Previous: Process/Device Optimization
Up: Examples
Next: Process Control Recipe Generation
Previous Page: Process/Device Optimization
Next Page: Process Control Recipe Generation
DOE/Opt has been applied to simulator model tuning. In this case,
the goal was to determine ionization coefficients in an internally
enhanced version of PISCES [31] to match measurements of
substrate current. The model is of the form
.
PISCES simulation decks were developed which calculate substrate
current (
) at drain voltages
, 6, and 7 volts, and gate
voltage
, 2, 3, and 4 volts. The input parameters to the
script are the ionization model coefficients
,
, and exponent
. A Box-Wilson experimental design
varying these parameters was performed, and regression models of
as a function of the ionization model coefficients
were constructed. Optimization was performed using the response
surface models to find ionization model coefficients that achieved the
best fit to the experimental data (minimum sum of squared error). The
result was approximately 15% average error across several
values. A comparison between the substrate currents from
measurements, simulation before calibration, and simulation after
calibration is made in Fig. 8.
Two observations on the use of DOE/Opt in this problem bear special mention. First, we found that DOE/Opt is well suited to optimization and fitting to a small and mixed number of targets and constraints. In its current form, it is not ideal for ``parameter extraction'' where one desires to fit to large sets of data. Work to improve DOE/Opt's capability by interfacing to the MACH system is underway [32]. Second, we find that the most difficult task facing a user of DOE/Opt is writing the data extraction routines used within a script body. We found it necessary to develop and make available to users utility routines that extract data (e.g. current-voltage curves) from process and device simulation output files. To support higher level tools such as DOE/Opt, it is important that the textual and binary output formats of simulation tools be well documented and available.