Hybrid Design Space Exploration Methodology for Application Specific System Design

Abstract

During the application specific system design process, the designer has to take early decisions for selecting the optimal system components from the available huge design alternatives. To obtain the optimal design configuration from the available design alternatives, an efficient Design Space Exploration (DSE) process is required. This paper extends our previous work by integrating Bayesian Belief Network (BBN) based design space pruning methodology with the proposed heuristic algorithm for appropriate selection of memory configuration during the system design process. A complete and well structured DSE strategy has been formulated through the combination of BBN and heuristic approaches. The BBN performs design space pruning from the available huge design alternatives, resulting in near Pareto-optimal solution. The proposed heuristic algorithm performs the selection of the most optimal cache options. This paper mainly focuses on integrating the BBN with the proposed heuristic algorithm for providing efficient DSE strategy that aids the system designers during the application specific system design process. The experimental results in support of the proposed heuristic show a considerable reduction in the number of simulations required for covering the design space and also the algorithm finds the most optimal cache configurations for the given application with less number of iterations.