An intelligence driven test system for detection of stuck-open faults in CMOS sequential circuits

Abstract

This paper discusses an intelligence driven test system for generation of test sequences for stuck-open faults in CMOS VLSI sequential circuits. The networks in system evaluation are compiled from an RTL representation of the digital system. To excite a stuck-open fault it is only necessary that the output of the gate containing the fault take on opposite values during two successive clock periods. Excitation of the fault must therefore constrain two successive input/present-state vectors, referred to in the paper as the pregoal and goal nodes respectively. An initialization procedure is used to determine the pregoal state. Two theorems are proved establishing a 1-1 correspondence between stuck-at and stuck-open faults. As a result the D-algorithm may be used to determine the goal node. Determining the nodes was tried on many circuits and a high success rate was achieved. The pregoal is observed to have more "don't care" values. The next step is a "sensitization search" for an input sequence (X(s)) that drives the memory elements to the determined pregoal and goal states over two consecutive clock periods. It is easier for the search to reach the pregoal due to the greater number of "don't cares." Following a "propagation search" for an input sequence (X(p)) to drive the effect of the fault to an external output, the sequence of vectors (X(s)), (X(p)) will be passed to an "ALL-Fault Simulator" for verification. The simulation will be clock mode but will represent the output retention resulting from the stuck-open faults. One measure of the value of a special search procedure for stuck-open faults can be obtained by comparing the results employing this search with results obtained by searching only for the analogous stuck-at faults. A first order prediction would be a likelihood less than 0.5 that the predecessor of a stuck-at goal node would excite an opposite output in the gate containing the fault. A comparison of the two methods using the stuck-open "All-Fault Simulator" is presented.