Cloud bursting an EDA workload with ML-driven technique for future SoC development

Abstract

SoC design traditionally relies on Electronic Design Automation (EDA) tools executed on locally deployed computing resources, typically known as on-premises machines. Its design efficacy such as placement optimization within a time limit, hence, heavily relies on the available on-premises computing resources. At IBM Research, we recently explored a new promising approach for SoC design by leveraging cloud computing. We utilize open-source RTL-to-GDS tools and machine learning-driven optimization techniques to demonstrate the potential of scalable cloud computing. Unlike a traditional design paradigm, computing resource is dynamically allocated on the cloud to handle variable requirements at each design stage. This not only reduces anticipated infrastructure operation costs but also lowers the entry barrier for new SoC designers to develop innovative designs and keep pace with the rapidly growing trends in various SoC manufacture.