Soft error reliability predictor based on a Deep Feedforward Neural Network
- 1 March 2020
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-5
Abstract
Statistical fault injection is a widely used methodology to early evaluation of soft error reliability of microprocessor based systems. Due to the increasing complexity of the software and hardware stack, the simulation of faults on modern processors is becoming a computationally demanding task even for ISA-equivalent models and virtualization tools. This paper proposes and explores the use of a supervised machine learning technique, Deep Feedforward Neural Network, to design a predictor which drastically reduces the computing time of fault injection campaigns. In addition, a novel approach is presented to increase the training data from a limited set of benchmarks. Thanks to this approach, the predictor can be modeled with an extensive data set comprising not only of millions of fault injections but also thousands of different benchmarks. Experiments show promising results for estimating the applications fault tolerance when they run on state of the art ARM processor.Keywords
This publication has 7 references indexed in Scilit:
- A Compact Model to Evaluate the Effects of High Level C++ Code Hardening in Radiation EnvironmentsElectronics, 2019
- Nonintrusive Automatic Compiler-Guided Reliability Improvement of Embedded Applications Under Proton IrradiationIEEE Transactions on Nuclear Science, 2019
- Using Machine Learning Techniques to Evaluate Multicore Soft Error ReliabilityIEEE Transactions on Circuits and Systems I: Regular Papers, 2019
- Finding Resilience-Friendly Compiler Optimizations Using Meta-Heuristic Search Techniques2016
- Fault Modeling of Extreme Scale Applications Using Machine Learning2016
- Understanding the propagation of transient errors in HPC applications2015
- Application-Based Analysis of Register File Criticality for Reliability Assessment in Embedded MicroprocessorsJournal of Electronic Testing, 2015