Crosstalk-Aware Placement

摘要

With today's rapidly shrinking process geometries, designers must address crosstalk, electromigration, IR drop, and other effects earlier in the design cycle to achieve signal integrity. The authors present a new placement algorithm that minimizes crosstalk and increases design speed 8% on average, in comparison with a traditional timing-driven, congestion-aware placement flow. Advances in semiconductor fabrication technology, such as smaller minimum feature sizes, increased metal layers, and stacked vias, have contributed to the fast and steady enhancement of VLSI circuit performance and density for more than two decades. However, for VLSI designs in processes under 0.18 micron, electrical effects such as crosstalk, electromigration, wire self-heating, and IR drop negatively affect signal integrity, which is critical to the performance and reliability of electronic systems. Of these problems, crosstalk induced by capacitive coupling is the one back-end vendors see most often. Crosstalk typically occurs between two adjacent wires when their cross-coupling capacitance is large enough to influence one another's electrical characteristics. The two major effects of crosstalk are delays (which change signal propagation time and thus can cause setup or hold time failures) and glitches (which can cause voltage spikes on wires, resulting in false logic behavior). Researchers have proposed models for extracting RCL networks from postrouting circuits. Others have proposed optimization techniques that reduce crosstalk effects after high-coupling nets are identified. However, most optimization techniques work with a postrouting circuit, so their effectiveness is limited. For example, if the initial placement introduces many potential couplings, the router cannot fix all of them without violating design constraints. For such cases, we need new techniques that prevent signal net coupling earlier in the design cycle. Although other factors such as driver strength and transition time also affect noise, coupling capacitance dominates the peak noise computation. Here, we present a crosstalk-aware placement algorithm that reduces signal net coupling capacitance. This algorithm uses a novel probabilistic model to estimate coupling capacitance at the placement stage and controls placement density to reduce the number of highly coupled regions.