Application of Convolutional Neural Network to TSOM Images for Classification of 6 nm Node Patterned Defects

TL;DR

This paper demonstrates that a CNN combined with TSOM imaging can accurately classify 6 nm node patterned defects, including very small size variations, using simulated optical images at 193 nm wavelength.

Contribution

It introduces a novel approach integrating CNN with TSOM imaging for classifying extremely small semiconductor defects at the 6 nm node.

Findings

Successful classification of eight defect variations including 3 nm size difference

CNN-TSOM method works with simulated images at 193 nm wavelength

Achieved classification of features much smaller than the illumination wavelength

Abstract

With the rapid growth in the semiconductor industry, it is becoming critical to detect and classify increasingly smaller patterned defects. Recently machine learning, including deep learning, has come to aid in this endeavor in a big way. However, the literature shows that it is challenging to successfully classify defect types at the 6 nm node with 100% accuracy using low-cost and high-volume-manufacturing compatible optical imaging methods. Here we combine a convolutional neural network (CNN) with that of an optical imaging method called through-focus scanning optical microscopy (TSOM) to successfully classify patterned defects for the 6 nm node targets using simulated optical images at the 193 nm illumination wavelength. We demonstrate the successful classification of eight variations of the defects, including the 3 nm difference in the defect size in one dimension, which is over 50 times smaller than the illumination wavelength used.