Using circuit model from layout-level synthesis as coarse model in space mapping and its application in modelling low-temperature ceramic cofired radio frequency circuits
The technique of using a coarse model obtained from layout-level synthesis of radio frequency components in neural inverse space mapping (NISM) optimisation technique is proposed. Layout-level synthesis is based on a combination of segmented lump-circuit modelling, nonlinear mapping using neural network (NN) and circuit-level optimisation. The circuit model based on the electromagnetic (EM) simulation describes the important parameters and parasitic parameters and serves as a high-quality coarse model in applying NISM. In implementation of NISM optimisation, trained NN is introduced as a buffer space between the original coarse model space and the fine model space, where the subset of the coarse model space is completely different from that of the fine model space. Stoer–Bulirsch adaptive frequency sampling (S–B AFS) technique is used in frequency sweep of the fine model to further reduce the computing time. An application design of low-temperature ceramic cofired filter is used to demonstrate the proposed technique. Synthesis and optimisation results of the filter meet the design specifications.
