Human based knowledge for the probe failure pattern classification with the use of a backpropagation neural network. Application on submicron linear technologies

Abstract

The practical use of what is known as soft computing (neural networks, fuzzy logic, genetic algorithms, etc.) is starting to offer important advantages in several fields. In particular, in a high-cost environment like the semiconductor arena, the application of those, up to now, research techniques offers an attractive alternative to the traditional approaches of yield enhancement. For increasing wafer diameters and more compact technologies, where the effect of tiny defects produces fatal consequences, a yield enhancement strategy based on inspections requires the synergy of intelligent new tools that, on the other hand, have a fraction of cost of the current inspection machines. This new strategy is used to classify and analyse all the production of a fab in a systematic way, providing new possibilities to improve yields without penalising cycle time, cost and reaching inspection levels impossible to achieve without this new approach.