Neural networks as massively parallel automatic test pattern generators

Neural networks are characterized by small-grain parallelism where a large number of inexpensive neurons (processors) can act simultaneously toward solving a given optimization problem. Neural networks present a promising paradigm for computationally intensive CAD applications like automatic test pattern generation (ATPG) for digital circuits. The performances of 2-Valued and 3-Valued neural networks as ATPGs are compared. The performance data are obtained by implementing the neural network-based ATPGs on the Myrias Scalable Parallel Supercomputer.<
>