Performance-Optimal Behavioral Synthesis with Degenerable Compound Functional Units

Abstract

Since utilizing compound functional units (e. g., multiplier-accumulate) designed with shorter delay and/or smaller area than cascaded basic functional units is a well-known technique in system design, this paper presents an ILP-based approach for performance-driven behavioral synthesis with compound functional units. The algorithm maximizes the performance of a design under resource constraints by deriving all possible degenerative forms of compound functional units and by maximizing the utilization of resources. The proposed method exploits patterns acquired from available resource types to cover the target design optimally, and then guarantees that the synthesis result is with minimal latency. Experimental results show that our approach achieves significant performance improvement as compared with the prior art.