This is a many-layered question and one which simulation
can only approximate.
Minimum detectable, vs minimum -reliably- detectable, just
by itself, has dark current, telegraph noise, detector circuitry
threshold / noise contribution aspects. Not that almost all of
these push into areas that often are poorly modeled or not
well modelable (dark current and RTN being extremely process
and device-construction-details variable).
Without preceding material characterization, the models have
a pretty poor chance of being good for anything but basic
functionality. How broad based and production-relevant
that material, and how diligent, process-comprehensive and
accurate the modeling, that would be between you and the
foundry.