In this paper, the natures of random and pseudo-random input sequences and their influence on permanent and intermittent fault detecting are analyzed. The aliasing fault coverage between the pseudo-random and random sequences is estimated. The activity probability features of the intermittent faults are considered. The selftest circuits of the intermittent faults are illustrated. The experimental results based on real circuits are obtained through simulation. The mathematical analysis and experimental results show that the quality of the pseudorandom testing is better than that of the random testing for the permanent and intermittent faults. The Markov chain models are used in obtaining the input sequence length needed for determining if a circuit fault is intermittent or permanent.
I IntroductionThe continually growing complexity of VLSI circuits makes the built-in self-test (BIST) techniques especially attractive [1~3]. In the techniques, two modules: the pseudorandom sequence generator and the output responses compactor, have to be integrated into a VLSI chip. The pseudo-random sequences generated by the generator is applied to the inputs of the circuit under test (CUT) and at the same time, the compactor compresses the CUT responses set into a signature.In most BIST techniques, the linear feedback shift register (LFSR) [4,5] is commonly used in generating the pseudorandom sequences. The responses compaction can be implemented by using the multiple input shift register (MISR) [6], the multiplexed parity trees [7], the transition count [8], the state-difference count [9], and so on. When the pseudo-random input sequences are applied to the CUT, the compactor compresses the CUT output responses simultaneously. After finishing the testing, the compressed signature will be compared with the expectant reference value produced by a corresponding fault-free circuit. If the two values are the same, the CUT testing will be considered pass; otherwise, the CUT will be considered faulty.However, the pseudo-random sequences generated by the LFSR used in the BIST techniques are not completely random. Therefore, the natures of the pseudo-random input sequences and the effectiveness of the compact approaches need to be analyzed.Some efforts have analyzed the pseudo-random testing [10][11][12][13]. These efforts used the combinatorial analysis and the differential solution to obtain the detection probability, the input sequence length, the fault coverage, the test confidence, and so on. The results of these efforts showed that the random test model is not a better approximation to the pseudo-random testing. However, the relationship among the input sequence length, the testability, and the fault coverage was not described in these efforts. Especially, the aliasing fault coverage between the random and pseudo-random sequences was not discussed.In this paper, we will analyze the relationship among the input sequence length, the testability, and the fault coverage for the random and pseudo-random testing, and estimate ...