Perhaps the epoxy is poorly cured and the bake makes it "freeze"
anything mobile when it finally does cure?
Might try, if this is consistently happening, pre-baking the diode
(quad) before assembly to test that one out. If it's true, field returns.
Including the ones already ruined.
Perhaps voltage or the leakage power initially, also accelerates some
sort of "curing". I don't usually deal with plastic packaging (you may
come around to the same view) and sounds like this packaging may
be inferior to standard mold compound (which can be had in super
low moisture absorbtion grade).
Have you searched for reliability related info on this part type and
manufacturer? Looking at the datasheet, is the package drawing one
that you can maybe also research, find source vendor and qualification,
reliability and angry-customer data?
Now you don't say quantitatively, just what "low resistance" means.
The POR reference makes me think maybe this clamp sits on the RC
timer node and at some point its leakage exceeds the pullup current
and the reset never releases.
If this is true then you may also have a design marginality or other
components to blame. For example if the C is leaky, same outcome?
Or the two gang up on the R and defeat the release because the
threshold is too high and somebody picked 10Mohms for R so
they could use a cheaper smaller cap (caps vary by V and C while
resistors pretty much not, so cheap it is). But then leakage matters,
where if you went down a decade on R and up a decade on C, no
prob.
Perhaps you have test access and info you are not using. Do you
measure POR release threshold (independent of UVLO)? Know
where threshold is and measure open circuit node voltage, are
you comfortably on the right side of it or one foot in the manure?
Good parts should not bake in or bake out. That's a fact. But you
might be more of a hero if you identified the contributing factors
and a resilient fix(es) as well. Like, if the ones passing are the ones
whose caps don't leak, anything in the field that develops a leak
might come back under warranty and then whoever sold it will be
looking for someone to pass the buck (or something that rhymes)
to.
One of my uncles had an Apple II that failed that way. Poking the
RSTb pin showed it stuck low-ish. New cap, all better.
Is the "failing" diode the one at the end or in the middle? Maybe
middle ones get residue that doesn't fully clean until cooked.
Is a 100% 50C unpowered bake too costly to get the job out the
door, while you investigate at more leisure? Still recommend the
full failure analysis, because gut says it's a "sum-of" kind of problem.