“…In the active case, explicit and dynamic redundancy is used to detect and manage the fault effects by mechanisms like detection, localization, adaptation, retraining, or self-healing, which may lead to complexity [46]. For example, in the case of the DNN accelerator, this approach can be used for fault recovery by remapping, error detection, or reconfiguration [47]. Overall, not all failure scenarios can be considered at design time to provide passive or active arrangements.…”