ANOVA Investigation of Neural Network Guided Spurious Modes Reduction in Lithium Niobate MEMS Resonators

In this work, X-cut Lithium Niobate MEMS resonators are fabricated and tested to further identify and confirm the impact of design features on the rise of spurious modes. Building upon previous work, where a neural network identified the key features to be the number of electrode finger pairs, electrode length, and anchor width, 432 resonators were designed and fabricated to experimentally test and identify the contribution of these three features to spurious mode production. Each resonator had a unique combination of the design geometries spanning across three operating frequencies: 100 MHz, 200 MHz, and 400 MHz. By measuring the admittance and assigning a unique metric based on the severity of the spurious modes to each resonator, it was confirmed through analysis of variance (ANOVA) that each of the three design features and the frequency of operation have statistical significance in contributing to the rise of in-band spurious modes. Additionally, a linear regression model of the measured data concluded that the optimal Lithium Niobate resonator design to reduce the severity of in-band spurious modes requires a full anchor width, short aperture length, and a reduced number of IDE pairs. These results open the path to using advanced algorithms for further hardware optimization.