Robust correction of repeating errors by non-linear codes

For many devices and circuits, even a single fault can result in errors of a large Hamming weight at the output of a device, complicating the task of error detection and correction. Traditional error detection and correction techniques are based on linear error-detecting codes that are designed to detect or correct errors of a small multiplicity or burst errors. The authors present codes and architectures suitable for both online and offline error detection and correction which can detect and correct any error of any multiplicity by exploiting the laziness of error manifestation. Faults in many combinational circuits can lead to `lazy` or repeating errors at the outputs of a device to be protected. The proposed decoding algorithms can correct any error of any Hamming weight if the same error affects at least three different consecutive distorted output vectors of a protected device for binary algebraic errors. Likewise, the proposed decoding algorithms can correct any error of any Hamming weight if the same error affects two different output vectors of a protected device for arithmetic errors.