E-RESO: An Enhanced Time Redundancy-based Error Detection Approach for Arithmetic Operations

In this paper, two time redundancy-based error detection methods are proposed for critical real-time embedded systems. These approaches are considered for error detection in computation and mainly focused on arithmetic operations especially the addition due to its frequent and critical usage. The proposed methods detect the occurred faults by duplicating the computation and modify operands of two versions by encoding schemes to avoid error propagation in the system. Here, a tradeoff is formed between the complexity of the encoding mechanism that is applied to the computation operands and the detection coverage of the proposed method. More complicated encodings avoid the propagation and masking of faults but in terms of more overhead. our proposed methods try to consider this trade-off and improve the detection capability of error detection methods with low overhead. To this aim, we have proposed two detection approaches based on recomputing and encoding the operands and utilized basic operations such as shift and swap in the encoding. The effectiveness of our proposed method is evaluated by several experiments and compared to basic and accepted methods of this field. Experimental results show that our proposed methods are capable of detecting more than 99% of injected stuck-at faults in one and two bits of the data with a very low-performance overhead, averagely.