In reconstructing 3-D from images based on feature points, one usually defines a triangular mesh that has these feature points as vertices and displays the scene as a polyhedron. If the scene itself is a polyhedron, however, some of the displayed edges may be inconsistent with the true shape. This paper presents a new technique for automatically eliminating such inconsistencies by using a special template. We also present a technique for removing spurious occluding edges. All the procedures do not require any thresholds to be adjusted. Using real images, we demonstrate that our method has high capability to correct inconsistencies.

A three-dimensional mesh generation method in which triangulation of the domain boundary is performed first is desirable since such a method would make it easier to achieve the requirements for the mesh around the boundary. We have developed a mesh generator for a 3D device simulator based on this approach. This mesh generator recursively subdivides a box that includes the whole domain into smaller boxes (cells), a method known as the octree technique. Although our mesh generator is similar to previously reported mesh generators in the sense that it utilizes recursive subdivision of elements, its major difference is that it constructs a triangular mesh upon boundaries of the domain first and this triangular mesh is not changed in the following processes. In order to generate a mesh suitable for the control volume method, a "forbidden region" is introduced and mesh points in the domain are allocated outside of this region. Since the triangular mesh is determined prior to tessellation of the domain, this method is suitable for handling layered mesh along the boundary, which is often necessary to estimate large flows parallel to the boundary precisely. A simple method to provide a layered mesh for a planar boundary is incorporated into the mesh generator. This mesh generator is integrated within our in-house three-dimensional device simulation system. The simulator's practicality is demonstrated through analysis of the reverse narrow channel effect for MOSFETs with LOCOS isolation structures. The effect of protection of the boundary by the layered mesh is also examined by calculating Id-Vg characteristics of a MOSFET with an oblique Si surface, and it is shown that protection of the whole surface of the channel region is necessary to estimate drain current correctly.

A surface extraction algorithm with NURBS has been developed for the mesh generation from the scattered data after a cell-based simulation. The triangulation of a surface is initiated with a step of describing the geometry along the polygonal boundary with multiple points. In this work, an NURBS surface can be generated with scattered data for each polygonal surface by employing a multilevel B-spline surface approximation. The NURBS mesh in accordance with our algorithm excellently represents the surface evolution of the topography on the wafer. A dynamically allocated topography model, so-called cell advancing model, is proposed to resolve an extensive memory requirement for the numerical simulation of a complicated structure on the wafer. A concave cylindrical DRAM cell capacitor was chosen to test the capability of our model. A set of capacitance present in the cell capacitor and interconnects was calculated with three-dimensional tetrahedral meshes generated from the NURBS surface on CRAY T3E supercomputer. A total of 5,475,600 (130 156 270) cells was employed for the simulation of semiconductor regions comprising four DRAM cell capacitors with a dimension of 1.3 µm 1.56 µm 2.7 µm . The size of the required memory is about 22 Mbytes and the simulation time is 64,082 seconds. The number of nodes for the FEM calculation was 70,078 with 395,064 tetrahedrons.

After a brief discussion of the demands in meshing for semiconductor process and device simulation, we present a three-dimensional Delaunay refinement technique combined with a modified advancing front algorithm.

A simple and efficient adaptive mesh generation for the approximate scalar analysis of optical waveguides is proposed. Two types of local weight estimates which can take into account both a field amplitude and its variation on a problem domain are introduced. One is a difference between linear and quadratic element solutions and the other is a residual for the partial differential equation to be solved. To show the validity and usefulness of the present scheme, the guided-mode analysis of a rib waveguide and the beam propagation analysis of a tilted slab waveguide and a Y-branching rib waveguide are performed.

In order to achieve an efficient and reliable prediction of device performance by numerical device simulation, a discretization mesh must be generated with an adequate, but not redundant, density of mesh points. However, manual mesh optimization requires user's trial and error. This task annoys the user considerably, especially when the device operation is not well known, or the required mesh-point density strongly depends on the bias condition, or else the manipulation of the mesh is difficult as is expected in 3D. Since these situations often happen in designing advanced VLSI devices, it is highly desirable to automatically optimize the mesh. Adaptive meshing techniques realize automatic optimization by refining the mesh according to the discretization error estimated from the solution. The performance of mesh optimization depends on a posteriori error indicators adopted to evaluate the discretization error. In particular, to obtain a precise terminal-current value, a reliable error indicator for the current continuity equation is necessary. In this paper, adaptive meshing based on the current continuity equation is investigated. A heuristic error indicator is proposed, and a methodology to extend a theoretical error indicator proposed for the finite element method to the requirements of device simulation is presented. The theoretical indicator is based on the energy norm of the flux-density error and is applicable to both Poisson and current continuity equations regardless of the mesh-element shape. These error indicators have been incorporated into the adaptive-mesh device-simulator HFIELDS, and their practicality is examined by MOSFET simulation. Both indicators can produce a mesh with sufficient node density in the channel region, and precise drain current values are obtained on the optimized meshes. The theoretical indicator is superior because it provides a better optimization performance, and is applicable to general mesh elements.