Search

A new approach to high level synthesis with micro-rollback and self-recovery method is presented. The objective of this approach is to generate a reliable system from the behavioral descriptions. The approach is formulated as a rollback point insertion problem based on the probability function of being inserted rollback points after the given CDFG is scheduled. This function allows us not only to search the problem space efficiently but also to avoid the exhaustive search.

In this paper, we propose a new heuristic scheduling algorithm based on the statistical analysis of the cumulative frequency distribution of operations among control steps. It has a tendency of escaping from local minima and therefore reaching a globally optimal solution. The presented algorithm considers the real world constraints such as chained operations, multicycle operations, and pipelined data paths. The result of the experiment shows that it gives optimal solutions, even though it is greedy in nature.