The Research On N-detection Tests Generation

With the technology of Very Large Scale Integrated circuits technology hasentered into the age of nanometer, its function becomes more complicated and theintegrated density increase quickly. As a result, the complexity of circuits’ faultssharply increased. Therefore, it is not realistic to establish fault model and generatetest sets for all faults. One efficient solution to this problem is the N-detection testthat uses conventional stuck-at fault model to generate patterns that attempt to detecteach fault at least N times. This test method is based on the theory that the times of astuck at fault to be detected more, the probability of activation the non-model fault ofthis node is greater. Previous experimental results have shown that N-detection testsets have high fault coverage for stuck open fault, bridging faults and other faultmodels. However, it is hard to generate test sets for N-detection and which needs hugetest data volume.The lower bond on the size of the single detection test is the size of theindependent fault set (IFS). People want to obtain minimal N-detection test sets as thesmallest single-detection test generation studies have been very mature. In theory, theminimal of the N-detection test set is N times of IFS. But the N-detection test setsobtained by the existing ATPG tool are much larger than the theoretical valueattributing to hard fault.An efficient method to generate N-detection tests is proposed in this paper. First,traditional single-test ATPG tool is used to generate a non-optimal M-detection testsets. Then the compression problem is transformed into an integer linear programming(ILP) model. So the minimum N-detection test sets could be obtained by solving theILP.The detection times of hard fault by M-detection test sets which generated bysingle detect ATPG tool are larger than N, but smaller than easy fault, which affectsthe compression efficiency. So this paper proposes an extra vector which containsdon’t care bit for all hard fault by atalanta algorithm. These vectors are compressedand then the don’t care bit is filled randomly. The number of fault’s detection times issignificantly increased by this mothod. The compression ratio is also increased by thisimprovement.It is time consuming to find optimal solutions by ILP. Therefore, a row dominance relationship in a matrix is defined in this paper. The dimension ofconstraint matrix is reduced by removing the dominant row, which improves timeefficiency of the N-detection test generation method and adaptability for largecircuits.