Performance driven multi-layer general area routing for PCB/MCM designs

In this paper we present a new global router appropriate for Multichip Module (MCM) and dense Printed Circuit Board (PCB) design, which utilizes a hybrid of the classical rip-up and reroute approach, and the more recent iterative deletion[9] method. The global router addresses performance issues by utilizing recent results in high performance interconnect design, while still effectively minimizing global congestion. With experiments on the maze-routing component of our global router, we show that the choice of routing cost functions can have a significant impact on final solution quality. The results of a number of previously proposed routers may be improved dramatically by adopting the cost functions we suggest here. We also find little evidence of the "net ordering problem" when our cost functions and routing model are applied. The iterative deletion method is shown to improve global solution quality, particularly when high performance interconnect is required. We evaluate the performance of our global router by comparing the congestion of routes produced by our global router to those of a well known MCM router, V4R [14]. Our global router, MINOTAUR, supports arbitrary numbers of routing layers, differing capacities for each layer, preexisting congestion and obstacles, and; high performance interconnect structures (including those which require variable width interconnect).