Substrate Signal Routing Solution Exploration for High-Density Packages with Machine Learning

• Published in: Proceedings of Technical Program of International Symposium on VLSI Design, Automation and Test pp. 1 - 4
• Main Authors: Yeh, Yeu-Haw, Chen, Simon Yi-Hung, Chen, Hung-Ming, Tu, Deng-Yao, Fang, Guan-Qi, Kuo, Yun-Chih, Chen, Po-Yang
• Format: Conference Proceeding
• Language: English
• Published: IEEE 18.04.2022
• Subjects: Design automation; Machine learning; Manuals; rip-up and reroute; Routing; substrate routing; Time to market; Very large scale integration; Wires
• ISSN: 2472-9124
• DOI: 10.1109/VLSI-DAT54769.2022.9768081

Off-chip substrate routing for high-density packages is on the critical path for time to market. There are several substrate routing algorithms have been proposed in previously. Although routers can rapidly that produce routing results, these results might not be satisfied universally from expert's experiences. In other words, different routers tend to have strength and weakness from different SOC designs. In this paper, we propose a novel reroute framework to remedy the defect of substrate routers by using supervised machine learning. We build a classification model which extracts features from expert's experience. It will identify suboptimal routings that do not conform to manual routing style. Then, reroute these areas using different routers and produce diverse results, then feed to classification model until they are acceptable. Guided by the model, suboptimal routing areas are replaced by results that are closer to expert's manual routing. Experiments show that our rerouting framework achieves 36.5% improvement on the number of wire bends and 1.6% wirelength improvement, compared with initial results routed by recent related work.