SEDSR: Soft Error Detection Using Software Redundancy

Seyyed Amir Asghari,Saadat Pourmozaffari,Hassan Taheri,Hossein Pedram,Atena Abdi

doi:10.4236/jsea.2012.59078

Seyyed Amir Asghari, Saadat Pourmozaffari + Show 3 more

Open Access

https://doi.org/10.4236/jsea.2012.59078

Copy DOI

Abstract

This paper presents a new method for soft error detection using software redundancy (SEDSR) that is able to detect transient faults. Soft errors damage the control flow and data of programs and designers usually use hardware-based solutions to handle them. Software-based techniques for soft error detection force less cost and delay to systems and do not change their configuration. Therefore, these kinds of methods are appropriate alternatives for hardware-based techniques. SEDSR has two separate parts for data and control flow errors detection. Fault injection method is used to compare SEDSR with previous methods of this field based on the new parameter of “Evaluation Factor” that takes in account fault coverage, memory and performance overheads. These parameters are important in real time safety critical applications. Experimental results on SPEC2000 and some traditional benchmarks of this field show that SEDSR is much better than previous methods of this field. SEDSR’s evaluation factor is about 50% better than other methods of this field. These results show its success in satisfaction of the existing tradeoff between fault coverage, performance and memory overheads.

Highlights

It is proved that decreasing in feature size of integrated circuits and the increased complexity of computer architectures will lead to less reliable systems
Fault injection method is used to compare soft error detection using software redundancy (SEDSR) with previous methods of this field based on the new parameter of “Evaluation Factor” that takes in account fault coverage, memory and performance overheads
It is proved that about 33% to 77% of transient faults are converted to Control Flow Errors (CFE) and the remained percentage is converted to data errors [1,2]

Summary

Introduction

It is proved that decreasing in feature size of integrated circuits and the increased complexity of computer architectures will lead to less reliable systems. Recent researches show that soft errors damage control flow or data of a program. It is proved that about 33% to 77% of transient faults are converted to Control Flow Errors (CFE) and the remained percentage is converted to data errors [1,2]. The methods based on hardware redundancy have better fault coverage but impose more cost and overhead to the system and may not satisfy users in general purpose applications because of system reconfiguration. Software-based methods have less fault coverage and more delay; they have less cost and do not change the configuration of the system and are utilized in different applications due to their flexibility [3,4,5,6,7,8,9,10,11]

Methods

Results

Conclusion