Parallel Monte Carlo simulation of multi-lattice thin film growth

This paper describes a new parallel algorithm for the multi-lattice Monte Carlo atomistic simulator for thin film deposition (ADEPT), implemented on a parallel computer using the PVM (Parallel Virtual Machine) message passing library. This parallel algorithm is based on domain decomposition with overlapping and asynchronous communication. Multiple lattices are represented by a single reference lattice through one-to-one mappings, with resulting computational demands being comparable to those in the single-lattice Monte Carlo model. The Multi-lattice Monte Carlo model of deposition includes two basic events: The deposition event and surface diffusion event. Since the diffusion event constitutes more than 90% of the total simulation time, we concentrate on the implementation of a new parallel diffusion simulation aiming to reduce the communication overhead during simulation. Asynchronous communication and domain overlapping techniques are used to reduce the waiting time and the communication time among parallel processors. Results show that the algorithm is highly efficient with large number of processors. The algorithm was implemented on a parallel machine with 50 processors, and it is suitable for parallel Monte Carlo simulation of thin film growth in either a distributed memory parallel computer or a shared memory machine with message passing libraries.