In this analysis (actual EVA.SCP file), for each device containing toleranced parameter values, the parameters are varied to their extreme value such that scalar measurements (i.e., Vout, iR2, etc.) are maximized. The extreme value for each device tolerance with respect to a measurement is first based on the sign of a previously-run sensitivity analysis, whereby the device's tolerance is slightly perturbed. If the sensitivity analysis produced a positive change in measurement for a specified node or device, then the device under sensitivity analysis is railed to its maximum parametric value, or to its lowest value if a negative change was incurred at the measured node or device. This is the basis of the EVA-HI routine. The EVA-LO routine sets parameterized devices to their maximum or minimum value, based on which ever produced a minimized measurement result at a specified node or device during the sensitivity analysis. A simulation is then run using the new extreme parametric value for the toleranced device. The result for the node or device measurement is saved in the "evahi" plot, or "evalo" plot. The aforementioned sensitivity analysis and simulation are repeated for every parameterized device in the design, and for each measurement specified by the user. When all simulations are finished, the measurements are printed to the IsSpice4 output file in a format that can be read back in by SpiceNet for recording of measurements specified in Simulation Control's "Results" dialog. Additionally, the output file contains a summary report for the user's records. If making an evalo analysis is not opted, the user can set measurement min/max test limits by expanding the measurements to "pass with symmetry" (amongst other choices) in the "Results" dialog as shown below.
The extreme value in this analysis refers
to the parameters, not the resultant measurements. For most
moderately complex circuits, the extreme value of the resultant
measurement occurs when some of the parameters are at an intermediate
rather than an extreme value. However, we usually find that
EVA results produce wider measurement test limits than Monte
Carlo — making it a worthwhile investment. Finding the true
extreme value of the resulting measurements requires solution
of a multi-parameter optimization problem. This becomes nearly
impossible for larger circuits because the number of simulations
grows as the product of parameters times the vectors. The EVA
in this script runs an analysis for
each toleranced parameter to get perturbation results and another
for each measurement to get the final results.
