ANN Characterization Of Noise Parameters In RF And Microwave Transistors

The metal oxide semiconductor field effect transistor(MOSFET)with characteristic size of tens of nanometers has become the mainstream of the realization of RF/microwave front-end integrated circuits in modern and next generation wireless communication systems due to its advantages of low cost,low power consumption and high integration.The accurate modeling of MOSFET high frequency small signal equivalent noise circuit is the first problem to be solved in the simulation optimization design of MOSFET low noise amplifier in the frequency band of RF microwave wave.At present,the parameters of the active device equivalent noise circuit model are obtained by measuring the scattering(S)and four-noise parameters of the device under a fixed operating bias voltage,and by studying the high-fidelity parameter extraction method.Given the equivalent circuit of MOSFET intrinsic element parameters and the mechanism of device noise bias dependency,therefore,in order to ensure the accuracy of MOSFET small signal equivalent circuit noise model,must be in a wider range of test frequency,to the work under different bias voltage of MOSFET scattering parameters and four noise to carry on the intensive sampling,thus increasing the microwave and millimeter wave research and development costs and cycles of MOSFET circuits.At the same time,the processing of massive test data also leads to more mathematical problems in the complex parameter extraction algorithm.For huge amounts of data in the modeling in order to reduce the MOSFET and high precision measurement and the dependence of the complex parameter extraction algorithm,this paper using Artificial neural network(Artificial neural network,ANN)in the active device modeling aspects of validity,accuracy and simplicity,RF microwave frequencies for40 nm MOSFET in strong inversion region under high frequency noise modeling problem,has carried out the following three aspects: the application of research work.Firstly,by analyzing the characteristics of ANN modeling technology for semiconductor active devices and summarizing the experience of existing modeling cases,it is found that MOSFET modeling can take advantage of the strong generalization ability of ANN modeling technology for active devices and refer to the mature ANN topology structure and the matching optimization training algorithm.Therefore,the key to high precision ANN modeling for MOSFET is to master the nonlinear relationship between the working conditions(input of ANN model)and the performance parameters(output of ANN model)of the device.More importantly,the data representing the nonlinear relationship must be accurate and reliable,because they will be used as the training data and test samples of the ANN model.Therefore,in the second chapter,two aspects of research work are carried out.On the one hand,the direct extraction method of the parameters of the equivalent circuit model of the 40 nm MOSFET small signal is studied.Among them,the equivalent circuit of full characterization of MOSFET,on the basis of the quasi static effect,through the equivalent circuit admittance parameter simplified modeling,to achieve the accurate extraction of equivalent circuit parameters,thus for training intrinsic element parameters ofthe ANN model built in this paper provide an accurate simulation data laid a foundation.On the other hand,this paper presents a parameterized model of the power spectrum density of the noise current source of the 40 nm MOSFET admittance equivalent noise model,which provides a test sample for verifying the accuracy of the ANN model with four noise parameters.Secondly,in the third chapter of this paper,the ANN model of intrinsic element parameters of 40-nanometer MOSFET is established by referring to the experience of ANN modeling of RF and microwave devices,using the generalization ability of ANN modeling,selecting the BP network structure and using the levenbert-marquardt algorithm as the optimization training algorithm.The test results show that the ANN model not only has high prediction accuracy in the bias voltage range of the strong inverse region,but also has good S parameter interpolation ability.Therefore,by using the ANN model of eigenelement parameters,it is easy to continuously obtain S parameter data in any bias point and frequency range,which can avoid the complicated S parameter measurement process and tedious equivalent circuit parameter extraction process.The high precision simulation data generated by the ANN model of eigen element parameter will be used as the knowledge,which provides a guarantee for improving the accuracy of the ANN model of four-noise parameter.Finally,in the fourth chapter of this paper,a four noise parameter ANN model based on knowledge is established.The model is BP network based on prior knowledge,and the training algorithm still adopts levenbert-marquardt algorithm.Is the purpose of the model in order to build up a wide frequency band and bias voltage range,able to quickly and continuously characterization of 40 nm MOSFET the offset and frequency dependence of the high frequency noise,which can avoid the MOSFET high-frequency noise analysis and modeling in the characterization of the four parameters of noise measurement requirements,easy to obtain any bias point and frequency noise model data.The experimental results show that the results of simulation extraction based on the established knowledge based four-noise parameter ANN model are consistent with the results of simulation calculation based on the physical two-port equivalent noise current accurate model,and the validity and accuracy of the proposed model are verified.