Machine Learning Assisted AlN Resonator Optimization Achieving 7.8% Coupling Coefficient with 70% Reduction in Time Expenditure

In our work, we present a machine learning-assisted optimization method that can generate the optimal design parameters for the laterally coupled alternating thickness (LCAT) mode resonator, with significantly reduced time expenditure compared to the finite element method (FEM). Trade-offs among electrode width (w), pitch width (p), and electrode thickness (h) were made by neural network to obtain the optimal design parameters. In the experiment, more than 2,000 sets of data with their corresponding impedance spectra were used for training. The results predicted by the neural network demonstrated a 7.8% effective electromechanical coupling coefficient $\left( {k_{{\text{eff}}}^2} \right)$ and a 70% reduction in total computational time, highlighting the vast potential of machine learning-assisted design in acoustic resonators.