Machine Learning for Analog/Mixed-Signal IC Design: Scaling - From Circuits to Systems - Dissertation

Aug 3, 2026

Upload by Anda

Analog and mixed-signal integrated circuits power many emerging applications, yet rising demand requires shorter design cycles and faster time-to-market. Traditional analog design relies heavily on manual expert iteration between circuit sizing and layout, while existing automation methods often fail to scale from small components to system-level designs and rarely incorporate post-layout parasitic effects during sizing. This dissertation applies machine learning to practical, scalable analog/mixed-signal design automation by learning layout symmetry constraints, predicting layout quality and parasitics without extraction or simulation, and enabling efficient sizing with Bayesian optimization and in-the-loop layout generation. Circuit simulations and real chip tape-out measurements validate results for ADC-scale system design.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages239

Reading Time10 h 2 min

AccessFree to download

Acknowledgments

Abstract

List of Tables

List of Figures

Chapter 1

Copyright by Mingjie Liu 2022 1 The Dissertation Committee for Mingjie Liu certifies that this is the approved version of the following dissertation: Machine Learning for Analog/Mixed-Signal IC Design: Scaling From Circuits to Systems Committee: David Z. Pan, Supervisor Nan Sun | Jaydeep P. Kulkarni | | --- | | Yaoyao Jia | Haoxing Ren Machine Learning for Analog/Mixed-Signal IC Design: Scaling From Circuits to Systems by Mingjie Liu DISSERTATION Presented to the Faculty of the Graduate School of The University of Texas at Austin in Partial Fulfillment of the Requirements for the Degree of DOCTOR OF PHILOSOPHY THE UNIVERSITY OF TEXAS AT AUSTIN December 2022 Acknowledgments I want to express my sincerest gratitude to my advisor, Dr. David Z. Pan, who I met at the start of my academic journey, for his invaluable guidance and support [throughout my Ph.D. study. Professor Pan played a decisive role in leading](throughout my Ph.D. study. Professor Pan played a decisive role in leading)[ ](throughout my Ph.D. study. Professor Pan played a decisive role in leading)me to solve critical and challenging research problems independently in the field of electronic design automation. He also provided generous suggestions in terms of technical writing and presentations skills that benefited me to a great extent. Forme, Professor Pan is not only a wise, patient, encouraging, and inspiring research advisor, but also a kind friend in the daily life who has given me insightful advice. I am very fortunate to have worked with him and learned much from him. I would also like to extend my deepest gratitude to other committee members, Prof. Nan Sun, Prof. Jaydeep Kulkarni, Prof. Yaoyao Jia, and Dr. Haoxing Ren, for their precious efforts and contributions to this dissertation. I very much appreciate Prof. Nan Sun for the great number of helpful discussions and collaborationson various research projects, where he offered practical advice with his extensive knowledge of analog/mixed-signal circuit design. I want to thank Prof. Jaydeep Kulkarni for his valuable comments and generous support during the development of this dissertation. I am very grateful to Prof. Yaoyao Jia for her constructive technical suggestions which contribute to the completeness of this dissertation. In particular, I want to express my gratitude to Dr. Haoxing Ren for the significant number of helpful discussions and collaborations on various research projects, from whom I have received generous support and insightful guidance during my internships at Nvidia Corporation. It has been a great honor to have worked with my fellow MAGICAL team members, Dr. Yibo Lin, Dr. Biying Xu, Dr. Shaolan Li, Dr. Xiyuan Tang, Dr. Keren Zhu, and Hao Chen, for numerous technical discussions, suggestions, and collaborations. It has been a truly amazing journey to have worked with a team with such diverse background, from hardcore design automation algorithm gurus to expert circuit designers. I want to thank the UTDA lab members: Dr. Meng Li, Dr. Wuxi Li, Dr. Mohamed Baker Alawieh, Dr. Wei Shi, Ahmet F. Budak, Rachel S. Rajarathnam, Chenghao Feng, Zixuan Jiang, Jiaqi Gu, Hanqing Zhu, Xuyang Jin, and Hyunsu Chae. I would like the thank research associates from Peking University: Xiaohan Gao, and Zizheng Guo. I am also very grateful to other industrial mentors and collaborators: Dr. Brucek Khailany, Dr. Walker Turner, Dr. George Kokai, and Dr. Haoyu Yang. It has been my great pleasure working with all of them. Finally, I cannot begin to express my thanks to my dear family. Many thanks to all of my family members who offer support and encouragement. With much love, I thank my wife and my parents for seeing me through all the hardship and sharing every achievement along my Ph.D. study journey. Machine Learning for Analog/Mixed-Signal IC Design: Scaling From Circuits to Systems Publication No. Mingjie Liu, Ph.D. The University of Texas at A

FAQs

为什么传统模拟/混合信号IC设计难以满足快速交付需求？

传统流程依赖人工专家在电路尺寸与版图实现之间反复迭代，耗时且难以缩短设计周期。同时，已有自动化工作往往难以从小规模电路扩展到更大的系统级设计。

本论文解决了哪些从小电路到系统级扩展的关键限制？

论文指出：现有方法难以扩展到系统级，且很少在电路尺寸阶段考虑后版图寄生效应，从而限制了面向真实芯片流片的应用。

论文如何利用机器学习实现可扩展的自动化设计？

论文提出多种机器学习方法：从未标注网络表自动分配版图对称约束；用模型量化版图质量并估计寄生效应（无需寄生提取与电路仿真）；并结合贝叶斯优化与在环版图生成，以保证后版图性能并支持ADC等系统级设计。

论文如何验证所提出方法的有效性？

通过电路仿真结果与真实芯片流片测量相结合，证明将机器学习用于自动化模拟电路设计的有效性，并展示其对ADC等模拟系统设计的可扩展性。

Aug 3, 2026

Upload by Anda