Linear-feedback shift registers (LFSRs) are often used to compact test responses. Prior analyses, based on statistically-independent error models, have predicted that aliasing probability 'converges' to 2/sup -k/ for LFSR polynomials of degree k, and that primitive polynomials perform better than nonprimitive polynomials. This paper presents the first statistical results based on full fault simulation that confirm these predictions. However, the average aliasing probability is not by itself a useful measure of the loss of fault detection information; the authors introduce an upper confidence limit (UCL) for the loss of fault coverage. The 'ideal' UCL is shown to match closely the empirically-derived UCL.
展开▼