In the simulation of semiconductor processes and devices it may be necessar
y to generate surface parallel meshes. One important example occurs in MOS
transistors where the electrons flow along the silicon surface underneath a
gate. It is desired beneficial in terms of accuracy to have rather long me
sh edges parallel and rather small edges orthogonal to those currents. For
most of the devices quadtree techniques have been used with big success (Ga
rreton G. A hybrid approach to 2D and 3D mesh generation for semiconductor
device simulation. PhD Thesis, Integrated Systems Laboratory, ETH Zurich, 1
999. Garreton G, Villablanca L, Strecker N, Fichtner W. Unified grid genera
tion and adaptation for device simulation. Proceedings of SISDEP'95, Erlang
en, Germany, 6-8 September 1995; 6:468-471.) If the interface is not axis a
ligned, however, a quadtree-based approach does not generate meshes of this
quality, resulting in a larger numerical error or in convergence problems
during equation solution.
We present here a modified advancing front grid generator that inserts surf
ace parallel mesh lines; the interior of the region is filled with layers o
f nearly rectangular quadrilaterals, and not triangles as in conventional a
dvancing front generators (see George FL, Sveno E. The advancing front mesh
generation method revisited. international Journal for Numerical Methods i
n Engineering 1994; 37:3605-3619 and Schoberl. Computing and Visualization
in Science 1997; 1:41-52). Here we follow references of Johnston BP, Sulliv
an JM. Fully automatic two dimensional mesh generation using normal offsett
ing. International Journal for Numerical Methods in Engineering 1992; 33:42
5-442; Blacker TD, Stephenson MB. Paving: a new approach to automated quadr
ilateral mesh generation. international Journal for Numerical Methods in En
gineering 1991; 32:811; Rees M. Combining quadrilateral and triangular mesh
ing using the advancing front approach. Proceedings of the 6th Internationa
l Meshing Roundtable 1997; 337-348; White DR, Rinney P. Redisign of the pav
ing algorithm: robustness enhancements through element by element meshing.
Proceedings of the 6th International Meshing Roundtable 1997; 323-335, but
we use a different point location scheme, in the sense that the opposite ed
ge of the quadrilateral is kept parallel if possible. At each layer the mar
ching distance is increased by a coarsening factor; refinement is therefore
controlled by the initial marching distance and the coarsening factor. A m
aximum edge length is guaranteed.
The generation of offsetting layers stops when the front intersects itself.
The remaining polygon is triangulated. As a final step the mesh is convert
ed to a Delaunay conforming mesh by swapping edges and inserting points.
The implementation in two dimensions has been tested successfully using rea
listic examples from device simulations. Copyright (C) 2000 John Wiley & So
ns, Ltd.