Author: Keren Zhu (Ph. D.)

Publisher:

Published: 2022

Total Pages: 0

ISBN-13:

DOWNLOAD EBOOK

The performance of analog circuits is critically dependent on layout parasitics, but the layout has traditionally been a manual and time-consuming task. Analog and mixed-signal (AMS) circuits often impose specific parasitics and mismatch requirements on their layout implementation. Designers leverage their prior experience to place devices in specific patterns and configurations to reduce parasitics, the effects of local variation gradients, and layout-dependent effects. The reason behind this is from both the algorithm and software. Automated AMS layout synthesis faces challenges in developing effective place-and-route (PNR) algorithms for high-performance AMS circuits and lacks easily usable and accessible software. This dissertation covers several analog PNR algorithms to improve the quality of automated layout synthesis and the circuit learning methodology targeting further reducing human efforts. The proposed techniques have become critical parts of the open-source AMS layout synthesis software MAGICAL. This dissertation first proposes a novel analog routing methodology. The proposed framework, GeniusRoute, leverages machine learning to provide routing guidance, mimicking the sophisticated manual layout approaches. This approach allows the automatic analog router to follow the design expertise of human engineers while no additional manual effort is required to code the layout strategies. The proposed methodology obtains significant improvements over existing techniques and achieves competitive performance to manual layouts while capable of generalizing to circuits of different functionality. This dissertation also proposes a practical mixed-signal placement framework. Unlike the existing techniques, which mainly focus on geometric constraints in analog building blocks, the proposed framework formulates and effectively optimizes the system-level signal flow for sensitive mixed-signal circuits. Leveraging prior knowledge from schematics, we propose considering the critical signal paths in automatic AMS placement and presenting an efficient framework. The proposed framework shows efficiency and effectiveness with a reduced routed wirelength compared to a state-of-the-art AMS placer and improved post-layout performance. Furthermore, the well generation in the analog layout synthesis flow is revisited. Instead of treating well generation as an isolated process, we propose a new methodology of well-aware placement. We formulate the well-aware placement problem and propose a machine learning-guided placement framework. By allowing well sharing between transistors and explicitly considering wells in placement, the proposed framework achieves more than 74% improvement in the area and more than 26% reduction in half-perimeter wirelength over existing placement methodologies in experimental results. Finally, this dissertation revisits and explores the fundamental problem of analog circuit learning. A novel unsupervised circuit learning framework is proposed to leverage the human layout as a training label. The machine learning model is pre-trained with automatically extracted labels and then transferred to other downstream tasks. The transferrable circuit representation model demonstrates the possibility of a machine learning model to understand the circuits

Download or read book Fully-automated Layout Synthesis for Analog and Mixed-signal Integrated Circuits written by Keren Zhu (Ph. D.) and published by . This book was released on 2022 with total page 0 pages. Available in PDF, EPUB and Kindle. Book excerpt: The performance of analog circuits is critically dependent on layout parasitics, but the layout has traditionally been a manual and time-consuming task. Analog and mixed-signal (AMS) circuits often impose specific parasitics and mismatch requirements on their layout implementation. Designers leverage their prior experience to place devices in specific patterns and configurations to reduce parasitics, the effects of local variation gradients, and layout-dependent effects. The reason behind this is from both the algorithm and software. Automated AMS layout synthesis faces challenges in developing effective place-and-route (PNR) algorithms for high-performance AMS circuits and lacks easily usable and accessible software. This dissertation covers several analog PNR algorithms to improve the quality of automated layout synthesis and the circuit learning methodology targeting further reducing human efforts. The proposed techniques have become critical parts of the open-source AMS layout synthesis software MAGICAL. This dissertation first proposes a novel analog routing methodology. The proposed framework, GeniusRoute, leverages machine learning to provide routing guidance, mimicking the sophisticated manual layout approaches. This approach allows the automatic analog router to follow the design expertise of human engineers while no additional manual effort is required to code the layout strategies. The proposed methodology obtains significant improvements over existing techniques and achieves competitive performance to manual layouts while capable of generalizing to circuits of different functionality. This dissertation also proposes a practical mixed-signal placement framework. Unlike the existing techniques, which mainly focus on geometric constraints in analog building blocks, the proposed framework formulates and effectively optimizes the system-level signal flow for sensitive mixed-signal circuits. Leveraging prior knowledge from schematics, we propose considering the critical signal paths in automatic AMS placement and presenting an efficient framework. The proposed framework shows efficiency and effectiveness with a reduced routed wirelength compared to a state-of-the-art AMS placer and improved post-layout performance. Furthermore, the well generation in the analog layout synthesis flow is revisited. Instead of treating well generation as an isolated process, we propose a new methodology of well-aware placement. We formulate the well-aware placement problem and propose a machine learning-guided placement framework. By allowing well sharing between transistors and explicitly considering wells in placement, the proposed framework achieves more than 74% improvement in the area and more than 26% reduction in half-perimeter wirelength over existing placement methodologies in experimental results. Finally, this dissertation revisits and explores the fundamental problem of analog circuit learning. A novel unsupervised circuit learning framework is proposed to leverage the human layout as a training label. The machine learning model is pre-trained with automatically extracted labels and then transferred to other downstream tasks. The transferrable circuit representation model demonstrates the possibility of a machine learning model to understand the circuits