Modeling And Parameter Extraction Methods For Impact Ionization Coefficients In Silicon-based Lateral Power Devices

Lateral Double-Diffusion Metal Oxide Semiconductor（LDMOS）have become a major component of the Hight Voltage Integrated Circuit（HVIC）due to their high breakdown voltage,low on-resistance,low power consumption,and high integration.Specific on-resistance(R_on,sp)and Breakdown Voltage（BV）,as the core design parameters of lateral power devices,has been the hot topic of research in the field of power devices.The compromise between breakdown voltage and on-resistance is a key issue in the design of lateral power devices.The avalanche breakdown voltage of lateral power devices is determined by the impact ionization of carriers in the semiconductor layer,where the impact ionization coefficient is a key parameter for analyzing the impact ionization process and the avalanche withstand voltage.However,due to the inherent two-dimensional coupling effect of lateral power devices,the traditional Fulop approximation model for solving the avalanche ionization integral in a one-dimensional PN junction is no longer applicable to the solution of the avalanche breakdown voltage of lateral power devices.In this paper,the study is centered on the impact ionization coefficient model for lateral power devices.Firstly,it is verified that the conventional silicon-based impact ionization approximation model based on one-dimensional PN junctions is no longer applicable in the analysis of two-dimensional power devices based on analytical models and numerical simulation tools.Further,an improved empirical model for the impact ionization coefficient is derived based on virtual retarded junction theory and avalanche breakdown theory for lateral power devices.A prediction model from device structural parameters to collisional ionization coefficients is finally developed using deep neural networks,which achieves an accuracy of 97%for the impact ionization coefficient prediction model.The study includes.First,a two-dimensional impact ionization model study.This study illustrates the variation of the impact ionization coefficient with the device structure parameters when the drift region of the lateral power device is partially depleted and completely depleted.Based on the virtual retarded junction theory and avalanche breakdown theory,an empirical model of the two-dimensional impact ionization coefficient is proposed.This empirical model can quantitatively describe the variation of the impact ionization coefficient with the device structure parameters,which provides guidance and assistance for the accurate solution of the avalanche breakdown voltage of lateral power devices.Secondly,the two-dimensional impact ionization coefficient prediction method is studied.This thesis proposes a new two-dimensional impact ionization coefficient prediction model based on deep learning networks to realize the prediction of impact ionization coefficients for lateral power devices with two-dimensional voltage breakdown structures.Compared with the manual parameter extraction method based on TCAD simulation software,the accuracy of the prediction model can reach 97%,and the prediction time can be greatly reduced while ensuring the prediction accuracy.Thirdly,an optimization criterion for Triple RESURF（T-RESURF）lateral power devices based on virtual retarded junction theory is proposed.The method bypasses solving the complex two-dimensional Poisson equation by equating the complex two-dimensional drift region of the T-RESURF transverse power device to a one-dimensional virtual retardation junction.The consistency of the model with the TCAD simulation results demonstrates the accuracy of the method.And thanks to its simplicity and accuracy,the proposed method not only enables qualitative and quantitative analysis of T-RESURF effects but also provides guidance for the design and optimization of T-RESURF lateral power devices.