Neural computing for built-in self-repair of embedded memory arrays

Abstract

A demonstration is presented of how to represent the objective function of the memory repair problem as a neural network energy function, and how to utilize the neural net's convergence property to find near-optimal solutions. Two algorithms have been developed using a neural network, and their performance is compared with the 'repair most' algorithm that is used commercially. For randomly generated defect patterns, the proposed algorithm with a hill-climbing capability has been found to be successful in repairing memory arrays in 98% of the cases, as opposed to the repair most algorithm's 20% of cases. >