Semiconductor yield loss' causes identification: A data mining approach

Abstract

Semiconductor manufacturing processes are known to be long and complex. Starting from a silicon wafer, multiple treatments are applied for about three months. Mastering the manufacturing process as well as a rapid identification of yield loss' causes are the keys to a successful semiconductor fabrication plant. The production cycle is composed of a combination of production and quality inspection steps. Data collected at production and quality control steps, lead to huge heterogeneous databases. In order to understand yield loss causes, we propose a KDD (Knowledge Discovery from Databases) approach, which explores the knowledge hidden in these multiple databases, by identifying, first, clusters in the different databases and, second, relational patterns between them. These relational patterns represent potential yield loss' causes.