To cover all demands in TCAD expected in the near future in a robust, practical, computationally manageable, and extendable manner, the standalone three-dimensional mesh generator deLink was embedded into a generic meshing framework with the capacity to use a set of meshing rules for n-dimensional mesh generation, where the rule set describes how to connect elements in two and three dimensions. To keep the implementation simple, the specification of the meshing rules is not part of the application. Therewith the code complexity is independent of the number of rules. Within this generalization the very important initial surface triangulation for the advancing front Delaunay method is handled just as the problem of very thin layers with a set of coordinated specific rules for surface and volume meshing. The problem of graded and anisotropic mesh generation is handled by applying transformations to the rule set where the generation of points at run time can be achieved easily by defining rules which allow point insertion. All different kinds of geometrical and topological adaptation can also be handled by geometrical and topological adaptation rules if the classification of degeneration is suitably chosen. For this reason, the developed meshing framework can also easily be used for error estimation and adaptive remeshing.
In interconnect structure modeling a large number of comparatively simple geometric structures with very different spatial dimensions is used, which can be seen in Figure 1. It is essential that very thin layers should not impose a lot of mesh points, since they are normally not regions of simulation. To generate these three-dimensional structures quickly and accurately, the solid modeling language Laygrid of SAP was enhanced to meet the present requirements in TCAD (tapered shapes, slanted geometries) with a constructive solid geometry based language.
Within process simulation the surface elements (for instance, moving boundary) are of great interest. On the one hand, process simulation depends strongly on the ability to refine and unrefine (coarse) the surface elements, and on the other hand, topography simulation has to be linked with interconnect simulation to simulate realistic structures. Especially topography simulation (Figure 2) can yield surface structures of arbitrary complexity containing degenerated or even faulty elements, where a subsequent surface mesh adaptation step must be performed in order to enhance the quality of the surface triangles. This reduces the number of mesh points which are not necessary for subsequent simulation steps.
Device simulation, in strong contrast to process simulation, requires direction-dependent mesh densities in some special locations to resolve highly non-linear quantities and to provide a very high quality of the mesh elements. In the example depicted in Figure 3, small and flux-aligned elements are required in the sensitive channel regions of the structure, while the element count outside this domain should be kept to a minimum to reduce computation time. Therefore the mesh generation engine can use orthogonally distributed points within special regions to generate highly isotropic and flux-aligned elements.