Application of Parallel Adaptive Computing Technique to Polysilicon Thin-Film Transistor Simulation

Abstract

In this paper, parallel adaptive finite volume simulation of polysilicon thin-film transistor (TFT) is developed using dynamic domain partition algorithm on a PC-based Linux cluster with message passing interface libraries. A set of coupled semiconductor device equations together with a two-dimensional model of grain boundary for polysilicon TFT is formulated. For the numerical simulation of polysilicon TFT, our computational technique consists of the Gummel’s decoupling method, an adaptive 1-irregular meshing technique, a finite volume approximation, a monotone iterative method, and an a posteriori error estimation method. Parallel dynamic domain decomposition of adaptive computing technique provides scalable flexibility to simulate polysilicon TFT devices with highly complicated geometry. This parallel approach fully exploits the inherent parallelism of the monotone iterative method. Implementation shows that a well-designed load balancing simulation significantly reduces the execution time up to an order of magnitude. Numerical results are presented to show good efficiency of the parallelization technique in terms of different computational benchmarks.