Research On Microwave GaN HEMT Large-Signal Model Parameter Extraction

Gallium nitride?Ga N?high electron mobility transistors?HEMTs?have become the research focus of microwave power devices because of their high operation frequency and output power density.The large-signal model for GaN HEMTs is the foundation for circuit design and plays a guiding role in the optimization of device process and structure.However,the existing large-signal equivalent circuit models for GaN HEMTs are mostly empiricism-based or semi-empiricism-based.In order to account for the significant self-heating effects,trapping-effects and ambient temperature effects of GaN HEMTs,these models contain a large number of fitting parameters.This makes it difficult to extract these models and update models according to process variations.Therefore,reseach on efficient parameter extraction of GaN HEMT large-signal models is in urgent need to improve the efficiency and accuracy of device modeling,shorten the development cycle of devices and circuits and promote development of GaN devices and circuits.In this dissertation,efficient parameter extraction method for GaN HEMTs empirical large-signal models,emerging quasi-physical GaN HEMT large-signal modeling and statistical modeling and corresponding parameter extraction have been deeply studied.The main research content is as follows.1.The existing small-signal model parameter extraction methods rely on manual tuning,thus have a low modeling efficiency,and hasnot considered the influence of exterior elements error accumulation on the accuracy of inner elements extraction.An iteration algorithm to extract extrinsic parameters is proposd to solve this problem.The validation results of 0.25?m process GaN HEMTs show that the proposed algorithm can reduce the S-parameters error by 40%in less than ten times of iterations and accurately predict S-parameters of GaN HEMTs up to 40 GHz.The proposed parameter extraction method can effectively improve the accuracy of GaN HEMT small-signal modeling.2.Considering that the empirical drain current I_dss model is complex due to self-heating and trapping effects and the parameter extraction is also difficult,a new parameter extraction procedure for empirical drain current I_dss model is proposed.The self-heating and trapping effects related parameters of an I_dss model are firstly divided into blocks according to their meaning.The parameters in each block are then extracted by fitting pulsed I-V transfer characteristic curves at different quiescent bias points.The proposed method follows a fixed procedure to extract I_dss model parameters step-by-step.The application of the proposed parameter extraction method in Angelov model shows that the extracted large-signal model can accurately predict dc I-V,pulsed I-V at different quiescent bias points,multi-bias S-parameters,output power,power added efficiency,gain and impedance charateristics of the device.At X and Ku band,the accuracy of output power and gain is more than 95%and the accuracy of power added efficiency is more than 90%.The proposed model parameter extraction methods are em-beded into the self-developed microwave device modeling and analysis software to greatly improve the efficiency of empirical model parameter extraction.This implements highly efficient and automatic parameters extraction of GaN HEMT empirical large-signal model.3.Compared with the empirical models for GaN HEMTs,the physics-based large-signal models contain less fitting paramters,have good physical meaning and play a guiding role in the optimization of device process and structure.However,the accuracy and convergence of the exsiting physics-based models still need to be improved.In order to solve this problem,a quasi-physical zone division?QPZD?GaN HEMT large-signal model is proposed in this dissertation.The proposed large-signal model is based on zone division method and surface-potential theory,it makes a breakthrough in the accurate and analytical modeling of self-heating effects,ambient temperature effects and trapping effects.The validation results of 0.15?m process GaN HEMTs with different gate width show that the proposed model can accurately predict dc I-V,pulsed I-V at different quiescent bias points,multi-bias S-parameters,output power,power added efficiency,gain,third-order intermodulation and impedance charateristics of the devices.The proposed large-signal model is also validated by Ka band power amplifier monolithic microwave integrated circuit?MMIC?design.Results show that within the frequency range of 32³8 GHz,the accuracy of output power and power added efficiency is more than 94%.This promotes the engineering application of GaN HEMT physics-based large-signal models.Compared with the Angelov empirical model,the proposed model in this dissertation reduces the number of parameters by55%.For the latest reported physics-based surface potential model,the percentage is20%.Moreover,the proposed model is easy to extract and has good convergence and physical meaning.4.The exsiting statistical models for GaN HEMTs are mostly empiricism-based.These statistical models contain too much fitting parameters and heavily rely on measurements and parameter extraction that contains manual tuning.In order to solve this problem,a full physical paramters based large-signal statistical modeling method based on the QPZD model is proposed in this dissertation.The proposed statistical model contains barrier layer thickness,electron saturated velocity,electron sheet density and critical electric field as parameters.The proposed statistical modeling method can avoid information loss from the original dataset.Validation results of 0.15?m process GaN HEMTs from different batches show that the proposed statistical model can accurately predict the means,standard deviations and correlations of the physical parameters of GaN HEMTs.The simulated probability density of I-V curves,transconductance and pinch-off voltage also agrees well with the measured sample.The proposed statistical model is also validated by 32³8 GHz power amplifier MMIC design.Results show that the accuracy of means of output power and power added efficiency is more than 95%.The proposed statistical model represents a significant step from empiricism-based to physics-based statistical models that can be used for both improving the deep-submicron device process and circuit yield estimation.Moreover,the proposed statistical model can provide a guidance to develop process-circuit coordination design.