Inspection system and method

Patent metadata

Field	Value
Publication number	US-2020334800-A1
Application number	US-201916388435-A
Country	US
Kind code	A1
Filing date	Apr 18, 2019
Priority date	Apr 18, 2019
Publication date	Oct 22, 2020
Grant date	—

Abstract

A system includes an inspection device and an image processing unit. The inspection device is configured to scan a wafer to generate an inspected image. The image processing unit is configured to receive the inspected image, and is configured to analyze the inspected image by using at least one deep learning algorithm in order to determine whether there is any defect image shown in a region of interest in the inspected image. When there is at least one defect image shown in the region of interest in the inspected image, the inspection device is further configured to magnify the region of interest in the inspected image to generate a magnified inspected image for identification of defects.

First claim

What is claimed is:

1 . A system, comprising:
an inspection device configured to scan a wafer to generate an inspected image; and an image processing unit configured to receive the inspected image, and configured to analyze the inspected image by using at least one deep learning algorithm in order to determine whether there is any defect image shown in a region of interest in the inspected image, wherein when there is at least one defect image shown in the region of interest in the inspected image, the inspection device is further configured to magnify the region of interest in the inspected image to generate a magnified inspected image for identification of defects.

2 . The system of claim 1 , wherein when there is the at least one defect image shown in the region of interest in the inspected image, the image process unit is configured to analyze the inspected image by using the at least one deep learning algorithm to generate at least one reference image, and to compare the at least one reference image with the inspected image in order to determine whether there is any defect image shown in the region of interest in the inspected image.

3 . The system of claim 2 , further comprising:
an image database unit configured to store the at least one reference image, wherein the at least one reference image, wherein the at least one reference image are generated by using the at least one deep learning algorithm including a convolutional neural network (CNN) with multi-box feature mapping, a transfer learning, or the combination thereof.

4 . The system of claim 1 , wherein when there is no defect image shown in the region of interest in the inspected image, the inspection device is further configured to generate a real-time image of a region different from the region of interest.

5 . The system of claim 1 , wherein the at least one deep learning algorithm comprises:
a convolutional neural network; a multi-box feature mapping; and a transfer leaning.

6 . The system of claim 1 , wherein the image processing unit is further configured to train the at least one deep learning algorithm by learning the inspected image.

7 . The system of claim 1 , wherein the inspection device comprises a scanning electron microscope (SEM) type device, and the image processing unit is configured inside the SEM type device.

8 . A method, comprising:
generating, by an inspection device, a first inspected image of a region on a wafer; identifying if there is any defect image shown in a region of interest in the first inspected image by applying at least one deep learning algorithm; and when there is at least one defect image shown in the region of interest in the first inspected image, generating, by the inspection device, a second inspected image including magnification of the at least one defect image in the region of interest.

9 . The method of claim 8 , further comprising:
when identifying if there is at least one defect image, the inspection device generating no reference image to be compared with the first inspected image.

10 . The method of claim 8 , further comprising:
when identifying if there is no defect image, the inspection device generating a real-time image to be compared with the first inspected image.

11 . The method of claim 8 , further comprising:
training, by an image processing unit, the at least one deep learning algorithm based on the first inspected image and the second inspected image.

12 . The method of claim 11 , wherein training the at least one deep learning algorithm comprises:
optimizing labels of defects on the region of interest; randomly modifying the first inspected image and the second inspected image to create plurality of inspected images; balancing numbers of defect sampling; and applying a nuisance training.

13 . The method of claim 8 , wherein applying at least one deep learning algorithm comprises:
applying a convolutional neural network, a multi-box feature mapping, a transfer learning, or a combination thereof.

14 . The method of claim 8 , further comprising:
cleaning, by a cleaning device, the wafer based on the second inspected image.

15 . A method, comprising:
scanning a wafer, by an inspection device, to generate a first inspected image; applying at least one deep learning algorithm to analyze the first inspected image, to determine whether there is any defect image shown in a region of interest in the first inspected image; and when there is at least one defect image shown in the region of interest in the first inspected image, magnifying, by the inspection device, the region of interest in the first inspected image, to generate a second inspected image for identification and classification of defects.

16 . The method of claim 15 , wherein when there is no defect image shown in the region of interest in the first inspected image, scanning the wafer, by the inspection device, to generate a third inspected image as a reference image to be compared with the first inspected image, for identification of defects.

17 . The method of claim 15 , further comprising:
when analyzing the first inspected image, the inspection device generating no reference image to be compared with the first inspected image.

18 . The method of claim 15 , further comprising:
training, by an image processing unit, the at least one deep learning algorithm by optimizing labels, modifying the first inspected image and the second inspected image, balanced sampling, and nuisance training.

19 . The method of claim 15 , wherein applying the at least one deep learning algorithm comprises:
applying a convolutional neural network, a multi-box feature mapping, a transfer learning, or a combination thereof.

20 . The method of claim 15 , further comprising:
training, by an image processing unit, the at least one deep learning algorithm based on the first inspected image and the second inspected image, in order to perform determining whether there is any defect image in a new inspected image.

Assignees

• Taiwan Semiconductor Mfg Co Ltd

Inventors

• Chou Chung-Pin
• Lin In-Tsang
• Huang Sheng-Wen
• Wang Yu-Ting
• Lai Jui-Kuo
• Chou Hsin-Hui
• Liu Jun-Xiu
• Wang Tien-Wen

Classifications

• H10P70/20

  Cleaning during device manufacture · CPC title

• H10P74/203 Primary

  Structural properties, e.g. testing or measuring thicknesses, line widths, warpage, bond strengths or physical defects · CPC title

• G06V30/2504

  Coarse or fine approaches, e.g. resolution of ambiguities or multiscale approaches · CPC title

• G06V10/82

  using neural networks · CPC title

• G06V10/764

  using classification, e.g. of video objects · CPC title