that give guidance on such design based on real research.

This board will be much bigger for mechanical reasons that it needs to be for even a spacious circuit to achieve the function with discrete parts. Things like wide traces is a no-brainer.

The customer wants to minimize total parts, and wants them to be thru hole. I see the point about minimizing parts, but which parts also matters a lot, and being able to get replacements in the future is important. This function can be implemented with a handful of discrete transistors and resistors, but the customer would rather use a single logic IC in DIP package. He thinks thru hole is more reliable, but I think I remember seeing a study that says the opposite. Also, I'm worried about availability of a 16 or 20 pin DIP logic chip in 20-50 years. But, are SOT-23 transistors and 0805 resistors a better bet? There will be some opto-isolators. It seems to me those will swamp everything else in terms of reliability and future availability. Yes, I'll run the LEDs at a small fraction of the rating to increase life.

So, I'm looking for real definitive research-based information on designing for long-term reliability. This is an area where it's easy to think about the 10% problem but miss the 90% problem that makes the 10% issue irrelevant.

Added:

I'm looking for evidence-based answers. I like to think I know electronics pretty well, and can come up with various plausible-sounding reasons why one approach may be better than another, and I'm sure others can too. However, I don't trust those because what sounds plausible and is based on sound physics may be correct but missing some other more dominant affect. I'm worried that this is where educated guessing could lead to significantly wrong conclusions. That's why I'm asking for evidence-based answers, papers from actual studies, rules NASA might insist on, etc.

Added 2:

Consider the environment "industrial". I'm not sure how well the environment is controlled if at all. The boards will be protected from the elements, but possibly no air conditioning or heating. I don't know about vibration, probably not much.

These boards will be installed in a cabinet that houses other parts of the electrical system. Service technicians can walk up to the cabinet when necessary. Difficulty of servicing isn't the issue, but downtime is. This is not what's going on, but imagine if a interstate highway was shut down until the system is up and running again. Of course there is redundancy already, but failure is something you really want to avoid.

Answer 1

NASA has a lot to say about long term reliability of electronics. Here's one example -> https://nepp.nasa.gov/files/20223/09_109_1%20JPL_Spence%20Longterm%20Reliability%20of%20Hand%20Soldering%20M55365%20Ta%20Capacitors%2009_30%2011_09%203_2_10.pdf one example (references are at the end).

I can't give you good link to everything related (NASA web site is quite messy), however, googling 'nasa long term reliability electronics' gives a lot of links to papers on the topic.

Answer 2

I will add to this answer what I know.

To begin with "creep corosion", you have an investigation here

It is actually related to enviroments containing sulfur. It's worth a read if nothing else it's an interesting topic.

There are a lot of articles related to ROHS and tin whiskers from NASA, links.

Another thing to consider is the FR4 material itself and CAFing. This is not a study, but it ilustrates the issue.

About the reliability of SMD, a study was conducted in 1993 and there are some interesting letters in the appendix. Link.

For capacitors I would say to go with ceramic MLCC, here is a comparison between precious-metal-electrode and base-metal-electrode. Included is a table with tested units.

For ceramics there are capacitor designs that have a "soft electrode" and the ones that are more likelly to fail in "open mode". Generaly speaking you want to get parts that are at least automotive qualified.

According to The capacitor handbook (Cletus J. Kaiser) the glass capacitors are the most reliable, and I remember NASA used them. I did not found reliability data yet.

Try this for reliability data. Also for other capacitor types.

Answer 3

My answer is not based on real research, but rather on real application. I recommend that you use the most reliable components and and create a board with them. Determine its MTBF. Based on this MTBF, assemble enough boards to cover the total time this design is supposed to last, and double that number. For example if the MTBF is 10 yrs, and the time the design is supposed to last is 50 yrs, then you need to make 10 boards.
To minimize the "down" time, a set of "switches" can be automatically activated to disconnect the bad board and connect a good board in its place. The bad board can then be replaced with a good one and be ready for the next board failure. You will not need to worry about repair parts not being available - you already have them!