Computer Aided Design of RF MEMS enables a reliable, cost effective and time saving alternative to the experiment guided development process. Full-wave analysis techniques including the finite element method and finite difference time domain have been applied and good results were obtained. However, a large memory/time overhead and inability for effective modeling on a system level limit their use for primarily individual device level analysis. In this paper, artificial neural network based macro-modeling approach for design and optimization of integrated circuits with RF MEMS resonators is proposed. Proposed approach combines the finite element analysis and electromechanical analogies to obtain an ANN trained blackbox model, which is fully compatible for analysis, tuning, design and optimization in a selected circuit simulator environment. The ANN algorithm is the centerpiece of this approach since it provides the intelligence for the model, i.e., mapping the device geometry dataset into corresponding output performance dataset. Although ANNs require extensive time and effort needed to prepare the training dataset, once the ANN is trained, it predicts unknown input/output relationships with significant savings in time/cost/effort while retaining the accuracy of full-wave analysis.
Journal: TechConnect Briefs
Volume: 3, Technical Proceedings of the 2005 NSTI Nanotechnology Conference and Trade Show, Volume 3
Published: May 8, 2005
Pages: 565 - 568
Industry sector: Sensors, MEMS, Electronics
Topic: MEMS & NEMS Devices, Modeling & Applications