Recent years has witnessed a large increase in the use of vibrating Micro-Electro-Mechanical-Systems (MEMS) especially in the expanding wireless telecommunication industry. In particular, the use of microresonators to generate or filter signals has facilitated a reduction in the size of many popular cell phones. Advances in microfabrication have increased the ability to create complex MEMS devices. Finite Element Analysis (FEA) has widely been used in the design of these devices. To obtain accurate simulations of complex MEMS devices, a dense FEA mesh is required resulting in computationally demanding simulation models. Arnoldi Model Order Reduction has been investigated and implemented to improve the computational efficiency of MEMS simulations. Using ANSYS, a popular FEA program, a micro resonator model was created. With Arnoldi, a Krylov subspace was extracted from the model and the model was projected onto the subspace reducing the model size. A harmonic simulation over normal operating frequencies was performed on the reduced model and compared with a simulation of the original model. It was found that the computational time was drastically reduced through the use of Arnoldi while achieving similar accuracy as compared to the original model.
Neural network closures for nonlinear model order reduction
