High-fidelity simulation, modeling, and optimization of devices, circuits, and systems exploiting engineering surrogates

点击翻译按钮获取中文摘要

First-pass success and time-to-market are crucial for manufacturability-driven design of modern telecommunication circuits. This requires powerful, accurate and fast CAD tools for component modeling, circuit simulation, and automated design optimization. High-fidelity or "fine" models based directly on physics and electromagnetics offer accuracy and handle arbitrary geometrical shapes. However, for design optimization, statistical (yield-driven) optimization and design with tolerances, they are computationally intensive, even prohibitive. To address this issue we focus on iteratively refined, physically-based (engineering) surrogates: we take the high-fidelity simulator out of the classical optimization loop. Our expertise in design centering, tolerance assignment and statistical design for manufacturability, tuning and yield-driven optimization (recognized by the IEEE MTT-Society's 2004 Microwave Application Award) will evolve into a flexible theoretical and software framework for electromagnetics-based, computer-aided engineering of wireless, radio-frequency, and microwave components and circuits. This framework will encompass the surrogate methodology we pioneered in 1993 called "space mapping," a process which continues to yield significant modeling and design successes in the engineering community in Canada and internationally. For example, our special case of "port tuning" has now been adopted by major electromagnetic software vendors. The next phase of research and development must include available adjoint sensitivities, in addition to the physics-based surrogate models and a trust region methodology, for even more effective device modeling and design optimization. Our algorithms should deliver, in only a handful of fine model evaluations, the accuracy expected from classical direct optimization. We will collaborate with Dr. Kaj Madsen (Technical University of Denmark) and Dr. S. Koziel (University of Reykjavik) on efficient, robust algorithms and convergence theories and applications of surrogate-based optimization. We will work closely with Dr. Q.J. Zhang (Carleton), Dr. Natalia K. Nikolova (McMaster), and Dr. M.H. Bakr (McMaster), and their respective HQP groups.