GA Based On-line Testing and Recovery for Critical Digital Systems

This work proposes a Genetic Algorithm (GA) based on-line testing and recovery mechanism for reconfigurable critical digital circuits. Given the input / output specifications for a circuit, multiple versions (n) of the circuit are evolved using GA techniques. The n different versions are evolved to provide 100% fault coverage. A TMR fault tolerant circuit is designed using three different versions out of these ‘n’ versions and a voting mechanism that chooses the majority output. Continuous on-line testing/fault detection is achieved by systematically changing the version used in the three modules. This is done so as to exercise all the components present in the modules. As one fault-free circuit provides the output, the external software control downloads the new configuration of a different version on to one of the other two circuits for testing. This process is cyclically repeated between the three circuits, with the various versions being periodically downloaded and tested, thus covering all the components. During the period of download, only one of the circuits is out of action and so the output of the circuit is not affected. Once a fault is detected, the faulty version is replaced with a fault-free version and the process continues. The efficacy of the proposed technique has been studied using simulated faults and hardware implementation of a 2*2 multiplier as a proof of concept. It is found that the proposed technique provides on-line self-testing and recovery of 100% single component failures.