КО-ЭВОЛЮЦИОННЫЙ АЛГОРИТМ РАЗМЕЩЕНИЯ НА ОСНОВЕ ВЗАИМОДЕЙСТВИЯ СУБПОПУЛЯЦИЙ, ОТЛИЧАЮЩИХСЯ СТРАТЕГИЯМИ ПОИСКА

Abstract

A new methodology and method for placing VLSI elements has been developed, which differin that the solution of the placement problem is based on the use of a fixed order of position selection,focused on the effective solution of the placement problem, and a heuristic procedure fordistributing elements by positions, which reduces the overall complexity and improves the qualityof the solution. The process of forming a list of positions on the switching field is carried out usingthe mechanisms of the wave algorithm. The choice of the final list is based on the principle of constructinga route with a minimum estimate of the total linear length of distances between routepositions. To solve the placement problem, a search algorithm based on the modified ant colonymethod has been developed. To exclude premature convergence and localization of the globalextremum of the problem, the development of the algorithm was carried out on the basis of the coevolutionaryapproach. The architecture of the co-evolutionary placement algorithm is developedon the basis of the ant colony algorithm paradigm. In the search space, sub-populations implementfour optimization strategies in parallel. In the work, the coevolution process is implemented on thebasis of the interaction of subpopulations that differ in search strategies. A distinctive feature ofthe co-evolutionary approach used is that subpopulations of solutions are actually virtual. Theprocess of co-evolution is implemented by one population of agents Z by sequential formation andmerging of virtual subpopulations of solutions into one population. In this paper, the solution ofthe placement problem is aimed at improving traceability by minimizing the resources required toimplement connections. A significant contribution to minimizing the spatial and temporal complexityof the search procedure was made by: the use by virtual sub-populations of a common evolutionarymemory, a common solution search graph, the formation of a single interpretation of thesolution in the form of a route on a complete directed graph with binary directed edges. Testingwas carried out on benchmarks 19s, PrimGA1, PrimGA2. The results compared to existing algorithmsare improved by 7-8%. The probability of obtaining a global optimum was 0.96. On average,solutions differ from the optimal by less than 1.5%. The time complexity of the algorithm forfixed values of the population size and the number of generations is O(n). The total time complexityof the hybrid algorithm is O(n2)−O(n3).