Abstract
We propose combining net-based and path-based timing-driven placement for better trade off between accuracy and complexity. We employ a force-directed iteration improvement approach and design a slack-based cost function that accurately captures timing violation and maneuverability. Our cost function can adjust weights on late, early, total, and worst slack for maneuverability. We evaluate our approach using the ICCAD2014 ITDP contest framework. It sets a displacement constraint on cell movement to preserve global placement result. Experiment results show that our approach significantly reduces timing violation from the initial legalize global placement. On average, total negative late slack, worst negative late slack, total negative early slack and worst negative early slack have been improved by 99.98%, 96.28%, 32.23% and 34.87%, respectively.