Machine-Learning-Assisted Anchor Loss Reduction of MEMS Resonator With One-Dimensional Phononic Crystal Tether

In this article, a machine learning (ML)-based anchor loss reduction approach for microelectromechanical systems (MEMSs) resonators with 1-D gourd-shaped phononic crystal (PnC) tether is reported here. A forward artificial neural network (ANN) is used to accurately predict the dispersion characteristics of the PnC tether. Then, a tandem network comprising inverse ANN and forward ANN is used to predict the corresponding geometries of PnC tether according to the expected bandgap target. The final optimal geometries of PnC tether corresponding to the desired bandgap, for width-extensional (WE) mode piezoelectric MEMS resonators, are determined with the aid of the tandem network. To verify the reliability of the proposed tandem network, the dispersion characteristics of the predicted geometries obtained from ML are also simulated by finite element method (FEM) analysis. The forward ANN serves as a high-speed and high-accuracy tool for calculating the dispersion characteristics of the PnC, performing calculations <roman xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</roman>×<roman xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">10</roman><sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><roman>4</roman></sup> times faster than FEM analysis, with an accuracy of 99.9%. ML provides an intelligible and effective approach to screen thousands of design candidates with the forward ANN, as well as optimizing the PnC design with the tandem network.