The measured subthreshold slope is usually expressed in
mV/decade. But the model parameters used to fit it, are
another thing entirely. Back in the day it was NSS (num
of surface states), I don't know about BSIM or more
advanced models, which get fitted by specialist types
with fancy software to crunch the I-V data.
You have to read the docs for the particular MOS model
you are using, to know what the relevant "handles" are.
And then most likely run some parametric sweeps to see
their ranges and effects. You may also find some second
order params that control the "blend" from subthreshold
to "on" regions, that have to be smoothed in after the
main slope is roughly-right or you'll get weird singularities
or nonmonotonicities that bother convergence and near-
threshold accuracy.