基于MIP的力导向大规模布局细化算法.pdf

1、ASPASP-DAC 2025DAC 202530thAsia and South Pacific Design Automation ConferenceZewen Li 1,Ke Tang 1,Lang Feng 2and Zhongfeng Wang 1,2An MIP-based Force-directed Large Scale Placement Refinement Algorithm1School of Electronic Science and Engineering,Nanjing University,China2School of Integrated Circui

2、ts,Sun Yat-sen University,China Background and Related Works Algorithms Experiments ConclusionOutline2ASP-DAC 2025,Tokyo,Japan01/22/2025 Background and Related Works Algorithms Experiments ConclusionOutline3ASP-DAC 2025,Tokyo,Japan01/22/2025Background4ASP-DAC 2025,Tokyo,Japan01/22/2025 Given an init

3、ial placement,move the standard cells again Legality:stay within bounds;no overlap;row/site alignment Objective:minimize total wirelengthBackground5ASP-DAC 2025,Tokyo,Japan01/22/2025C1C2C3C4C7C8C5C6C9C10C11C12C13C14C15 Mixed Integer Programming(MIP)Linear programming with integer variables Formulate

4、 the problem flexibly Precise refinement Challenges Too many integer variables slow down the speed How to model wirelength?HPWL?RSMT?Background6ASP-DAC 2025,Tokyo,Japan01/22/2025Related Work7ASP-DAC 2025,Tokyo,Japan01/22/2025C1C2 Every cell can move everywhere in the boundary.CRP:generate discrete c

5、andidate positions for each cellRelated Work8ASP-DAC 2025,Tokyo,Japan01/22/20251 Erfan Aghaeekiasaraee et al.2023.CRP2.0:A Fast and Robust Cooperation between Routing and Placement in Advanced Technology Nodes.ACM Transactions on Design Automation of Electronic Systems 28,5(2023),142.Too small searc

6、h spaceLimit the optimization levelC1C2C1Candidate 1Candidate 2Candidate 1C2Candidate 2 Displacement:generate a potential region for each cellRelated Work9ASP-DAC 2025,Tokyo,Japan01/22/20252 Ke Tanget al.2024.Mixed Integer Programming based Placement Refinement by RSMT Model with Movable Pins.ACM Tr