Application-driven co-design of fault-tolerant industrial systems

Abstract

This paper presents a novel methodology for the HW/SW co-design of fault tolerant embedded systems that pursues the mitigation of radiation-induced upset events (which are a class of Single Event Effects - SEEs) on critical industrial applications. The proposal combines the flexibility and low cost of Software Implemented Hardware Fault Tolerance (SIHFT) techniques with the high reliability of selective hardware replication. The co-design flow is supported by a hardening platform that comprises an automatic software hardening environment and a hardware tool able to emulate Single Event Upsets (SEUs). As a case study, we selected a soft-micro (PicoBlaze) widely used in FPGA-based industrial systems, and a fault tolerant version of the matrix multiplication algorithm was developed. Using the proposed methodology, the design was guided by the requirements of the application, leading us to explore several trade-offs among reliability, performance and cost.