Date of Award
Doctor of Philosophy
School of Engineering and Mathematics
Faculty of Communications, Health and Science
Professor Kamran Eshraghian
Dr Daryousb Habibi
Very Large Scale Integrated-circuit (VLSI) routing involves many large-size and complex problems and most of them have been shown to be NP-hard or NP-complete. As a result, conventional approaches, which have been successfully used to handle relatively small-size routing problems, are not suitable to be used in tackling large-size routing problems because they lead to 'combinatorial explosion' in search space. Hence, there is a need for exploring more efficient routing approaches to be incorporated into today's VLSI routing system. This thesis strives to use intelligent approaches, including symbolic intelligence and computational intelligence, to solve three VLSI routing problems: Three-Dimensional (3-D) Shortest Path Connection, Switchbox Routing and Constrained Via Minimization. The 3-D shortest path connection is a fundamental problem in VLSI routing. It aims to connect two terminals of a net that are distributed in a 3-D routing space subject to technological constraints and performance requirements. Aiming at increasing computation speed and decreasing storage space requirements, we present a new A* algorithm for the 3-D shortest path connection problem in this thesis. This new A*algorithm uses an economical representation and adopts a novel back- trace technique. It is shown that this algorithm can guarantee to find a path if one exists and the path found is the shortest one. In addition, its computation speed is fast, especially when routed nets are spare. The computational complexities of this A* algorithm at the best case and the worst case are O(Ɩ) and 0(Ɩ3), respectively, where Ɩ is the shortest path length between the two terminals. Most importantly, this A' algorithm is superior to other shortest path connection algorithms as it is economical in terms of storage space requirement, i.e., 1 bit/grid. The switchbox routing problem aims to connect terminals at regular intervals on the four sides of a rectangle routing region. From a computational point of view, the problem is NP-hard. Furthermore, it is extremely complicated and as the consequence no existing algorithm can guarantee to find a solution even if one exists no matter how high the complexity of the algorithm is. Previous approaches to the switch box routing problem can be divided into algorithmic approaches and knowledge-based approaches. The algorithmic approaches are efficient in computational time, but they are unsucessful at achieving high routing completion rate, especially for some dense and complicated switchbox routing problems. On the other hand, the knowledge-based approaches can achieve high routing completion rate, but they are not efficient in computation speed. In this thesis we present a hybrid approach to the switchbox routing problem. This hybrid approach is based on a new knowledge-based routing technique, namely synchronized routing, and combines some efficient algorithmic routing techniques. Experimental results show it can achieve the high routing completion rate of the knowledge-based approaches and the high efficiency of the algorithmic approaches. The constrained via minimization is an important optimization problem in VLSI routing. Its objective is to minimize the number of vias introduced in VLSI routing. From computational perspective, the constrained via minimization is NP-complete. Although for a special case where the number of wire segments splits at a via candidate is not more than three, elegant theoretical results have been obtained. For a general case in which there exist more than three wire segment splits at a via candidate few approaches have been proposed, and those approaches are only suitable for tackling some particular routing styles and are difficult or impossible to adjust to meet practical requirements. In this thesis we propose a new graph-theoretic model, namely switching graph model, for the constrained via minimization problem. The switching graph model can represent both grid-based and grid less routing problems, and allows arbitrary wire segments split at a via candidate. Then on the basis of the model, we present the first genetic algorithm for the constrained via minimization problem. This genetic algorithm can tackle various kinds of routing styles and be configured to meet practical constraints. Experimental results show that the genetic algorithm can find the optimal solutions for most cases in reasonable time.
Tang, M. (2000). Intelligent approaches to VLSI routing. Retrieved from https://ro.ecu.edu.au/theses/1375