Simulation Studies of Two-Layer Hopfield Neural Networks for Automatic Wafer Defect Inspection

Abstract

The occurrence of defect on a wafer may result in losing the yield ratio. The defective regions were usually identified through visual judgment with the aid of a scanning electron microscope. Dozens of people visually check wafers and hand-mark their defective regions leading to a significant amount of personnel cost. In addition, potential misjudgment may introduce due to human fatigue. In this paper, a two-layer Hopfield neural network called the competitive Hopfield wafer-defect detection neural network (CHWDNN) is proposed to detect the defective regions of wafer image. The CHWDNN extends the one-layer 2D Hopfield neural network at the original image plane to a two-layer 3D Hopfield neural network with defect detection to be implemented on its third dimension. With the extended 3D architecture, the network is capable of incorporating a pixel’s spatial information into a pixel-classifying procedure. The experimental results show the CHWDNN successfully identifies the defective regions on wafers images with good performances.