An AI-Enhanced STA Framework Supporting CCS and SI with Signoff Accuracy

Chao Yan
Hexin


Abstract

STA tools play a critical role in signoff, requiring exceptional accuracy to ensure circuit reliability. OurTool is an industry-grade STA solution designed to support advanced modeling with high precision. This work introduces the following key contributions: (1) a sophisticated stage delay calculation methodology that incorporates Composite Current Source (CCS) driver and receiver models with distributed RC networks, achieving superior timing accuracy; (2) a hybrid framework that combines traditional SI flow with AI predictors, utilizing internal tool data for both training and inference, which significantly enhances accuracy and performance in delta delay estimation and (3) a stage-based testing and comparison framework for comprehensive validation. OurTool error is within 3% of SPICE results, while maintaining performance comparable to industry-leading tools.