Original Date: 01/24/1994
Revision Date: 01/18/2007
Best Practice : Response Surface Methods
Harris Semiconductor engineering personnel regularly use Response Surface Methods (RSMs), a statistical analysis method, for acquiring and applying process control knowledge. RSM is part of the standard statistical analysis tool set available to Harris engineers for evaluation of processes. The use of the RS/Discover software has been implemented in a way that is convenient for engineers to use RSMs without the assistance of a statistician.
The RSM methodology is similar to design of experiments (DOE) methodology, as it is useful to determine the results of varying process control parameters over a given range. However, RSM is a powerful tool since a number of related nonlinear process performance points such as process contours can be studied as opposed to the typical high and low levels for DOE. RSM is also able to determine if a process' controlling factor is actually the result of interactions between two or more factors  not always possible with the DOE methodology. Analysis time to determine process controlling factors with RSM methodology is frequently one week when working with eight to ten factors compared to two to three weeks for a similar analysis with DOE methodology.
A particular example of Harris' use of RSMs was the development of an empirical model for the surface charge in a gate oxide unit for a cryogenic application. Harris designed and characterized the gate oxide unit process as part of a complete silicon process, particularly the relationship between the electrical surface charge (Q_{ss}) versus five key process factors. Two experiments were conducted  from the first experiment, all five factors were confirmed to be important and an empirical model was developed. From the second experiment, the empirical model was confirmed, and the shape of the process curvature was explored for the most important factor. Regression analysis of the model showed high correlation coefficients and that Q_{ss} was predictable with the empirical model. In addition, a series of graphs prepared via the RSM software aided in interpretation of the process model.
Harris has facilitated the efficient use of RSM statistical methods in the analysis of process controls by providing training in, and access to, the RS/Discover software. In return, RSM methodology is paying dividends to Harris through reduced analysis time and costs and improved accuracy in determining process control interactions.
For more information see the
Point of Contact for this survey.
