This question is a follow up to Can chips be sensitive to mechanical stress after an electrical incident?. It turns out that our problem was nothing to do with failed chips. It was the PCBs all along.
Background: We had a number of flexi-rigid PCBs fabricated and assembled by a company in the UK. We had several different designs of board made, but all had the same stackup: 6-layers of rigid at 0.8mm, and 4-layer flex.
When the boards arrived and we tested them, we discovered a couple of failed vias, which we fixed with jumper wires. Everything else seemed fine, and we put these boards into a longer term test. After more than 100 hours of testing at 70ºC with some mild vibration, we started to see some off behaviour from one of the microcontrollers on one of the boards. To cut a long story short, it turned out that four vias had failed under a QFN package, entirely cutting off that chip's ground!
Question: How common is it that PCBs arrive with A) totally unconnected vias, or B) vias that fail after some time? Should we be highly suspicious of vias in general, or is this a freak manufacturing accident?