Research On Multi-Physics Modeling Method Of Passive Devices Based On Neural Network

With the rapid development of electronic technology,research on device modeling methods has become an important development direction in the field of microwave design.The structural complexity of microwave passive devices,the high requirements for product performance,and the shortened research and development cycle lead to passive device models in the microwave RF field that must not only describe the ideal electromagnetic characteristics of microwave devices,but also accurately reflect the impact of multiple physical parameters on device performance.The traditional multi-physics modeling methods of microwave passive devices often have problems such as complex modeling processes,long modeling cycles,and low model accuracy,which have been unable to meet existing device modeling requirements.In recent years,artificial neural network(ANN)techniques have been recognized as useful alternatives to conventional approaches in microwave modeling.Based on ANN,this paper proposes two multiphysics modeling methods for microwave passive devices:First,this paper proposes an artificial neural network method for modeling the multi-physics electromagnetic response of microwave passive devices.The data for modeling are generated using multiphysics simulations was imported into Neuro Modelerplus,a software specially used for neural network modeling,for training and testing,and finally an ANN multi-physics model was established.The trained multi-physics model can provide accurate and fast multi-physics analysis of device with lower computational cost and shorter design time.The accuracy and cost comparison between case modeling and traditional modeling methods proves the feasibility of the ANN multiphysics modeling method proposed in this paper.Second,based on the existing neural network space mapping(Neuro-SM)structure,this paper proposes a new Neuro-SM multiphysics modeling method suitable for microwave passive devices.The new Neuro-SM multi-physics model is a multi-physics model structure of general-purpose microwave device including a coarse model(empirical model)and two mapping networks.The model adds a mapping network to the input and output of the coarse model,and uses fewer neurons to obtain higher model accuracy,which reduces the complexity of the model and enhances the robustness of the model.When the multiple physical characteristics of the microwave device are complex or the accuracy of the coarse model is low,the two mapping networks adjust each other,and the model can accurately reflect the multiphysics characteristics of the device.The accuracy and cost comparison between case modeling and traditional modeling methods proves the feasibility of the Neuro-SM multiphysical modeling method proposed in this paper.