For reliability "guarantees" to apply, none of the Jxyz
limits shall be violated. Each applies to a specific type
of electrical stress (DC=Jmax, unidirectional pulsed =
timer-averaged Javg, symmetric reversing AC-only is
Jrms. The case of AC-riding-on-DC is a judgment call
I suppose as now there is a reversible and irreversible
component to material migration, summed.
Re thickness, you are unlikely to know, care or find out
what the true thickness is (or was in the foundational
experiment behind all the rules derivations) and so
everyone just gets to pretend that nominal is the case
at hand, regardless. Unless you want to go get SEM
photos (at the worst case step, which you get to
find and indicate for the sample sectioning) and then
repeat the reliability analysis with the new numbers
(a fool's errand since the original sample metal layer
thicknesses are likely to be either never known or
simply not worth digging up and providing, so your
relation between material in hand and material in
rules, can never be known beyond process tolerance
slop).
I'd forget about delta-T until you get close to a rule
violation and need a shave.