| Abstract: | The detection of and recovery from transient faults that cause errors in microprogrammed control units are addressed. Error detection is based upon a particular form of signature analysis, namely the run-time computation of a value (the signature) that depends strictly on the microprogram path (the sequence) under execution. The microprogram is statistically segmented (sequences are defined in a fixed way, at design time), and an anticipated signature is associated with each sequence. At run time, at the start of each sequence the computation of the signature is restarted, whereas at the end the computed signature is compared with the anticipated signature of the sequence. If the check is passed, the status of the control unit is saved in suitable registers, and the previously saved status is discarded. If the check fails, the last-saved status is restored (rolled back) and microprogram execution restarts from this last-saved point |
