Neural Network Approaches To The Modeling Of Nonlinear Microwave Devices

With constant development of new technology in the semiconductor industry,the existing models may not be accurate when new devices are used.However,the accuracy of the model is important to enhance the product properties,shorten the design cycles and reduce costs.In order to solve the problem,the existing models have to be modified or new models have to be developed.However,developing a new equivalent circuit model,which usually requires manual trial-and-error effort,is time consuming.The physical-based models are computationally expensive.Therefore,neural network approaches to the modeling of nonlinear devices are proposed.These approaches can effectively build accurate and more general models through a systematic computation process.In order to create an accurate model,a noval non-quasi-static microwave device modeling technique based on time delay neural network(TDNN)is proposed in this thesis.TDNN has been used in the behavior modeling of PA with time domain information.Here,a general TDNN technique for nonlinear microwave device modeling is proposed,which contains DC,small-signal and large-signal information to build a full large-signal nonlinear model considering non-quasi-static effects.A new formulation to train the proposed TDNN with DC,small-signal,and large signal data is proposed.Examples on modeling GaAs metal-semiconductor-field-effect transistor(MESFET)and GaAs high-electron mobility transistor(HEMT)demonstrate that the TDNN technique is an effective and accurate method to represent the large-signal nonlinear behavior for nonlinear device modeling.In order to build device models with good properties,a new Wiener-type dynamic neural network(DNN)approach for nonlinear device modeling is proposed in this thesis,where Wiener system is used in microwave device modeling area for the first time.The proposed analytical formulation of Wiener-type DNN structure consists of a cascade of a simplified linear dynamic part and a nonlinear static part based on a Wiener system formulation.The simplified linear dynamic equations are obtained by vector fitting enhancing the efficiency of the model.The number of state variables can be directly determined by vector fitting.A new formulation and new sensitivity analysis technique is derived to train the proposed Wiener-type DNN with DC,small-signal,and large-signal data.GaAs Pseu-domorphic HEMT(pHEMT)and GaAs HEMT examples have been used to illustrate how to determine the structure of the Wiener-type DNN.These examples also demonstrate that the Wiener-type DNN structure is applicable and the Wiener DNN model can accurately match the original microwave device data.A new training algorithm utilizing gradient based optimization is also developed for fast training of the proposed Wiener-type DNN model.Application examples on modeling GaAs MESFET and a real 2?50 gatewidths GaAs pHEMT are presented.These examples demonstrate that this new gradient based training algorithm is effective.The proposed Wiener-type DNN model can be trained to be accurate relative to device data.Furthermore,the proposed Wiener-type DNN provides enhanced convergence properties over existing neural network approaches such as TDNN and TDNN with extrapolation.The use of Wiener-type DNN model in harmonic balance simulations demonstrates that the Wiener-type DNN is a robust approach for modeling various types of microwave devices.