Permanent fault detection and diagnosis in the lightweight dual modular redundancy architecture

The Lightweight Dual Modular Redundancy (LDMR) is a fault tolerant architecture for low-latency soft error correction. The LDMR introduces a software compilation strategy that enforces error containment inside a basic block, allowing for a simplified error correction policy. This paper evaluates how the LDMR and its architectural components behave in the presence of permanent faults. It also classifies how sensitive the error detection and rollback machinery is to hard faults. By including permanent fault detection and diagnosis, the LDMR becomes a comprehensive fault tolerant architecture for embedded computing, covering a broad range of fault models. This paper also evaluates the LDMR´s performance overhead using a MiBench subset, which is currently 1.54 in average.