Ant Colony Algorithm for Steiner Tree Problem in CGRA Mapping

CGRA (Coarse-Grained Reconfigurable Array) is competitive hardware architecture for streaming applications due to its excellent tradeoff between programmability and efficiency. However, the efficiency of CGRA highly depends on how applications are mapped onto it with optimized operation placement and data routing. In the routing procedure, data are transferred through different processing elements by occupying the data path so that reduce hardware consumed by routing is important. This paper presents a 2-step ant colony algorithm for the route resource sharing optimization to obtain better data transfer path for CGRA application mapping, which can be modeled as a Steiner Tree problem. By dividing the searching procedure of ants into two steps, the execution time of proposed method was 32% less than the traditional ant colony algorithm. Meanwhile, the quality of result only deteriorates 0.4%. Compared to the simple heuristic method MPH, the optimized ant colony algorithm shortens the length of data transfer path by 4-6%.