Machine learning as a new strategy for designing surface acoustic wave resonators

Abstract

Surface Acoustic Wave (SAW) technology has been widely applied in the fields such as communication and sensing. The performance of SAW devices is significantly influenced by designs of their key component, Interdigital Transducers (IDTs), and thus Coupling of Modes (COM) theory has been used as one of the most employed design tools for SAW devices due to its fast computational speed. Accuracy of this model is primarily dependent upon the COM parameters, but the traditional approach to obtain these parameters is heavily relied on accuracy of the input material properties, which has become a key issue for successful applications of this model. This paper proposed a new strategy to utilize the results obtained from the COM model as a dataset and then employ five different machine learning models for performing regression analysis and accurately extracting the COM parameters. To validate the accuracy of this approach, experimental verifications were performed using a 128°Y-X LiNbO3 based SAW resonator as an example. The machine learning model with the best predictive performance, i.e., Extreme Gradient Boosting, was used to predict the COM parameters corresponding to experimental results, which were subsequently used in conjunction with the COM model for further calculations and comparisons with the experimental results. Results showed that the calculated results exhibit the same trend of resonance Q-values as the experimental results, demonstrating its effective solution for accurately extracting COM parameters.