An Embarrassingly Simple Approach for Wafer Feature Extraction and Defect Pattern Recognition

TL;DR

This paper introduces a simple, fast, and explainable feature extraction method for wafer defect pattern recognition that outperforms traditional deep learning models without requiring training or fine-tuning.

Contribution

It presents an extremely simple, non-parametric feature extraction technique that is efficient, interpretable, and suitable for online manufacturing defect detection.

Findings

Outperforms conventional deep learning models in accuracy.

Requires no training or fine-tuning.

Preserves shape and location information of data points.

Abstract

Identifying defect patterns in a wafer map during manufacturing is crucial to find the root cause of the underlying issue and provides valuable insights on improving yield in the foundry. Currently used methods use deep neural networks to identify the defects. These methods are generally very huge and have significant inference time. They also require GPU support to efficiently operate. All these issues make these models not fit for on-line prediction in the manufacturing foundry. In this paper, we propose an extremely simple yet effective technique to extract features from wafer images. The proposed method is extremely fast, intuitive, and non-parametric while being explainable. The experiment results show that the proposed pipeline outperforms conventional deep learning models. Our feature extraction requires no training or fine-tuning while preserving the relative shape and location of data points as revealed by our interpretability analysis.