Behavior-to-placed RTL synthesis with performance-driven placement

Interconnect delay should be considered together with computation delay during architectural synthesis in order to achieve timing closure in deep submicrometer technology. In this paper, we propose an architectural synthesis technique for distributed-register architecture, which separates interconnect delay for data transfer from component delay for computation. The technique incorporates performance-driven placement into the architectural synthesis to minimize performance overhead due to interconnect delay. Experimental results show that our methodology achieves performance improvement of up to 60% and 22% on the average.