A heuristic for channel routing

Abstract

In designing the layout of VLSI chips, channel routing plays a central role. A frequently used channel routing mode is known as Manhattan routing, where a two-layer channel is used to route a specified set of nets between two rows of terminals. If the net list contains long runs, the only way to reduce the number of channel tracks is by splitting horizontal segments of nets (jogging).

A new algorithm is proposed which uses a preprocessing to decide, before producing the routing, which nets must be split and where. Jogging a net effectively splits a run into two runs with approximately the same total length. The objective is achieved by finding a parameter m which represents the maximum desirable length for a run and minimizing the number of runs with length greater than m. This minimization is obtained by finding a minimum-cost matching in a weighted bipartite graph G _m = (V, E). Node set V is partitioned in two sets: J and R, where J is a set of jogging requests associated to nets chosen as possible candidates for joggings while nodes in R are associated to channel columns. A heuristic is proposed to assign priorities to vertices of J and costs to vertices of R. The running time of the algorithm is O(t·n·logn) +O(n✓n), where n is the number of nets and t is the number of tracks used. Quite satisfactory experimental results are also reported.

This work has been supported in part by Ministero della Pubblica Istruzione of Italy and by the Semiconductor Research Corporation under Contract 87-DP109.