More Effective Power Network Prototyping by Analytical and Centroid Learning

• Main Authors: Chuang, Yu-Hsiang, 莊宇翔
• Other Authors: Chen, Hung-Ming
• Format: Others
• Language: en_US
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/7dxgfg

碩士 === 國立交通大學 === 電子研究所 === 106 === As the development of Moore's Law, many knotty problems come up and multiple design constraints need to be satisfied in IC design. The IR-Drop problem has been an important and serious issue for a long time. The voltage drop will reduce noise margin and increase gate delay; besides, the supply voltage of IC is much lower with the fast shrinking of CMOS process technology node. If the driving voltage of cell is not sufficient, it will deteriorate the signal integrity and cause functional failure. Therefore, a robust power distribution network that satisfies given constraints is more concerned than before. In this thesis, we present a more effective design flow to automatically generate a power distribution network(PDN) verified by state-of-the-art commercial tool without IR violation. We improve the previous works to acquire a more comprehensive design flow to ensure the quality of our PDN. Firstly, we propose an analytical model which contains overall power metal resource and consider the different types of macros to determine the total metal width of PDN. Moreover, the optimization is based on a centroid learning method from unsupervised learning to consolidate PDN so that we can obtain a more adequate topology of PDN. Our work has experimented on real designs in 65 nm LP process, 0.18 um generic process, and 40 nm LP process. The results show that our framework can satisfy the given IR-Drop and simultaneously save lots of metal resource.