Scalable diversified antirandom test pattern generation with improved fault coverage for black-box circuit testing

Abstract

Pseudorandom testing is incapable of utilizing the success rate of preceding test patterns while generating subsequent test patterns. Many redundant test patterns have been generated that increase the test length without any significant increase in the fault coverage. An extension to pseudorandom testing is Antirandom that induces divergent patterns by maximizing the Total Hamming Distance (THD) and Total Cartesian Distance (TCD) of every subsequent test pattern. However, the Antirandom test sequence generation algorithm is prone to unsystematic selection when more than one patterns possess maximum THD and TCD. As a result, diversity among test sequences is compromised, lowering the fault coverage. Therefore, this thesis analyses the effect of Hamming distance in vertical as well as horizontal dimension to enhance diversity among test patterns. First contribution of this thesis is the proposal of a Diverse Antirandom (DAR) test pattern generation algorithm. DAR employs Horizontal Total Hamming Distance (HTHD) along with THD and TCD for diversity enhancement among test patterns as maximum distance test pattern generation. The HTHD and TCD are used as distance metrics that increase computational complexity in divergent test sequence generation. Therefore, the second contribution of this thesis is the proposal of tree traversal search method to maximize diversity among test patterns. The proposed method uses bits mutation of a temporary test pattern following a path leading towards maximization of TCD. Results of fault simulations on benchmark circuits have shown that DAR significantly improves the fault coverage up to 18.3% as compared to Antirandom. Moreover, the computational complexity of Antirandom is reduced from exponential O(2n) to linear O(n). Next, the DARalgorithm is modified to ease hardware implementation for on-chip test generation. Therefore, the third contribution of this thesis is the design of a hardware-oriented DAR (HODA) test pattern generator architecture as an alternative to linear feedback shift register (LFSR) that consists of large number of memory elements. Parallel concatenation of the HODA architecture is designed to reduce the number of memory elements by implementing bit slicing architecture. It has been proven through simulation that the proposed architecture has increased fault coverage up to 66% and a reduction of 46.59% gate count compared to the LFSR. Consequently, this thesis presents uniform and scalable test pattern generator architecture for built-in self-test (BIST) applications and solution to maximum distance test pattern generation for high fault coverage in black-box environment.