The upset-fault-observer: A concept for self-healing adaptive fault tolerance

Advancing integration reaching atomic-scales makes components highly defective and unstable during lifetime. This demands paradigm shifts in electronic systems design. FPGAs are particularly sensitive to cosmic and other kinds of radiations that produce single-event-upsets (SEU) in configuration and internal memories. Typical fault-tolerance (FT) techniques combine triple-modular-redundancy (TMR) schemes with run-time-reconfiguration (RTR). However, even the most successful approaches disregard the low suitability of fine-grain redundancy in nano-scale design, poor scalability and programmability of application specific architectures, small performance-consumption ratio of board-level designs, or scarce optimization capability of rigid redundancy structures. In that context, we introduce an innovative solution that exploits the flexibility, reusability, and scalability of a modular RTR SoC approach and reuse existing RTR IP-cores in order to assemble different TMR schemes during run-time. Thus, the system can adaptively trigger the adequate self-healing strategy according to execution environment metrics and user-defined goals. Specifically the paper presents: (a) the upset-fault-observer (UFO), an innovative run-time self-test and recovery strategy that delivers FT on request over several function cores but saves the redundancy scalability cost by running periodic reconfigurable TMR scan-cycles, (b) run-time reconfigurable TMR schemes and self-repair mechanisms, and (c) an adaptive software organization model to manage the proposed FT strategies.