Research and System Implementation of Machine Vision-Based Critical Dimension and Overlay Metrology

Abstract

This paper proposes a machine vision-based method for chip appearance inspection to address issues in chip manufacturing, such as reliance on manual inspection, low efficiency, poor repeatability, and operator-dependent results. The method uses images from an industrial Charge-Coupled Device (CCD) camera and microscope, employing grayscale processing and Region of Interest (ROI) template construction for detection region initialization. Normalized Cross-Correlation (NCC) is used for automatic region localization. The method combines mean filtering, adaptive threshold segmentation, two-pass connected domain analysis, and Hough geometric fitting to automatically measure key indicators such as linewidth and alignment deviation. Experimental results show good repeatability and consistency, with standard deviations of the Overlay (OVL) errors in the X and Y directions being 0.00016 and 0.00062, respectively, and the Critical Dimension (CDX) measurement error being 0.044. The processing time per point is about 0.02 seconds, providing a 200-fold efficiency improvement over manual inspection. The method offers simplicity, strong interpretability, and edge-deployment friendliness, making it an effective solution for online appearance inspection in chip manufacturing.