Cluster-based parallel 3-D Monte Carlo device simulation

The recent increase in the performance of commodity computer hardware, along with the constant cost reductions, has created a very favorable medium for building cluster based distributed parallel machines. Such machines have been demonstrated to reach performance levels of supercomputers designed just a few years ago. During the same period, the developments in parallel programming techniques, standardization of message passing libraries and their interfaces with high-level programming languages made it possible to use distributed parallel computers for solving computationally very intensive problems. The uninterrupted reduction of device dimensions has highlighted issues such as granularity of doping, actual shape of device boundaries in the third dimension, and proper modeling of carrier-carrier interactions due to very high fields in such devices. Also a new interest in development of inherently 3-D dimensional devices, due to performance and integration density considerations has been created. We have designed and implemented a cluster based parallel 3-D Monte Carlo simulation core for Si semiconductor devices. The main focus has been on enabling the simulator to perform well on a cluster of commodity computers. The networking hardware used on such a distributed computer typically leads to a significantly larger communications cost (larger latency and reduce bandwidth) than that of a specially built parallel computer. For this reason code has been designed in such a way to minimize unstructured communication requirements among nodes.