Design and Prototyping of Hard Real Time Systems
CALIFORNIA UNIV BERKELEY ELECTRONICS RESEARCH LAB
Pagination or Media Count:
Recent advances in processing technology create the need of modeling physical phenomena described in terms of three-dimensional geometries. For instance, submicron technology for MOS devices requires the accurate modeling of narrow-chennel effects which, in turn, imply the use of 3-D discretization. Three-dimensional effects must be modeled also in advanced bipolar structures. However, including these new effects yields an enormous increase in CPU time if the present 2-D algorithms are extended to cover the three-dimensional case on a conventional processor. Large vector supercomputers have been used to simulate this class of problems 1 a three-dimensional device simulator was developed based on a seven-point finite-difference discretization and the biconjugate gradient method was used to solve the non-symmetric linear system arising from the previous discretization. The vectorization provided a speed-up of sixteen for typical problems. Since supercomputers offer such a limited degree of parallelism, other approaches are under investigation to efficiently solve very large simulations. Recently, massively parallel algorithms have been used for linear capacitance evaluation in three-dimensional structures showing good computational performance 2. In this work we present a full 3D device simulator developed on a Connection Machine 3. The CM is a massively parallel SIMD computer with up to 65,536 bit serial processors. Each processor has 64 Kbits of memory. Communication is either a fixed distance on an N-dimensional grid, or direct to an arbitrary processor based on a hypercube.
- Computer Hardware