| Semiconductor devices are usually simulated by the analytical models at the verybeginning. A analytical model is based on some analyzed assumptions. Under theseassumptions, the physical problems and the electrical characteristics of semiconductordevices can be mathematically expressed. Though the analytical models are convenient,their practical applications are restricted, because there are so many approximations inthe analytical models that only one-dimensional (1-D) model can be applied and thesimulation accuracy is lowered. With the development of large-scale integrated circuits(IC), the analytical models are beyond the engineering requirement. Therefore, thenumerical models have been developed. To compare with the analytical models, thenumerical models are established from the basic equations of semiconductor devices.Numerical models are strictly built upon the collective structure of devices and then canbe solved by numerical methods to obtain the characteristic parameters. Numericalmethods are more accurate than the analytical model, but the solution process is morecomplex.Hence, it is necessary to use advanced numerical techniques to solve the nonlinearpartial differential equations. Firstly, the scale described by the nonlinear partialdifferential equations needed to be divided into a series of limited sub-scales. Secondly,nonlinear partial differential equations on total scale can be simplified as s collection oflinear partial differential equations on sub-scales. Lastly, the solution of nonlinearequations can be replaced by the solution linear equations.During the solving course, meshing is the key to transfer nonlinear problems tolinear problems. On obtaining the numerical models, meshing the first importantconsideration. Meshing course is complex and its qualification is directly related to thesimulation accuracy and the corresponding time consumption. Therefore, the maincontent of this dissertation is as follows.1. To introduce the semiconductor device physics knowledge and the discretenumerical methods.2. Though there have many meshing generators complied for now, we need to write a numerical discrete meshing generator fit to semiconductor devices. In this paper,the Quad-tree method and the Delaunay triangulation method are discussed.3. After the completion of the basic grid structure, adapting meshing technique hasbeen applied in the preliminary meshes.4. The storage of the information for the meshing boundary grids are treated byDelaunay method with the help of the open source software Triangle. |
