A Novel Parametric System-Level Modeling Method for MEMS Devices Combining Artificial Neural Networks and Behavior Description

Abstract

System-level simulation with macromodel is an important way to help MEMS design and optimization. The applications of traditional MEMS macromodels are confronted with the challenge of incorporating a substantial number of parameters into the macromodel and reconstructing the macromodel when new parameters should be added into the model. To solve the above problem, we propose a novel method for parameterizing macromodels based on modularizing parameters. Firstly, a basic macromodel of a MEMS device with constant hypothetical parameters is constructed. Subsequently, two ways are used to modularize the parameters. The first one is that artificial neural networks (ANNs) are adopted to construct the relationship between non-intuitive parameters with fuzzy behavior and basic macromodel to acquire abstract equations. Another is that the behavioral models of parameters are directly constructed based on behavioral equations for intuitive parameters with clear behavior. Subsequently, a way to implement ANN models by using Verilog-A is also given. Finally, the basic macromodel is assembled with various parameters to obtain the parameterized macromodel of the MEMS device. The highlights of this method are manifested in two aspects. First, ANNs based on data-driven can be applied to various types of parameters, so it has good universality. Second, during the parameterization process, there is no need to reconstruct the basic macromodel. MEMS thermal wind sensor is used to demonstrate the proposed method. The accurate results indicate that this method can provide accurately parameterized macromodels for system-level simulation and has the potential to efficiently support the optimization design of MEMS.[2024-0133]