This dissertation presents comprehensive methodologies to improve the routability and design reliability of advanced integrated circuits under emerging technology nodes. As design rules become increasingly complex, challenges such as pin inaccessibili...
This dissertation presents comprehensive methodologies to improve the routability and design reliability of advanced integrated circuits under emerging technology nodes. As design rules become increasingly complex, challenges such as pin inaccessibility and routing congestion have led to a higher incidence of design rule violations (DRVs). To address these challenges, the first part of this work introduces a machine learning-based framework that predicts potential DRC hotspots by capturing the intricate relationship between pin accessibility and routing congestion. The proposed model integrates graph neural network and convolutional architectures through a newly defined pin proximity graph, effectively representing spatial and connectivity information within cells. Furthermore, by incorporating transfer learning and incremental learning strategies, the model is designed to maintain prediction accuracy across technology nodes while enabling efficient adaptation to new circuit designs.
In the second part, this dissertation presents an automatic synthesis framework for CFET (Complementary FET) standard cells to enhance pin accessibility and routing quality. The proposed framework systematically explores transistor placement and in-cell routing by making optimal use of the backside metal layers. Graph-based pruning
and satisfiability modulo theory (SMT) formulations are employed to achieve efficient and reliable metal connectivity with minimal design overhead. Experimental evaluations demonstrate that the CFET cells synthesized with this method achieve improved routing efficiency and reduced DRV occurrences at the chip level by utilizing less frontside metal compared to conventional CFET designs.
By combining predictive modeling and CFET-aware cell synthesis, this research provides a unified and sustainable approach to DRV reduction, routing optimization, and design scalability for future semiconductor technologies.