TCAD simulators tuning is a crucial step in technology characterization. With properly calibrated physical models, it is expected that TCAD simulation can accurately model integrated circuits manufacturing processes and predict the electrical characteristics of fabricated devices. A virtual factory, based on accurate process and device simulation, provides efficient means for exploring the manufacturing space and evaluating design alternatives. Coupled with statistical analysis techniques, it can be used to generate the necessary data for extracting the circuit model parameters needed for worst case technology characterization as explained in the next chapter. Such predictive capabilities reduce time and costs associated with the manufacturing of high performance semiconductor devices.
In general, the calibration methodology involves the use of nonlinear least-squares optimization to adjust model parameters to minimize the errors between the calculated simulator outputs and carefully selected experimental data. It is important that the direct effect of the parameters be dominant on the measurement data used in the extraction to ensure the validity of the determined values for other data sets.
Electrical data needed for device simulators calibration is readily available, but reliable two- and three- dimensional process characterization measurements are more difficult to obtain [60]. This is especially critical because device simulation input consists in part of the results of process simulation such as the device structure and doping profile. In this light, the importance of multi-dimensional process metrology tools, such as the inverse modeling method, is greatly enhanced.
Three example applications that demonstrate the use of the nonlinear
least-squares task module for the calibration of TCAD process and device
simulators are presented. In the first two examples, the
MINIMOS [92] mobility and avalanche generation model parameters are
adjusted to fit experimental drain (
) and substrate (
)
currents respectively. In the third example, the parameters of a new
model for as-implanted 1-D profiles in SUPREM-III [38] are
determined to match SIMS profile measurement.