SHF: Small: Uncertainty Modeling and Design Methods for Heterogeneous Embedded Systems

Marquette University, Milwaukee WI

Investigators

Abstract

Future embedded systems will contain tens and projected hundreds of heterogeneous cores. The increased complexity and heterogeneity, however, come with new design and optimization challenges including increased design uncertainties due to process, voltage, and temperature variations and poor reliability due to elevated rates of faults. Therefore, it is imperative to rethink existing embedded system design approaches in order to directly consider these issues during the design and optimization processes. To address these challenges, this research develops a new design method to address the increased uncertainties and to improve reliability and performance of future complex embedded systems. This new design method is developed with both general probabilistic and non-probabilistic uncertainty models. These uncertainty models are used to develop a multi-objective computer-aided design automation framework, which incorporates several algorithmic innovations based on Monte Carlo techniques and evolutionary algorithms. At the heart of the proposed design flow lies the hardware/software co-synthesis or mapping problem, which is solved with enhanced evolutionary algorithms. By bridging several areas of research including uncertainty modeling, hardware/software co-synthesis of embedded systems, robust multi-objective optimization, and design automation tools development, the proposed design method enables the development of better embedded systems, which have an increasingly dramatic impact on society via applications ranging from automotive and consumer electronics to military and space. The proposed research is also closely integrated with a broad and diverse education and outreach plan aimed at inspiring female students to pursue careers in science, technology, engineering, and mathematics. More broadly, the results of this project impact significantly the design of future integrated systems by building the foundation for subsequent studies aimed at establishing robust multi-objective optimization for next-generation embedded systems.