Previous: Process Control Recipe Generation
Up: Examples
Previous Page: Process Control Recipe Generation
Next Page: Gate Oxidation Problem Formulation

Design for Manufacturability

DOE/Opt was used with a statistical process simulator to study design for manufacturability. Several criteria for optimizing process recipes under random process variations have been described in the literature [39][38][37][36][35]. The concepts of parametric yield, standard deviation and ``signal to noise ratio'' have been implemented using DOE/Opt and used for process optimization. We present an example of gate oxidation process optimization under simulated manufacturing variations. Using this example, we compare three different design for manufacturability strategies.

The semi-analytic statistical process and device simulator FABRICS [41][40] is used to emulate the effects of process and material parameter disturbances on the resulting device structure. Given the nominal settings for a process step and the standard deviations for the process disturbances, we seek to determine the mean shift in the disturbances (or equivalently new process parameter nominal values) that will result in a design that is optimal under some manufacturability criterion.

We encapsulated FABRICS into DOE/Opt via a body Tcl script which generates FABRICS input files, runs the simulator, and parses the resulting output file. For each DOE/Opt run, FABRICS is executed in the statistical mode to generate samples, from which the statistical response parameters are calculated. As shown in Fig. 11, our approach is to construct response surface models for each of several responses as a function of the inputs and disturbances.

In the oxidation example below, a Box-Wilson on a cube design was executed and full quadratic models for each output were generated using least squares regression. We found that all of the models had excellent goodness of fits.

• Gate Oxidation Problem Formulation
• Gate Oxidation Optimization Results