Optimization of analog fault coverage by exploiting defect-specific masking

A new method is presented to detect catastrophic defects from the signal analysis of dynamic current consumption waveforms of analog circuits. While other techniques use the whole information in a Root-Mean-Square computation or in black-box techniques such as a neural network, the central point of this work resides in the selection of waveform samples to create a signature able to discriminate a defective circuit from a fault-free circuit. The selection of samples is implemented by the introduction of binary vectors to partially mask the data. Confronted with process variations, this technique offers the advantage of being straightforward and simple to implement in Automated Test Equipments. The generation of the masks is optimized to improve the defect coverage by means of a genetic algorithm maximizing the distance between the signature of the fault-free circuit and a faulty circuit. Results from simulations on industrial circuits show that the number of detected defects can be nearly doubled for specific stimuli.