This monograph represents a summary of our work in the last two years in applying the method of simulated annealing to the solution of problems that arise in the physical design of VLSI circuits. Our study is experimental in nature, in that we are con cerned with issues such as solution representations, neighborhood structures, cost functions, approximation schemes, and so on, in order to obtain good design results in a reasonable amount of com putation time. We hope that our experiences with the techniques we employed, some of which indeed bear certain similarities for different problems, could be useful as hints and guides for other researchers in applying the method to the solution of other prob lems. Work reported in this monograph was partially supported by the National Science Foundation under grant MIP 87-03273, by the Semiconductor Research Corporation under contract 87-DP- 109, by a grant from the General Electric Company, and by a grant from the Sandia Laboratories.

From my B.E.E degree at the University of Minnesota and right through my S.M. degree at M.I.T., I had specialized in solid state devices and microelectronics. I made the decision to switch to computer-aided design (CAD) in 1981, only a year or so prior to the introduction of the simulated annealing algorithm by Scott Kirkpatrick, Dan Gelatt, and Mario Vecchi of the IBM Thomas 1. Watson Research Center. Because Prof. Alberto Sangiovanni-Vincentelli, my UC Berkeley advisor, had been a consultant at IBM, I re ceived a copy of the original IBM internal report on simulated annealing approximately the day of its release. Given my background in statistical mechanics and solid state physics, I was immediately impressed by this new combinatorial optimization technique. As Prof. Sangiovanni-Vincentelli had suggested I work in the areas of placement and routing, it was in these realms that I sought to explore this new algorithm. My flJ'St implementation of simulated annealing was for an island-style gate array placement problem. This work is presented in the Appendix of this book. I was quite struck by the effect of a nonzero temperature on what otherwise appears to be a random in terchange algorithm.

Introduction The exponential scaling of feature sizes in semiconductor technologies has side-effects on layout optimization, related to effects such as inter connect delay, noise and crosstalk, signal integrity, parasitics effects, and power dissipation, that invalidate the assumptions that form the basis of previous design methodologies and tools. This book is intended to sample the most important, contemporary, and advanced layout opti mization problems emerging with the advent of very deep submicron technologies in semiconductor processing. We hope that it will stimulate more people to perform research that leads to advances in the design and development of more efficient, effective, and elegant algorithms and design tools. Organization of the Book The book is organized as follows. A multi-stage simulated annealing algorithm that integrates floorplanning and interconnect planning is pre sented in Chapter 1. To reduce the run time, different interconnect plan ning approaches are applied in different ranges of temperatures. Chapter 2 introduces a new design methodology - the interconnect-centric design methodology and its centerpiece, interconnect planning, which consists of physical hierarchy generation, floorplanning with interconnect planning, and interconnect architecture planning. Chapter 3 investigates a net-cut minimization based placement tool, Dragon, which integrates the state of the art partitioning and placement techniques.

Design and optimization of integrated circuits are essential to the creation of new semiconductor chips, and physical optimizations are becoming more prominent as a result of semiconductor scaling. Modern chip design has become so complex that it is largely performed by specialized software, which is frequently updated to address advances in semiconductor technologies and increased problem complexities. A user of such software needs a high-level understanding of the underlying mathematical models and algorithms. On the other hand, a developer of such software must have a keen understanding of computer science aspects, including algorithmic performance bottlenecks and how various algorithms operate and interact. "VLSI Physical Design: From Graph Partitioning to Timing Closure" introduces and compares algorithms that are used during the physical design phase of integrated-circuit design, wherein a geometric chip layout is produced starting from an abstract circuit design. The emphasis is on essential and fundamental techniques, ranging from hypergraph partitioning and circuit placement to timing closure.

The Best of ICCAD marks the 20th anniversary of the International Conference on Computer Aided Design. This book presents a selection of papers from among the best contributions presented in ICCAD based on their impact on research and applications. The Best of ICCAD contains overview articles solicited from leading EDA researchers that comment on the historical context of the selected papers and outline their impact on follow up work. Nine leading companies including Cadence, Synopsys, Fujitsu, IBM and Magma offer "Industry Viewpoints" outlining the impact of ICCAD on their businesses. The Best of ICCAD provides an insightful reminder on how much progress has been made in EDA in the past twenty years and will be a useful tool for professionals in the field and students in the pursuit to crack the next wave of emerging EDA problems.

Simulated Annealing is a probabilistic meta-heuristic that is based on statistical mechanics: while at high temperatures molecules in a liquid move freely, the slow reduction of temperature decreases the thermal mobility of the molecules. The final state forms a pure crystal which also corresponds to a state of minimum energy. We encourage readers to explore SA in their work, mainly because it is simple and because it can yield very good results.

The VHSIC Hardware Description Language (VHDL) provides a standard machine processable notation for describing hardware. VHDL is the result of a collaborative effort between IBM, Intermetrics, and Texas Instruments; sponsored by the Very High Speed Integrated Cir cuits (VHSIC) program office of the Department of Defense, beginning in 1981. Today it is an IEEE standard (1076-1987), and several simulators and other automated support tools for it are available commercially. By providing a standard notation for describing hardware, especially in the early stages of the hardware design process, VHDL is expected to reduce both the time lag and the cost involved in building new systems and upgrading existing ones. VHDL is the result of an evolutionary approach to language devel opment starting with high level hardware description languages existing in 1981. It has a decidedly programming language flavor, resulting both from the orientation of hardware languages of that time, and from a ma jor requirement that VHDL use Ada constructs wherever appropriate. During the 1980's there has been an increasing current of research into high level specification languages for systems, particularly in the software area, and new methods of utilizing specifications in systems de velopment. This activity is worldwide and includes, for example, object oriented design, various rigorous development methods, mathematical verification, and synthesis from high level specifications. VAL (VHDL Annotation Language) is a simple further step in the evolution of hardware description languages in the direction of applying new methods that have developed since VHDL was designed.