A Practitioner's Guide to Software-Based Soft-Error Mitigation Using AN-Codes

Abstract

Arithmetic error coding schemes (AN codes) are a well known and effective technique for soft error mitigation. Although coding theory being a rich area of mathematics, their implementation seems to be fairly easy. However, compliance with the theory can be lost easily while moving towards an actual implementation - finally jeopardizing the aspired fault-tolerance characteristics. In this paper, we present our experiences and lessons learned from implementing AN codes in the Cored dependable voter. We focus on the challenges and pitfalls in the transition from maths to machine code for a binary computer from a systems perspective. Our results show, that practical misconceptions (such as the use of prime numbers) and architecture-dependent implementation glitches occur on every stage of this transition. We identify typical pitfalls and describe practical measures to find and resolve them. Our measures eliminate all remaining SDCs in the Cored voter, which is validated by an extensive fault-injection campaign that covers 100 percent of the fault space for 1-bit and 2-bit errors.