Skip to Main Content
Wire pipelining emerges as a new necessity for global wires due to increasing wire delay, shrinking clock period and growing chip size. Existing approaches on wire pipelining are mostly based on edge triggered flip-flops. In this paper, we demonstrate the advantages of using level sensitive latches in terms of both latency and area cost. The input-output timing coupling and the strict short path constraint for latches demand additional design elaborations compared with flip-flops. New approaches are proposed in this work to solve these difficulties so that the advantages of latches can be fully utilized. In particular, a deferred delay padding technique is developed to correct short path violations with the minimal extra cost. These techniques are integrated with a dynamic programming based concurrent synchronous element and repeater insertion framework. Experimental results confirm the advantages of using latches as well as effectiveness of our algorithms.