(A) delayed binding of design decisions dor synthesizing the data-paths in digital system based on path-search algorithm

Abstract

The major obstacle in the data path synthesis problem is interactions between synthesis subtasks. Interactions do not allow a foresight of the final synthesis result to estimate an effect of local design decision of any subtask. And the other obstacle is the size of problem space for getting the best implementation with minimum execution time and minimum area. It is so large and thus very hard to search the space efficiently. The conventional techniques such as iterations or heuristics tend to find the local optimum solution. This thesis describes a delayed binding of design decisions. It did not bind the design decisins at each synthesis subtask but normally defer important design decisions. An efficient method to synthesize the data paths of digital system ffrom behavioral specification is presented. This method is based on path-search algorithm that finds groups of compatible variables(CVGs) which are existing along proper paths in a modified data-dependency graph(MDDG). It does a limited search through the space of the proper paths in MDDG. It is proved that the upper bound on the path number of a MMDG is $O(n^2)$, where n is the number of nodes in MDDG. Thus, searching the proper paths of a MMDG can not be blown-up by the massive path number. The pathsearch algorithm is also applies to the behavioral specification with conditional statements or loop statements. The algorithm for incorporating with the results of the path-search algorithm is conflict-analysis phase. During this phase, two types of resource foldings, memory and functional resource folding, are performed simultaneously. This resource folding technique is independent of the number of resources. Finally, the bus-style synthesis algorithm is presented. It is conceptually simple and results in the implementation with low cost and short delay. The approach taken here has been tested for typical examples which were taken from HAL, Facet, Flamel, and etc. From these results, the proposed method perf...