Improved backwards analysis for architectural vulnerability factor estimation

The effects of single event upsets (SEU) are becoming increasingly important to circuit designers. Exact and detailed error rate estimations are needed to determine a system´s level of reliability. The architectural vulnerability factor (AVF) is a measure for the relative reliability of a circuit. In this paper we outline the properties of several known approaches such as statistical testing (fault injection), probabilistic error propagation and formal techniques. In addition we present necessary improvements to the Backwards Analysis method for AVF estimation. These improvements allow the analysis of any kind of RT-level circuit. Finally we present optimizations of the algorithm, which result in linear runtime overhead for the analysis of circuits compared to their simulation time.