DREAMPlace 3.0: Multi-Electrostatics Based Robust VLSI Placement with Region Constraints

Jiaqi Gu,Zixuan Jiang,Yibo Lin,David Z. Pan
DOI: https://doi.org/10.1145/3400302.3415691
2020-01-01
Abstract:Placement is a critical step for modern very-large-scale integrated (VLSI) design closure. Recently, electrostatics-based analytical placement frameworks (ePlace) demonstrate promising performance in both solution quality and runtime. However, existing ePlace-based placers fail to meet the versatility and robustness requirements on various placement workloads. We propose a versatile and robust placer to solve region-constrained placement problems with better solution quality and faster convergence. We formulate the region- constrained placement problem into a multi-electrostatics system via virtual blockage insertion and field isolation. To achieve robust wirelength minimization with aggressive density constraints, we adopt self-adaptive quadratic density penalty and entropy injection techniques to automatically accelerate and stabilize the nonlinear optimization. Our experiments on ISPD 2015 benchmarks with region constraints demonstrate an average of > 13% HPWL improvement and> 11 % top5 overflow improvement compared with advanced region-aware placers Eh?Placer and NTUplace4dr. Our robustness-boost techniques show an average of ~1% and ~10% improvement in HPWL and runtime compared to DREAMPlace on ICCAD 2014 and ISPD 2019 benchmark suites.
What problem does this paper attempt to address?