REDUNDANCY REMOVAL FOR SEQUENTIAL-CIRCUITS WITHOUT RESET STATES

In this paper, we propose methods of identifying and removing redundancy in synchronous sequential circuits that do not have a global reset state. All existing structure-level test generators use three-valued logic, which is not completely accurate, to process circuits that have an unknown initial state. A fault that is reported undetectable by such test generators is not necessarily redundant. We show that if a fault is potentially undetectable (p-undetectable), it is redundant. An algorithm that identifies p-undetectable faults is then described. For large circuits, we propose a practical procedure for removing redundancy in the feedback-free portion of the circuits. An alternative approach to identifying redundancy that does not require determining the potential detectability of faults is also presented. This paper contains derivations of conditions in which undetectable faults are redundant. The undetectable faults are classified into three categories: (1) unactivatable faults, (2) unpropagable faults, and (3) faults that are both activatable and propagable, but cannot be simultaneously activated and propagated by any vector sequence. It is shown that class (1) and (2) faults are redundant faults, while class (3) faults may not be. Algorithms for identifying unactivatable and unpropagable faults are also described. These algorithms are implemented and incorporated in a redundancy removal system MIRACLE. Experimental results of running MIRACLE-C (removing combinational redundancy), MIRACLE-FFS (removing feedback-free sequential redundancy) and MIRACLE-S (removing unactivatable and unpropagable sequential redundancy) on large MCNC sequential benchmark circuits are presented.