High-Performance Placement and Routing for the Nanometer Scale.

Abstract

Modern semiconductor manufacturing facilitates single-chip electronic systems that only five years ago required ten to twenty chips. Naturally, design complexity has grown within this period. In contrast to this growth, it is becoming common in the industry to limit design team size which places a heavier burden on design automation tools.

Our work identifies new objectives, constraints and concerns in the physical design of systems-on-chip, and develops new computational techniques to address them. In addition to faster and more relevant design optimizations, we demonstrate that traditional design flows based on ``separation of concerns'' produce unnecessarily suboptimal layouts. We develop new integrated optimizations that streamline traditional chains of loosely-linked design tools. In particular, we bridge the gap between mixed-size placement and routing by updating the objective of global and detail placement to a more accurate estimate of routed wirelength. To this we add sophisticated whitespace allocation, and the combination provides increased routability, faster routing, shorter routed wirelength, and the best via counts of published techniques. To further improve post-routing design metrics, we present new global routing techniques based on Discrete Lagrange Multipliers (DLM) which produce the best routed wirelength results on recent benchmarks. Our work culminates in the integration of our routing techniques within an incremental placement flow to improve detailed routing solutions, shrink die sizes and reduce total chip cost.

Not only do our techniques improve the quality and cost of designs, but also simplify design automation software implementation in many cases. Ultimately, we reduce the time needed for design closure through improved tool fidelity and the use of our incremental techniques for placement and routing.