No abstract

No abstract

Wafer acceptance test (WAT) is a key process of semiconductor manufacturing. The collected testing parameters can be used in identification of wafer defects, improvement of product yield, and control of production costs. However, WAT parameters regularly have characteristics such as high dimensions and strong redundancy, which prevent the wafer yield from accurate prediction and effective improvement. To overcome these shortcomings, a hybrid feature selection method is proposed to identify key WAT parameters influencing wafer yields. This method is composed of two stages, i.e. filter selection and wrapper selection. In filter selection, the minimum Redundancy Maximum Relevance (mRMR) filtering parameter pre-screening criterion based on mutual information (MI) is proposed. The relevance between each parameter and the wafer yield value is calculated by MI. At the same time, the criterion of MI is used to measure the redundancy between each parameter to select the minimum redundancy parameters, and reduce feature size for further searches. In wrapper selection, a wrapped key parameter identification model based on genetic algorithm (GA) and deep belief network (DBN) is designed. The coding and optimization of candidate input parameters are realized by GA. The wafer yield prediction error value of the DBN and the weight of the selected features are solved as the fitness function to realize the selection process of the combined parameters. In experiment, both testing data sets and industrial data are used to demonstrate the efficiency of this proposed method.

No abstract

No abstract