Suitable-Distribution Particle Swarm Applying To SOC Architectural Parameters Design

Recently SOC(System-on-chip) has been increasingly discussed in the field of micro-electronics. The constant increase in levels of integration and reduction in the TTM(Time to Market) has led to a big challenge to the SOC system design method based on IP(Intellectual Property).Traditional SOC design method is based on the configuration and execution of IP, which focus on mapping an application onto a previously designed SOC architecture with using IP as component and realize the design by configurating the parameters of IP. Due to the varieties of IP and the conflict of SOC optimal objective (power and executive time), the SOC design space is very complex. According to a certain application, searching a set of feasible IP configurations for SOC is the main task of the design and the whole process is more like finding solutions for multi-objective problem. A novel multi-objective optimization, suitable-distribution particle swarm (SDPS), will be proposed in this paper. After verifying the validity of SDPS, we apply SDPS to SOC architecture parameters design.SDPS maintain the characters of PSO such as quick convergence, easy calculation and good generalization. In term of multi solutions instead of single one for the Multi-objective problem, good algorithm is able to not only make the particles quick converge to the Pareto set, but also maintain the diversity of the final solutions, which should representatively distribute along the Pareto front. In that case, we combine suitable-distribution method (SDM) with PSO and SDM can ensure the final solutions more dispersive and symmetrical. SDM includes suitable-distribution value (SDV) and suitable-distribution radius (SDR). The definition of SDV and SDR will be presented in the paper. Dominance is no longer the only criterion for particles to enter the elitism set, but the particle with high SDV has great possibility to enter and be kept in the elitism set. SDR is proposed to guarantee the diversity of the particles in the elitism set and make them well distributed in the objective space. By comparing with normal multi-objective particle swarm optimization (MOPSO) and nondominated sorting genetic algorithm(NSGA-II), SDPS shows its advantage in solving multi-objective problem.SDPS is applied to SOC architectural parameters design and a triple discretization method is used to discretize the velocity vector of the particles. Comparing SDPS with parameters interdependence method, the experimental result indicates that, within the two dimension objective space composed of power consumption and executive time cost, the configurations gained by SDPS have well distribution and nondominance and SDPS is very efficient in SOC architectural parameters searching.