Nanometer VLSI placement and optimization for multi-objective design closure

Access full-text files




Luo, Tao, Ph. D.

Journal Title

Journal ISSN

Volume Title



In a VLSI physical synthesis flow, placement directly defines the interconnection, which affects many other design objectives, such as timing, power consumption, congestion, and thermal issues. With the scaling of technology, the relative interconnect delay increases dramatically. As a result, placement has become a bottleneck in deep sub-micron physical synthesis. In this dissertation, I propose several optimization algorithms from global placement, placement migration, timing driven placements, to incremental power optimizations for multi-objective VLSI design closure. The first work is DPlace, a new global placement algorithm that scales well to the modern large-scale circuit placement problems. DPlace simulates the natural diffusion process to spread cells smoothly over the placement region, and uses both analytical and discrete techniques to improve the wire length. However, global placement is never sufficient for multi-objective design closure, a variety of design objectives have to be improved incrementally, such as timing, routing congestion, signal integrity, and heat distribution. Placement migration is a critical step to address the cell overlaps appearing during incremental optimizations. To achieve high placement stability, I propose a computational geometry based placement migration flow to cope with placement changes, and a new stability metric to measure the “similarity” between two placements accurately. Our placement migration algorithm has clear advantage over conventional legalization algorithms such that the neighborhood characteristics of the original placement are preserved. For timing closure in high performance designs, I present a linear programming based incremental timing driven placement to improve the timing on critical paths directly. I further present an efficient timing driven placement algorithm (Pyramids). Two formulations of Pyramids are proposed, which are suitable for different optimization stages in a physical synthesis flow. Both approaches find the optimal location for timing of a cell in constant time, through computational geometry based approaches. For fast convergence of design closure, placement should be integrated with other optimization techniques. I propose to combine placement, gate sizing and Vt swapping techniques to reduce the total power consumption, especially the leakage power, which is becoming increasingly critical for nanometer VLSI design closure.