Ken Wang
wang@gloworm.stanford.edu


Abstract:

The need for improved computational performance in process modeling tools is being driven primarily by two factors: increasing diffusion model complexity and growing grid sizes. Since the ALAMODE program provides both flexibility in model specification (allowing models of arbitrary complexity) as well as dimension independence (allowing 1D, 2D or 3D grids), ALAMODE users are particularly well-positioned to leverage enhanced computational performance. This work focuses on computational methods and parallel algorithms for finite element analysis in the context of the ALAMODE program.

Progress:

Recent work has concentrated on the integration of multiple linear solvers into the ALAMODE framework using a generic solver interface, allowing users to choose from a variety of solution algorithms. A parallel multi-frontal solver has been integrated. Encapsulation of the PETSc 2.0 parallel solver package is currently in progress. Preliminary solver performance benchmarks have been conducted using a computationally intensive 9-species diffusion model which incorporates the effect of {311} defect aggregation over time.

Publications & Presentations this Quarter:

Preliminary solver performance results from the above-mentioned diffusion model (internal report).

Trips: None.