Previous: DOE/Opt: A System for Design of Experiments, Response Surface Modeling, and Optimization using Process and Device Simulation
Up: DOE/Opt: A System for Design of Experiments, Response Surface Modeling, and Optimization using Process and Device Simulation
Next: System Architecture
Previous Page: DOE/Opt: A System for Design of Experiments, Response Surface Modeling, and Optimization using Process and Device Simulation
Next Page: System Architecture

Introduction

Semiconductor process and device design can substantially benefit from the use of simulation and modeling to reduce the cost and time required to develop new or extend existing technology [1]. Technology development, however, requires substantially more than a fundamental simulation capability: tools and methods to assist in exploration of design tradeoffs and to optimize a design are becoming increasingly important.

Work on frameworks for the integration of technology CAD (TCAD) tools has focused primarily on underlying representations for the wafer and process, and on the interfaces between simulation tools and these representations [4][3][2]. Systems which achieve integration and help automate the execution of tools have also been demonstrated [7][6][5]. Less common are tools or systems which directly address how one uses such integrated simulation capability to solve design problems [8]. Alvarez explored statistical modeling methods for the design and optimization of processes and devices [9]. In that work, formal (Box-Behnken) design of experiments were used to construct response surface models. A grid search method was used to explore the design space and to identify feasible regions given goals and constraints on multiple performances and performance sensitivities. Other design exploration and optimization techniques have been investigated elsewhere, particularly for circuit yield optimization [12][11][10].

This paper makes contributions in two areas. First, we describe approaches for the use of experimental design, regression or direct simulation modeling, and optimization that we have found to be important in process/device design, simulator tuning, coupling to process control, and design for manufacturability. Second, we describe a solution to issues in the implementation of a general purpose optimization tool suitable for end TCAD users including graphical interface, user programmability via an extension language, and interfaces to existing simulation tools. DOE/Opt is a ``task level'' tool which complements existing TCAD Framework research on ``data level'' representations, and builds on ``tool level'' simulation capability [13]. Two audiences should thus benefit from this paper: (1) designers who will increasingly demand the types of simulation, modeling, and optimization capability represented by DOE/Opt, and (2) implementors who will find it necessary to implement similar high level utility programs in the future.

In Section 2, the overall architecture of the prototype DOE/Opt system is introduced. Each of the key conceptual components of the system is discussed in succeeding sections. The encapsulation and use of simulation or other models, and the construction of analytic response surfaces is discussed in Section 3. The integration of formal experimental design techniques with (1) simulation or experimental execution, and (2) design exploration and model construction are described in Section 4. The integration and use of nonlinear optimization is discussed in Section 5. Four examples are presented in Section 6 to illustrate the use of the DOE/Opt system. These include process/device performance optimization, simulator model tuning, process control recipe generation, and manufacturability optimization. Conclusions are drawn in Section 7.