Fault Coverage Enhancement via Weighted Random Pattern Generation in BIST Using a DNN-Driven-PSO Approach

Conventional pseudo-random testing in Built-InSelf-Test (BIST) usually requires a huge amount of testing time. This issue can be addressed with a Weighted Random Pattern generation that can produce test patterns in order to achieve high fault coverage with the fewer number of test vectors. Determining such input weights for a particular circuit is an NP-hard problem. In this paper, we have proposed a technique to converge to a high-quality input weight vector using a Particle Swarm Optimizer (PSO) with the help of a Deep Neural Network (DNN). The DNN prediction of fault coverage value as well as the parallel training of the DNN along with the evolution of the PSO makes this significantly fast. The technique has been tested with ISCAS'85 and ISCAS'89 benchmark circuits. The result shows that the DNN gets capable of predicting the fault coverage values accurately for weight assignments suggested by the particles in PSO. Also, it is observed that the proposed approach is very efficient in covering a large number of faults with less test vectors in self-testing circuits.