Sequential spectral ATPG using the wavelet transform and compaction

Abstract

We propose a simulation-based sequential automatic test pattern generation (ATPG) method for single stuck-at faults using the wavelet transform and linear reverse order restoration (LROR) compaction. Sequential ATPG is such a difficult problem that nearly all industrial circuits are tested using full-scan design and combinational ATPG. We describe the wavelet transform, a spectral noise substitution experiment showing the utility of the transform, the wavelet ATPG procedure, and its parameters. Our method outperforms all other methods, including the Hadamard transform method by Giani et al. (2001), PROPTEST by Guo et al. (2003), and the selfish gene (SG) method by Zhang et al. (2004). We present validated results. On the ISCAS '89 benchmarks, the fault coverage of the wavelet ATPG program is 2.226% greater than for the selfish gene ATPG method, and the vector length is 91.690% of the selfish gene vector length. On the 6 largest circuits, wavelet ATPG detected 2.534 % more faults than the selfish gene algorithm. On sl5850, wavelet ATPG detected 14.9% more faults.