KR100567896B1 - Automatic defect classification method - Google Patents

Abstract

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an automatic defect classification method, comprising: an automatic defect classification method for comparing coded defect information of a defect image with feature information of defect classifications stored in a database, and assigning a code of a classification in which a feature matches. Comparing the defect image imaged by the reference image with the reference image to extract the image of the defect portion, and comparing the extracted unnecessary image with the unwanted pattern image set to filter defects occurring only in a specific pattern with high intensity Removing unnecessary defect patterns from the defect image, quantifying the feature amount of the extracted defect as defect information, and removing the unnecessary defect pattern from the unwanted pattern filtering, and unnecessary defects set to filter unnecessary defects having low intensity. Filter unwanted defects by comparing information with quantified defect information The step includes comparing the quantized defect information passed through the filtering process with the feature information of the defect classifications and assigning and classifying codes of matching classifications. The inspection efficiency and automatic defect classification accuracy of the wafer inspection system are included. There is an advantage that is improved.

Wafer Inspection System, Defect Classification, Real-Time Classification, Defect Filtering

Description

Automatic defect classification method {AUTOMATIC DEFECT CLASSIFICATION METHOD}

1 is a block diagram of a general wafer inspection system;

2 is an example of a defect pattern for explaining a defect recognition method in a general wafer inspection system;

3 is a flowchart illustrating an automatic defect classification execution process according to the prior art;

4 is a flowchart illustrating an automatic defect classification execution process according to the present invention.

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to wafer inspection of wafers and, more particularly, to a defect classification method for automatically classifying defects.

In recent years, yield improvement is rapidly progressing in the manufacturing process of semiconductor devices. The improvement of process equipment and process technology greatly contributed to this yield improvement, but in-line monitoring applied in actual manufacturing stage also contributed greatly. That is, conventionally, defects that may occur in the manufacturing process of semiconductor devices were inspected by indirect monitoring. As a result, it was difficult to take immediate measures when problems occurred.

In-line monitoring, however, refers to the inspection of actual wafers on a per-process basis using a wafer inspection system. Therefore, based on the inspection results obtained from the wafer inspection system, process defects and equipment defects that occur aperiodically in each unit process can be detected quickly and normalized quickly. Therefore, it is possible to minimize the production cost by maximizing the yield and reducing the waste of the production cost caused by the failure by quickly finding and normalizing the occurrence of the defect in a short time.

In the in-line monitoring process, the wafer inspection system is used to inspect defects on the wafer using a laser or an optical system to detect defects on the wafer. ) And shape, color and density of defects, and location.

1 is a schematic diagram illustrating a data processing method of a general wafer inspection system.

Referring to FIG. 1, an image of a wafer on which a conductive pattern such as a metal pattern or a polysilicon pattern doped with impurities is formed is obtained through the imaging apparatus 10. The signal of the analog state is converted into the signal of the digital state through the analog-to-digital converter 20 for processing the recognized image in a digitized state in the computer. As described above, the image converted into the digital state is filtered by the image processing computer 30, and the defects included in the image are extracted by comparing them with a reference image. .

Subsequently, the image processing computer 30 detects the size, color, and shape of the extracted defect, and confirms the density and position of such defect. The image of which defects have been confirmed is clustered and reclassified in the review station 40 configured in a wafer inspection system and in-line or off-line. (classification). After the defect classification is completed, a defect file 50 for the wafer to be finally inspected is created, and used as a process monitoring or control means.

2 is an example of a defect pattern for describing a defect recognition method in a general wafer inspection system.

Referring to Fig. 2, the image processing computer adopts a method of comparing the image 60 obtained from the wafer to be inspected with the reference image 65 with each other. At this time, the conductive pattern image 71 normally formed on the chip in the reference image 65 is canceled compared with the image 60 to be inspected, and only the abnormal defect image is extracted, and then the position and number thereof are recognized. In the drawing, after the two images 60 and 65 are canceled by comparing with each other, it can be confirmed that a total of four defects 73, 75, 77 and 79 have been extracted.

Here, the defects 73, 75, 77, and 79 may be classified into fatal defects which directly affect the operation of the chip formed on the wafer and non-fatal defects having a relatively small influence. That is, the conductive pattern defects 73, 75, 77, and 79, which may be removed through a subsequent process such as cleaning and the conductive pattern defect 73 which causes a short between the conductive patterns and make the chip inoperable, are eliminated. It exists.

However, non-fatal defects, which are relatively insignificant as described above, may be included in images to be inspected from hundreds to tens of thousands in some cases. Such non-fatal defects may be included in a microscope or a scanning electron microscope (SEM). You should check it out with a review station like Microscope and reclassify it. These operations may show a lot of variation depending on the skill and experience of the operator, and the excessive time, manpower, and cost of the operation may be a major cause of reducing inspection efficiency of the wafer inspection system.

In order to make up for the shortcomings of the wafer inspection system, a real time classification method is employed. The real time classification method filters non-fatal defects to improve the inspection efficiency of the wafer inspection system. It is mounted in the form of a program on the processing computer.

3 is a flowchart of an automatic defect classification process employing a real-time classification method according to the prior art.

First, the operator classifies each defect based on experience and assigns a classification code to each classified group. The feature information of each group to which a classification code has been assigned in this way is stored in a database as information of classifications 1 to N.

Then, the operator collects defects which are not desired to be given a defect classification code as unnecessary defects in the defect classification, and sets and stores unnecessary defect information (S81).

Thereafter, only the image of the defect site is extracted by comparing the defect image imaged by the imaging device with the reference image, and the feature amounts such as the size and color of the extracted defect are digitized as defect information (S82 to S85).

Then, the numerical defect information and the unnecessary defect information set by the operator are compared to filter the defects that the operator does not want to be given a defect classification code. That is, the non-fatal defects are filtered out of the non-fatal defects which have a direct influence on the operation of the chip formed on the wafer and the non-fatal defects having a relatively small influence (S86).

Next, the numerical defect information of the fatal defect that passed the filtering process and the characteristic information of the classifications 1 to N stored in the database are compared, and the classification code corresponding to the characteristic is given, and the classification result is output to a file (S87). ).

However, the conventional automatic defect classification method as described above has the following problems.

In the case of metal wiring in the semiconductor process, various metals constituting the metal wiring and a silicon oxide film are present between the wirings, and thus, a high lock or color variation occurs according to the metal structure. These defects have very high intensity and are very difficult to remove.

Similarly, in the real-time classification method, defects such as high-lock are so high in intensity that they are so large that they can be eliminated even when the defects are set as unnecessary defects for removal.

 The present invention has been proposed to solve such a conventional problem, and the unnecessary defects having a high intensity and occurring only in a specific pattern are set to the unnecessary pattern image itself as well as unnecessary defects having low intensity as unnecessary defect information. After that, separate and filter defect images with high intensity and defect images with low intensity separately, thereby filtering non-fatal defects to improve inspection efficiency of the wafer inspection system, while preventing actual defects, thereby improving accuracy of automatic defect classification. The purpose is to make it possible.

The automatic defect classification method according to the present invention for realizing the above object is an automatic defect that compares the numerical information of the defect image of the defect image with the characteristic information of the defect classifications stored in the database and gives a code of classification matching the characteristics. A classification method, comprising: extracting an image of a defect site by comparing a defect image imaged by an imaging device and a reference image, and extracting an unnecessary pattern image and the extracted unnecessary pattern image set to filter defects generated only in a specific pattern with high intensity Comparing the defect images to remove the unnecessary defect patterns from the extracted defect images, and for the defect images not removed by the unnecessary pattern filtering, quantifying the feature amounts of the extracted defects as defect information, and having low intensity. Having unwanted defect information set up to filter unwanted defects having Comparing the quantized defect information to filter the unnecessary defects, and comparing the quantized defect information passing through the filtering processes with feature information of the defect classifications, and assigning and classifying a code of a classification having a matching characteristic. do.

Hereinafter, an automatic defect classification method of the present invention will be described in detail with reference to the accompanying drawings.

As shown in the flowchart of FIG. 4, the automatic defect classification process according to the present invention has a low intensity and a form that is not confused with actual defects among the defects that an operator does not want to be given a defect classification code as an unnecessary defect in a defect classification. Collecting and storing unnecessary defect information (S101), and setting and storing the pattern itself as an unnecessary pattern image which does not want to be given a defect classification code in order to filter defects that have high intensity and occur only in a specific pattern. Step S102, comparing the defect image imaged by the imaging device with the reference image to extract an image of the defect site (S103 to S105), comparing the set unnecessary pattern image with the extracted defect image, and Removing the unnecessary defect pattern from the extracted defect image (S106), and the unnecessary pattern The defect image that is not removed by filtering digitizes the feature amount of the extracted defect as defect information (S107), and compares the set unnecessary defect information with the digitized defect information to filter the unnecessary defect (S108). And comparing the digitized defect information passing through the filtering processes with the feature information of the defect classifications stored in the database, and assigning and classifying a code of a classification having a matching feature (S109).

Referring to the defect classification process by the automatic defect classification method of the present invention configured as described above in more detail as follows.

First, the operator classifies each defect based on experience and assigns a classification code to each classified group. The feature information of each group to which a classification code has been assigned in this way is stored in a database as information of classifications 1 to N.

Then, the operator collects defects which are not desired to be given a defect classification code as unnecessary defects in the defect classification, and sets and stores unnecessary defect information. In this case, the defects set as unnecessary defects are limited to defects having a low intensity and a form that are not confused with actual defects (S101).

In addition, the operator sets and stores the pattern itself as an unnecessary pattern image that does not want to be given a defect classification code in order to filter defects occurring only in a specific pattern with high intensity (S102).

Thereafter, only the image of the defective portion is extracted by comparing the defect image imaged by the imaging device with the reference image (S103 to S105).

Then, the extracted defective image is compared with the unnecessary pattern image stored in step S102 to remove the unnecessary defect pattern that the operator does not want to be given a defect classification code. In other words, the intensity defect or the unnecessary defect having a large size is filtered (S106).

Next, the defect image not removed by unnecessary pattern filtering digitizes the feature amount such as the size and color of the extracted defect as defect information (S107).

Then, the numerical defect information is compared with the unnecessary defect information set in step S101 by the operator to filter the defects that the operator wants not to be given a defect classification code. That is, the non-fatal defects are filtered out of the non-fatal defects which have a direct influence on the operation of the chip formed on the wafer and the non-fatal defects having a relatively small influence (S108).

Next, the numerical defect information of the fatal defects which passed all the filtering processes and the characteristic information of the classifications 1 to N stored in the database are compared, the codes of the classifications matching the characteristics are given, and the classification results are output to a file ( S109).

In the above description, but limited to one embodiment of the present invention, it is obvious that the technology of the present invention can be easily modified by those skilled in the art. Such modified embodiments should be included in the technical spirit described in the claims of the present invention.

As described above, in the present invention, a defect image having a high intensity and a defect image having a high intensity after setting a pattern itself as an unnecessary pattern image and unnecessary defects having a low intensity are set as unnecessary defect information By separately separating and filtering defect images having low intensity, the non-fatal defects are filtered to improve inspection efficiency of the wafer inspection system, but the actual defects are not filtered, thereby improving accuracy of automatic defect classification.

Claims (1)

An automatic defect classification method for comparing coded defect information of a defect image with feature information of defect classifications stored in a database and assigning a code of a classification in which the features match. Comparing the defect image imaged by the imaging device with the reference image to extract an image of the defect site; Comparing the extracted defect image with an unnecessary pattern image set to filter defects occurring only in a specific pattern with high intensity, and removing the unnecessary defect pattern from the extracted defect image; The defect image not removed by the unnecessary pattern filtering is digitized as defect information of the feature amount of the extracted defect; Filtering the unnecessary defects by comparing the unnecessary defect information set to filter unnecessary defects having a low intensity with the quantified defect information; Classifying by comparing the quantized defect information passed through the filtering processes with the feature information of the defect classifications and assigning a classification code corresponding to a feature to classify the defects. Automatic defect classification method comprising a.

Images