Regarding binning related to yields, AFAIK there are two forms to consider:
1. Yields due to actual defects (forgot the official name for this).
2. Parametric yields (defect is not present, but clocks and/or power consumption are not achieving established targets)
cbn, Arachnotronic is correct but I would like to expand on this subject as it is near and dear to my work life (if I may)...
In every sense of the concept, both functional and parametric yield are the result of "physical" defects being present in the form of atoms literally not being present, or being present, in physical locations where they are desired or not wanted.
If they are not present where they are desired then we refer to this as an implant or dopant deficiency (reducing parametric yield due to less dopants and/or less conductors where we want and need them to be).
Likewise if they are present where they are not desired then we refer to this as a contaminant or undesired variability (again reducing either the desired drive currents, the targeted TDDB, or leakage current).
Even the "physical" defects, i.e. random blobs of add-on defect from the ambient surroundings can be captured and considered to be "undesired contaminants and/or misplaced dopants".
All of which impact both the functional and parametric yields.
My point is merely to say don't get hung up on discerning between the physical domain (is atom xyz present in location abc?) and the electrical domain (is the conduction band in the region of coordinates xyz altered to be abc?).
On the outside it may seem neat and tidy to delineate between the two realms, but on the inside there is an interplay between the two that in itself creates the reality that we all get to purchase