Machine Learning and Genetic Algorithm-Assisted Design of MEMS Resonator With Multifold Beam Tether for Anchor Loss Reduction

In this work, a machine learning (ML) and genetic algorithm (GA) optimization framework is proposed for designing the square-extensional (SE) mode microelectromechanical systems (MEMS) resonator with multifold beam tether. The presented multifold beam tether is introduced to suppress anchor loss and is further optimized using the ML model and GA. First, the neural network (NN) is trained on a dataset generated from finite element method (FEM) simulations to accurately predict the anchor loss (${Q}_{\text {anchor}}$). Bayesian optimization is subsequently employed to determine the optimal hyperparameters of the NN model, which is then coupled with the GA to efficiently explore the multidimensional design space. The optimal tether geometries corresponding to the maximum predicted${Q}_{\text {anchor}}$are obtained through the GA search. FEM simulation based on the tether dimensions predicted by the surrogate model is conducted to verify the reliability of the optimization framework. Finally, the SE-mode resonators with the optimized multifold beam tether are fabricated and experimentally characterized. The measured frequency response exhibits a resonant frequency of 10.141 MHz and a high quality factor ($Q$) of${3}.{72}{\,}\times {\,}{10}^{{5}}$. The proposed surrogate-assisted GA framework provides a simple and efficient approach for tether optimization, enabling significant suppression of anchor loss and facilitating the realization of high$Q$MEMS resonators.