An Examination and Survey of Random Bit Flips and Scientific Computing

Abstract

Data error, modification, and loss can occur in scientific computing due to numerous potential causes, including those natural, accidental, or malicious in nature. Such errors and modifications can silently impact scientific computing results if not detected and corrected or compensated for through other means. The goal of this document is to describe the ways in which integrity faults due to bit flips occur, their potential odds of occurring, and potential mitigation mechanisms in high-level, practical terms, and is broken into individual computer components and environments describing each of those elements. We conclude the report by summarizing key issues and several best practices for mitigation of the effects of bit flip-induced errors.