I-Corps: Enabling Electronic Design using Data Intelligence

Texas A&M Engineering Experiment Station, College Station TX

Investigators

Abstract

The broader impact/commercial potential of this I-Corps project stems from its data intelligence approach to empower electronic design automation. The semiconductor industry provides vital hardware backbone of the information technology age through an extremely wide range of integrated circuits (ICs) in computing devices and consumer electronics. Modern IC development process is bottlenecked by growing chip design complexity, e.g. measured by large device count and functionality diversity, and ever-demanding requirements on computing performance and power/energy efficiency. Advanced IC manufacturing processes are costly, and yet have unavoidable process variations, making fabricated chips susceptible to failures. With its revenue reaching $7.8 billion in 2015, the electronic design automation (EDA) industry supplies indispensable tools and methodologies that make IC design possible. The potential market and societal impact of the proposed EDA innovation is substantial. This technology can help semiconductor and chip design companies develop integrated circuits of improved performance and robustness with a reduced time-to-market and development cost. This I-Corps project demonstrates novel machine learning algorithms targeting electronic design automation. As the complexity of integrated circuits scales up rapidly, the need for smart design tools is prominent. The EDA industry is in the early phase of rapid integration of machine learning algorithms into commercial IC design flows. The learning methods focused in this project significantly improve the accuracy of statistical regression and classification over the current-state-of-the-art, and offer the much needed understanding of the underlying structure of the data. Built upon the focused machine learning algorithms, the targeted EDA technology can efficiently process simulation or measured performance data of existing chip designs, and intelligently learn the complex hidden relationships between performance specifications, design parameters, and manufacturing conditions. As a result, it offers a powerful data science solution to IC design optimization, verification, and debug. Implemented as high-performance parallel software design tools, the technology will bring the power of machine learning to the field of electronic design.