| Zeit: | 14. Mai 2019 |
|---|---|
| Download als iCal: |
|
14:00-15:30, ITI-Seminarraum 3.175,
Department of Computer Science Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU)
The advent of new technology trends like the Internet of things and autonomous driving have made embedded systems more pervasive in the everyday life and has brought them to applications with strict non-functional requirements such as performance and power consumption. The continuous shrinkage in semiconductor devices has made the electronic components used in these systems increasingly susceptible to failure and degradation mechanisms like negative-bias temperature instability and gate oxide breakdown. This renders reliability a major concern in the design of embedded systems, and necessitates the automatic analysis and optimization of system reliability alongside other design objectives. To this end, this dissertation is based upon a system-level design methodology to overcome challenges and leverage opportunities in the design of reliable, yet efficient system implementations.
Due to variations in manufacturing, environment, and usage conditions, the reliability of modern electronic components is associated with various forms of uncertainty. The uncertainty in component reliability can propagate through the system, causing system reliability to be uncertain as well. Moreover, destructive effects such as high temperature have simultaneous impacts on several adjacent components, giving rise to correlation in their uncertainties. To consider uncertain characteristics and their correlations in system reliability analysis, this dissertation proposes to incorporate existing techniques, such as binary decision diagrams, and a Monte Carlo simulation into an uncertainty-aware reliability analysis. It models the reliability of each component using a reliability function with parameters characterized by probability distributions, and derives the probability distribution of system reliability. This necessitates the design space exploration to allow for the comparison of candidate system implementations with design objectives represented by probability distributions instead of single values. To this end, this dissertation introduces novel statistical and probabilistic comparison operators that enable to differentiate the quality of system implementations in the presence of uncertainty at low execution time overhead.