FBAR Filter Structural Parameters Optimizing With Deep-Learning Approach

When using genetic and gradient algorithms in ADS simulation software to design and optimize FBAR filters, it is usually using the scattering parameters of the FBAR filter as the optimization target and to adjust the 12 structural parameters (optimization variables) of the FBAR. However, this method, which is time-consuming, usually requires multiple iterations of optimization variables in a larger range to achieve convergence. This paper proposes an optimization model based on a 1-D CNN to construct the mapping from scattering parameter sequence of FBAR filter to 12 structural parameters. This model, giving the initial value and optimization range of structural parameters, will reduce the time of design and optimization of FBAR filter. The effectiveness of the method is proved by the example of the four-step FBAR filter.