WISP: Image Segmentation-Based Whitespace Diagnosis for Optimal Rectilinear Floorplanning

Abstract

The increasing number of rectilinear floorplans in modern chip designs presents significant challenges for traditional macro placers due to the additional complexity introduced by blocked corners. Particularly, the widely adopted wirelength model Half-Perimeter Wirelength (HPWL) struggles to accurately handle rectilinear boundaries, highlighting the need for additional objectives tailored to rectilinear floorplan optimization. In this paper, we identify the necessity for whitespace diagnosis in rectilinear floorplanning, an aspect often overlooked in past research. We introduce WISP, a novel framework that analyzes and scores whitespace regions to guide placement optimization. WISP leverages image segmentation techniques for whitespace parsing, a lightweight probabilistic model to score whitespace regions based on macro distribution, a Gaussian Mixture Model (GMM) for whitespace density scoring and direction-aware macro relocation to iteratively refine macro placement, reduce wasted whitespace, and enhance design quality. The proposed diagnostic technique also enables the reclamation of block-level unused area and its return to the top level, maximizing overall area utilization. When compared against state-of-the-art academia placer DREAMPlace 4.1, our method achieves an average improvement of 5.4% in routing wirelength, with a maximum of 11.4% across widely-used benchmarks. This yields an average of 41.5% and 43.7% improvement in Worst Negative Slack (WNS) and Total Negative Slack (TNS), respectively. Additionally, WISP recycles an average of 16.2% area at the block level, contributing to more efficient top-level area distribution.