Early Component-Based System Reliability Analysis for Approximate Computing Systems

A key enabler of real applications on approximate computing systems is the availability of instruments to analyze system reliability, early in the design cycle. Accurately measuring the impact on system reliability of any change in the technology, circuits, microarchitecture and software is most of the time a multi-team multi-objective problem and reliability must be traded off against other crucial design attributes (or objectives) such as power, performance and cost. Unfortunately, tools and models for cross-layer reliability analysis are still at their early stages compared to other very mature design tools and this represents a major issue for mainstream applications. This paper presents preliminary information on a cross-layer framework built on top of a Bayesian model designed to perform component-based reliability analysis of complex systems.