Crosstalk-Aware Signal Probability-Based Dynamic Statistical Timing Analysis

Yao Chen1,  Andrew Kahng2,  Bao Liu1,  Wenjun Wang1
1University of Texas at San Antonio, 2University of California, San Diego


Abstract

Crosstalk is an increasingly significant effect for VLSI timing performance. Traditional STA or SSTA techniques provide pessimistic crosstalk analysis based on timing window envelopes. In this paper, we present input-aware signal probabilitybased statistical timing analysis (SPSTA) taking crosstalk-induced delay variations into account. SPSTA achieves reduced pessimism and improved accuracy by signal propagation path/networkbased timing analysis and leveraging some existing automatic test pattern generation (ATPG) techniques. Our experiments based on the 45nm Nangate open cell library show that compared with 1.35 million Monte Carlo simulation runs, while PrimeTime-SI over-estimates by 14.64%, 16.39% and 19.69%, SSTA achieves an average of 2.85%, 2.89% and 3.59% inaccuracy, and SPSTA achieves an average of 1.85%, 2.54% and 1.53% inaccuracy in crosstalk-oblivious, crosstalk-aware non-iterative and crosstalkaware iterative analysis, respectively