MAPPING DYNAMIC-PROGRAMMING ONTO MODULAR LINEAR SYSTOLIC ARRAYS

In this paper we propose a novel way of deriving a family of fully-pipelined linear systolic algorithms for the computation of the solutions of a dynamic programming problem. In many instances, modularity is an important feature of these algorithms. One may simply add more processors to the array as the size of the problem increases. Each cell has a fixed amount of local storage alpha and the time delay between two consecutive cells of the array is constant. The time complexity and the number of cells in our array tend to n2 + O(n) and n2/alpha + O(n), respectively, as alpha increases. This represents the best known performance for such an algorithm.