Polycrystalline Silicon Wafer Scratch Segmentation based on Deep Convolutional Autoencoder

Navin Ranjan, Sovit Bhandari, Yeong-Chan Kim, Hoon Kim

Abstract: Silicon chips are the backbone of the current digital era, and it is crucial to find or detect defects like a scratch on the surface of the silicon wafer during production to improve the yield. In the past, traditional machine learning and image processing methods based on handcrafted models or features have heavily relied on expertise, which is inefficient and expensive. This paper presents a machine learning-based method to segment scratch from Integrated Circuit images without any handcrafted features. The proposed method is based on a deep Convolutional Autoencoder, which learns the spatial relations between the pixels at the encoder and generates the scratch segmented image at the decoder. Our experiments show that our proposed model can efficiently and effectively learn to produce a scratch segmented image.

External IDs:dblp:conf/elinfocom/RanjanBKK22