Research And Application Of Kinetic Monte Carlo Parallel Algorithm Based On The Heterogeneous Many-core Platform

Nowadays,as the computer system continues becoming heterogeneous many-core structures,its computing ability has also been continuously broken through and improved.Accordingly it also provides an opportunity for the research of large-scale and computation-demanding computing problems in computing physics and other fields.So,Monte Carlo method,including kinetic Monte Carlo method,which is a very important simulation tools,is of great significance for the research and implementation of its parallel algorithm.In this background,this thesis has studied the kinetic Monte Carlo parallel algorithm,relies on the many-core platform built by the lab.We studied and realized the parallel algorithm based on the thin film growth,and then applied it into the process of self-assembly of block copolymer on nanopatterned sub strates.In detail,the research work and results including the following parts:i.The problem and challenges of kinetic Monte Carlo algorithm are summarized.Among that,the traditional algorithm handling one event-at-a-time is the main difficulty in the parallelization.Hence,two relative parallel algorithms are introduced with the difference whether the advanced time is synchronized or not:Hybrid asynchronous algorithm and Null-event synchronous parallel algorithm.Combining the summary of relevant literature and results of two algorithms by pseudo-parallel implementation,it's considered that the null-event synchronous parallel algorithm is more suitable,and so we carried out the subsequent work on this method.ii.Based on the thin film growth process,the parallel simulation is implemented by OpenMP with periodic boundary conditions and others details.Comparing the result of parallel computing with that of serial,we believe that the parallel algorithm can maintain the correctness and dynamics of the process.As for the parallel performance,we found that the block division will increasing the time advancement step.Accordingly,a speedup ration formula for this algorithm is proposed,and so the algorithm actual speedup can reach around 37 with 60 threads,iii.In the background that the semiconductor manufacturing is in the bottleneck and the block copolymer self-assembly technology becomes an alternative,we applied the kinetic Monte Carlo parallel algorithm to this simulation.We proposed a solution of using the statistical mean value as the rate approximation of different class states,to solve the difficulty of calculating global event ratesiv.Parallel simulation of block copolymer self-assembly process is also implemented,and the speed up reach around 40 with 64 threads.The concept of weighted image information entropy is proposed and applied to verify that the algorithm still has good dynamics.Finally,with the advantages of parallel computing,we studied the effects and changes of pattern formed by block copolymer self-assembly on Nano-substrates under different simulation conditions.