Congestion-aware logic synthesis

A predictive distributed congestion metric with application to technology mapping

“…Assigning the congestion cost to the nets in the congested bins accounts for this discrete nature of routability and also allows the mapper to select area-optimal matches in the sparsely congested regions. Both of these purposes are critical and are served by our algorithm, while previous approaches [8], [9] have not addressed these.…”

“…This approach may result in long wires in the final netlist, and more importantly, may leave a significant portion of the design space unexplored. Pandini et al proposed wirelength as a metric to be minimized during technology mapping in order to reduce the congestion [9]. Although large wirelength may be correlated with high congestion, the correlation is rather poor, and therefore, the mapping based on such a metric may not result in an effective optimization.…”

“…Instead of applying an indirect metric such as wirelength [9], [11], we utilize probabilistic congestion estimates [12] to guide our technology mapping; these estimates are shown in [12] to have good fidelity with post-routing congestion measurements. Experimental results due to congestion-aware technology mapping algorithms on an industrial benchmark, ISCAS'85, and ITC'99 circuits show an improvement of 37% and 20%, on an average, in track overflows as compared to conventional mapping for area and delay minimization, respectively.…”

A predictive distributed congestion metric and its application to technology mapping

“…In case of Figure 4, wires w1 and w2 will have different congestion costs even though the shortest routes in both the cases may have the same length; the congestion cost of w1 will be zero and the congestion cost of w2 will have a positive value as its bounding box contains congested bins. The following equation captures this causality relation between routability and congestion while computing the congestion cost of a net, Ó×Ø Ò Ø , Ó×Ø Ò Ø ¦ Ò¾ ÓÙÒ Ò ÓÜ´Ò Øµ ´ Òµ Ñ Ü Ò Ò Ø (3) where ´ Òµ is the congestion in a bin, Ñ Ü is the threshold …”

“…This approach may result in long wires in the final netlist, and more importantly, may be so limiting as to leave a significant portion of the design space unexplored. Pandini et al proposed wirelength as a metric to be minimized during technology mapping in order to reduce the congestion [3]. Although large wirelength may be correlated with high congestion, the correlation is rather poor, and therefore, this may not result in an effective optimization.…”