Design of surface acoustic wave resonators based on a series neural network

The surface acoustic wave (SAW) filter is widely applied in mobile communication, and the resonator is its main component. However, the traditional simulation and design methods of resonators often require much computation because the resonator includes a multilayer structure and many interfinger pairs. In order to improve the efficiency of simulation and designing, this paper proposed a series neural network to design the structural parameters backward based on the performance indicators. We validate the method using a SAW resonator based on a 42°YX cut LiTaO3 substrate with an aluminum electrode. The device consists of an interdigital transducer and two reflector gates. The test set results from simulation data show that the trained model has a relative average error of less than 5% on the devices' structural parameters, and the coefficient of determination is more significant than 0.99. In addition, we compare the predicted and the experimental results, which show that the series neural network has excellent potential to infer the electrical response and structural parameters of SAW devices. The proposed method provides a potential solution for improving the efficiency of simulation and design of surface acoustic wave resonators.