User`s guide
13
-
26 Monte Carlo and Sensitivity/Worst-Case Analyses
Worst is user-defined as the highest (HI) or lowest (LO) possible
collating function relative to the nominal run.
Procedure
To establish the initial value of the collating function, worst-
case analysis begins with a nominal run with all model
parameters at their nominal values. Next, multiple sensitivity
analyses determine the individual effect of each model
parameter on the collating function. This is accomplished by
varying model parameters, one at a time, in consecutive
simulations. The direction (better or worse) in which the
collating function changes with a small increase in each model
parameter is recorded.
Finally, for the worst-case run, each parameter value is taken as
far from its nominal as allowed by its tolerance, in the direction
which should cause the collating function to be its worst (given
by the HI or LO specification).
This procedure saves time by performing the minimum number
of simulations required to make an educated guess at the
parameter values which produce the worst results. It also has
some limitations, which will be discussed in the following
sections.
Outputs
A summary of the sensitivity analysis is printed in the PSpice A/
D output file (
.out). This summary shows the percent change in
the collating function corresponding to a small change in each
MAX maximum output variable value
MIN minimum output variable value
YMAX output variable value at the point where it
differs the most with the nominal run
RISE_EDGE
(value)
sweep value where the output variable
value crosses above a given threshold
value
FALL_EDGE
(value)
sweep value where the output variable
value crosses below a given threshold
value
Analo
g
behavioral models can
be used to measure waveform
characteristics other than those
detected by the available
collatin
g
functions, such as rise
time or slope. Analo
g
behavioral
models can also be used to
incorporate several volta
g
es and
currents into one output variable
to which a collatin
g
function may
be applied. See
Chapter
6,Analog Behavioral Modeling,
for more information.