KR101936628B1 - Method of acquiring TSOM image and method of examining semiconductor device - Google Patents

Abstract

Acquiring actual images of a plurality of different focus positions and an out-of-focus distance (distance) of each actual image through an optical means with respect to the object to be inspected, based on the actual images and the degree of deviation of each actual image from the focus Acquiring a plurality of virtual images having different focus positions from an actual image and a focus position thereof; obtaining a TSOM image for the object to be inspected using the real image and the virtual image; A method is disclosed. According to the TOSM image acquisition method and the semiconductor device inspection method of the present invention, it is possible to obtain a highly accurate TSOM image even when a smaller effort and time are applied compared with the conventional technique, and thereby, the semiconductor device can be efficiently and economically tested .

Description

TECHNICAL FIELD [0001] The present invention relates to a TSOM image acquisition method and a semiconductor device inspection method,

The present invention relates to a semiconductor device inspection method, and more particularly, to a TSOM image acquisition method and a semiconductor device inspection method for inspecting a semiconductor device.

Recently, the National Institute of Standards and Technology (NIST), Ravikiran Attota et al. Have demonstrated the possibility of measuring three-dimensional fine patterns using a through focus scanning optical microscopy (TSOM) .

This technique uses a conventional optical microscope, but uses a method of creating a three-dimensional image data space for an object by collecting two-dimensional images at different focus positions for the same object. Thus, the resulting two-dimensional images constitute a through-focus image comprising a plurality of in-focus images and out-of-focus images that are out of focus . Computer processing of such a three-dimensional image data space is performed. The computer extracts a brightness profile from a plurality of through-focus images for the same subject being collected and uses the focus position information to create a through-focus scan optical microscope (TSOM) image.

The TSOM image formed by the images provided by the through-focus scanning optical microscope (TSOM) does not specifically represent the object but is somewhat abstract, but the difference between the images is inferred from the fine difference in shape of the measured three- .

Through simulation studies, it has been known that through-focus scanning optical microscopy (TSOM) images can measure characteristics below 10 nanometers, suggesting possibilities for fine three-dimensional structure shape analysis.

Although TSOM technology provides a method for inspecting semiconductor devices with the precision of electron microscope level using optical equipment, in order to use TSOM technology, it is necessary to acquire plural images with different depth of focus. For this purpose, It is necessary to change the setting of the optical equipment such as changing the position of the object or the position of the lens of the inspection equipment, thereby delaying the inspection of the semiconductor device quickly at the inspection site and making it difficult to apply in the field.

In other words, in order to improve the accuracy of the inspection through the TSOM image, it is necessary to acquire images for a larger number of focus positions and acquire the TSOM image through them. However, in order to do so, There is a disadvantage that it is necessary, costly and time consuming.

On the other hand, since the TSOM technique does not use the actual image of the object but determines the shape and size of a specific part of the actual semiconductor device through the pattern or determines whether the TSOM image is normal or normal, It is not possible to determine whether the TSOM image is normal or not. By knowing that a certain pattern of TSOM image corresponds to a certain structure, shape and size of a semiconductor device through actual inspection using other means such as SEM, Once an image is obtained, it is possible to know the shape, size, and abnormality of the object by using the corresponding data already obtained.

For example, a TSOM image for a part of a semiconductor device that knows at least one item to be inspected (parameter) and its related items (numerical value, numerical range) is acquired, a database is created with all items, If there is an unknown test object, it obtains the TSOM image for it, compares it with the TSOM image of each item in the already created database, and then selects the item corresponding to the TSOM image with the smallest difference in the image, Of the items.

In order to compare two TSOM images, it is possible to use the MSD method in which differences in brightness or color values of each pixel in the TSOM image are obtained, the square of the difference is added to all pixels, and then divided by the number of all pixels.

However, when using this method, it is necessary to acquire a TSOM image for each item and each category of the item. To do so, it is necessary to acquire a large number of TSOM images. In order to secure each TSOM image, If the image of the subject is needed, it is very difficult to do so much preliminary work for the object inspection using the TSOM image.

In addition, there is a problem that the accuracy of the measurement drops sharply when an error occurs because the images are not precisely matched, and the time required for the test calculation increases as the data in the database increases.

As a result, when acquiring a TSOM image for an object, it is highly desirable to obtain an accurate TSOM image with only images of at least two different focus positions on the object, since the cost and time required to obtain the TSOM image can be saved.

Thus, the possibilities and advantages of TSOM technology, the use of relatively inexpensive optical equipment, can be used to perform highly accurate inspection such as electron microscopy, and can be applied to process inspection without interrupting the production process or destroying the object Despite its advantages and potentials, it has not yet been utilized effectively for practical semiconductor device inspection. At present, it is required to speed up image acquisition and analysis required for application of TSOM technology to be applied to on-site inspection in the production process of a specific semiconductor device It is required to develop a specific inspection method that can automate the process.

Deep Learning, on the other hand, is a technique used to classify or cluster objects or data. Machine learning is a type of machine learning that trains computers to classify objects as if a human brain were to distinguish them.

The basic concept of deep running, which was first used by Professor Geoffrey Hinton of the University of Toronto, USA in 2006, is similar to the Artificial Neural Network (ANN). An artificial neural network is an algorithm that processes information in a way similar to a human brain. It learns information by repeating the process of combining and separating data of various elements such as objects' faces and shapes. Hinton suggested a way to overcome the disadvantages of existing artificial neural network models in his paper. The development of hardware capable of analyzing a large amount of data and the appearance of big data (artificial neural network) show an excellent result, which is called deep running.

One of the features of Deep Learning is data classification through Unsupervised Learning. In general, data classification methods of computers are divided into Supervised Learning and Unsupervised Learning. Map learning is a method of inputting classification criteria into a computer. Conventional machine learning algorithms usually classify data into a map learning method. In order to classify data without classifier, the computer classifies the data by searching similar clusters. In order to do this, it is necessary to have high computing ability. Deep learning performs artificial neural network optimization by adjusting data by pre-training using non-bipod learning method.

For example, a computer finds a certain pattern based on input data to form a feature map. It can be extracted from very small feature to big feature. The performance of the algorithm depends on how good the characteristics are. Through the CNN (convolutional neural network) algorithm, which extracts features through various stages, it is possible to learn difficult contents as it goes up to the upper layer. The term "deep running" comes from the fact that you learn to go down to the deep level in multiple stages. When a pattern is found through a feature map in a given data and the data can be classified as if a human being identifies an object, the computer can see the newly entered picture and correctly distinguish it according to the criterion. Deep learning is a feature of the algorithm from feature extraction to learning.

This deep running has achieved many achievements in the field of speech recognition and image recognition in recent years. In the field of image recognition, based on the bio-neural network, the basic model in which the input calculates the output by the logical function and the weight, and Convolunant The extraction of feature maps using the concept of pooling is important.

However, deep-running is not yet accessible because it has not been generalized yet due to its remarkable possibility, and it is difficult to access it. For proper understanding and utilization, computer hardware and software for cognitive science or artificial intelligence, statistics, And knowledge about the surrounding knowledge, such as knowledge about the field is not enough to be used in various fields.

KR 10-1523336 B1 KR 10-1507950 B1

&Quot; TSOM method for semiconductor metrology ", Proc. SPIE 7971, Metrology, Inspection, and Process Control for Microlithography XXV, 79710T (April 20, 2011)

The present invention is to overcome the above problems of the conventional semiconductor device inspection method using a TSOM image. The present invention provides a method of obtaining a TSOM image having a high precision with respect to a target while taking less effort and time, And to provide a method for inspecting an object.

The present invention is directed to obtaining images with a sufficient number of different focal depths to obtain an accurate TSOM image without varying the optical setting or the position of the subject's focal length direction largely or frequently when acquiring images with various focal depths And a semiconductor device inspection method using the same.

An object of the present invention is to provide a new semiconductor device inspection method capable of inspecting an inspection object more accurately and quickly by embodying an inspection tool through deep running through data.

It is an object of the present invention to provide a TSOM image acquisition method having a suitability to reduce the time and cost in constructing a verification data set or database which can be a key issue in the initial execution of a semiconductor device inspection method do.

It is an object of the present invention to provide a semiconductor device inspection method that can be used for inspection of semiconductor devices more realistically and more efficiently by overcoming the limitations of a semiconductor device inspection method using TSOM.

More particularly, the present invention aims at providing a realistic method for efficiently inspecting a semiconductor device at a low cost while making good use of the advantages of TSOM by incorporating a deep learning technology into a semiconductor device inspection using a TSOM.

The present invention is based on the fact that the deep learning technology is one of the fields where the most realistic approaches and achievements are currently being realized are the field of image recognition and the depth learning The present invention is directed to a semiconductor device inspection method that can be applied to a real process inspection of a semiconductor device.

Further, in the inspection step of the semiconductor device, an inspection object image is expressed by a probability distribution method for all categories of inspection target items, and an interpolation method for adding all the representative values of each category multiplied by the probabilities included in each category is used And an object of the present invention is to provide a semiconductor device inspection method capable of minimizing an inspection error by determining numerical determination values for inspection target items of an inspection target image.

In addition, the present invention provides an image processing method, an image processing method, an image processing method, an image processing method, an image processing method, an image processing method, By using a multi-channel image including the non-matching image as a verification data set for deep running and an input image for inspection, there is no need to form a TSOM image, a registration process is not necessary, And a semiconductor device inspection method capable of shortening the inspection time.

According to an aspect of the present invention, there is provided a method of acquiring a TSOM image,

Acquiring actual images of a plurality of different focus positions and an out-of-focus distance (distance) of each actual image through an optical means with respect to the object to be inspected;

Acquiring a plurality of virtual images and their focal positions that differ in focal position from the actual image based on the degree of deviation of the actual images and the actual images from the focal point; And

And obtaining a TSOM image for the inspection object using the actual image and the virtual image.

In the TSOM image acquisition method according to an embodiment of the present invention, in the step of acquiring the plurality of virtual images and the focus position thereof,

A plurality of virtual images obtained by obtaining the actual images with respect to a plurality of different focal positions and using an interpolation method based on the focal position and the information about the actual image, The process of acquiring the focus position can be performed.

Also, in the TSOM image acquisition method according to an embodiment of the present invention, it is assumed that the virtual image obtained using the interpolation method has a Gaussian distribution of the focal position centered on a focal distance, It is possible to adopt a weighted arrangement scheme in which the focus position near the focus position where the focus is made is dense for the focus position and the focus position for the off-focus position is less.

That is, a more reliable TSOM image can be obtained by weighting the focus position around the focal distance to obtain an out-of-focus image, and finally obtaining a TSOM image reflecting the weight.

In the method of acquiring a TSOM image according to an embodiment of the present invention, the actual images may include an image that is in focus with respect to the object to be examined, and a case where the distance from the lens to the imaging surface is shorter than the actual focal distance, An image, and three images including an image which is longer in focus than the actual focal length and whose distance from the lens to the imaging surface is longer.

On the other hand, in the step of acquiring a plurality of virtual images and their focal positions, an analysis for a given optical system and a conversion formula (conversion program) indicating this analysis can be used. For example, in order to acquire a plurality of virtual images and their focal positions that differ in focal position from a small number of actual images and an out-of-focus degree of each of the actual images, an Fourier Modal Method (FMM) And acquiring a plurality of virtual images having the same focal positions as the actual images using the setting (characteristics) of the optical system obtained through the optical system analysis, and comparing the virtual images with the actual images, Acquiring a conversion formula (conversion program) for the more appropriate analysis of the virtual image and the appropriate analysis, and acquiring the focal positions of the plurality of virtual images and the plurality of virtual images having different focal positions from the actual images through the conversion formula Can be achieved.

Meanwhile, in the step of acquiring a plurality of virtual images and their focal positions, an interpolation method and a conversion formula can be used together. In the present invention, it is difficult to sufficiently analyze an optical system through about two to three actual images. Therefore, a virtual image of a certain focal position and a focus position of an actual image are primarily determined by a simple optical system analysis and conversion formula And acquiring an additional virtual image based on the real image and the virtual image by an interpolation method.

In the method of acquiring a TSOM image according to an embodiment of the present invention, a light source for irradiating an object to be inspected for obtaining an actual image may be a planar light source having a single wavelength for optical system analysis through the FMM, Monochromatic light or a single wavelength light source can be used.

According to another aspect of the present invention, there is provided a method of inspecting a semiconductor device,

Acquiring an image of a plurality of semiconductor device parts that know at least one item to be inspected (parameter) and a category (class) in the item, and performing a deep learning ) In a storage list (database) as a verification data set for the verification data set;

(TOOL) for inspection in the form of combining computer hardware and software and performing a deep run on at least one of the items to be inspected based on the plurality of images in the storage list, When the result of discrimination is compared with the result of classification by the above storage list, deep running is performed until the predetermined standard is satisfied, and the state of having the software satisfying the condition Determining a tool as an inspection tool suitable for inspection; And

And an inspection step of acquiring an inspection object image for an unknown semiconductor device portion and using the inspection tool determined through the deep learning to find out what category the inspection object image belongs to which inspection object item.

In the inspection method of a semiconductor device according to an embodiment of the present invention, it is not a simple judgment method that the inspection target image belongs to a category of the inspection target item, and the probability distribution method Lt; / RTI >

The categories are expressed as a range of values,

The numerical value determination value for the inspection target item of the inspection object image can be determined by adding all of the representative values of the respective categories multiplied by the probabilities included in the respective categories.

In the inspection method for a semiconductor device according to an embodiment of the present invention, it is not a simple determination method that a certain inspection target image belongs to which category of an inspection target item, but a probability distribution method , And the probability of each of these categories can be used to determine by the interpolation or weighted averaging how the numerical value of the corresponding category of the corresponding item in the image has a specific value.

For example, when each category consists of a numerical range, the numerical determination of the item to be inspected of the image to be inspected can be performed by adding all the categories to the representative value of each category multiplied by the probability to be included in the category, It can be a median or mode often used in statistics.

In addition to the above, there are also methods of discarding values having a probability less than a certain threshold value among the probability values of the respective classes, eliminating outliers, obtaining a meaningful probability category, multiplying each representative value with a probability value, We introduce multidimensional equations by using probability values of categories instead of simple multiplication and linear interpolation methods, and apply it to interpolation method (nonlinear interpolation method). It is also possible to expand.

Further, in the semiconductor device inspection method according to an embodiment of the present invention, an image of a plurality of semiconductor device parts that knows at least one inspection target item (parameter) and a category (class) The image to be inspected for the semiconductor device portion is a through-focus scanning optical microscope (TSOM) image,

Wherein the TSOM image for a plurality of semiconductor device parts that know the at least one item to be inspected (parameter) and the category (class) within the item, and the TSOM image to be inspected for the unknown semiconductor device part,

Acquiring actual images of a plurality of different focal positions and an out-of-focus degree (distance) of each actual image through an optical means for an object to be inspected for the plurality of semiconductor device parts or the unknown semiconductor device part;

Acquiring a plurality of virtual images and their focal positions that differ in focal position from the actual image based on the degree of deviation of the actual images and the actual images from the focal point; And

A TSOM image for a plurality of semiconductor device parts that know the at least one inspected item (parameter) and the category (class) within the item using the real image and the virtual image, and a test for the unknown semiconductor device part And obtaining the target TSOM image using the TSOM image acquisition method.

In the semiconductor device inspection method according to an embodiment of the present invention, in the step of acquiring the plurality of virtual images and the focal position thereof,

A method for obtaining a plurality of actual images and real images for a plurality of different focal positions and for obtaining a plurality of focal positions different from a focal position by using an interpolation method based on the focal position and information And a process of acquiring the virtual image and the focus position of the virtual image may be performed.

In the method for inspecting a semiconductor device according to an embodiment of the present invention, it is assumed that the virtual image obtained using the interpolation method has a Gaussian distribution of the focal position around a distance at which the focal point is focused,

It is possible to adopt a weighted arrangement scheme in which the focus position near the focus position where the focus is focused is dense for the focus position for the virtual image and the focus position for the focus position that is out of focus is less.

In the method for inspecting a semiconductor device according to an embodiment of the present invention, the actual images may include an image focused on the inspection object of the plurality of semiconductor device parts or the unknown semiconductor device part, An image that is out of focus with a shorter distance from the lens to the imaging surface and an image that is out of focus with a longer distance from the lens to the imaging surface than the actual focal length.

In a method of inspecting a semiconductor device according to an embodiment of the present invention, a step of acquiring a TSOM image of a plurality of semiconductor device patterns that recognize the shape and size, and associating the TSOM image with a pattern type and size, To reduce the time and cost of acquiring the TSOM image, the photographs of the actual different focus depths obtained through the optical means may be three or more per pattern, interpolated to artificially include a greater number of different depths of focus Acquire images, and obtain more TSOM images from these more pictures.

According to another aspect of the present invention, there is provided a method of inspecting a semiconductor device, the method comprising the steps of: acquiring a focal position of each of the plurality of virtual images and the plurality of virtual images;

The optical system is analyzed using FMM (Fourier Modal Method)

Acquiring a plurality of virtual images having the same focal position and actual images using the setting (characteristics) of the optical system obtained through the optical system analysis, comparing the virtual images with the actual images,

Based on the comparison, a more suitable analysis of the optical system and a conversion formula (conversion program) for the appropriate analysis are obtained,

And acquiring a focal position of each of the plurality of virtual images and the plurality of virtual images having different focal positions from the actual images through the conversion formula.

In the method for inspecting a semiconductor device according to an embodiment of the present invention, a planar light source having a single wavelength may be used as a light source for irradiating an object to be inspected to obtain the actual image for optical system analysis through the FMM.

Further, in the semiconductor device inspection method according to an embodiment of the present invention, an image of a plurality of semiconductor device parts that knows at least one inspection target item (parameter) and a category (class) The image to be inspected for the semiconductor device portion includes,

Wherein an image focused on a plurality of semiconductor device parts or an unknown semiconductor device part that knows the at least one inspection target item (parameter) and the category (class) in the item is focused and compared with an actual focal distance M images that are shorter than the actual focal length, and M images that are shorter in the distance from the lens to the imaging plane and N images that are longer in focus than the actual focal length, M and N may be arbitrary integers of 1 or more and 4 or less.

Also, in the method of inspecting a semiconductor device according to an embodiment of the present invention, M and N may be the same, and M and N may be one.

Further, in the semiconductor device inspection method according to the embodiment of the present invention, since the tool is adapted to the inspection target item through the deep learning, it is checked whether the result of the tool is correct at the intermediate stage where the deep learning tool is not determined The data in the storage list is firstly divided into items and categories so that the individual images can be identified to which items and categories.

However, it is very difficult at the initial stage to prepare sufficient images for all items and categories and for each category at this time, so the prepared images can be prepared through simulation. Of course, at this time, the simulation program confirms at least a part of the images prepared by the simulation through actual measurement, prepares it so that there is not much variation, and implements the simulation by reflecting the interpolation method in general.

Further, in the method of inspecting a semiconductor device according to an embodiment of the present invention, when deep running is performed on at least one of the inspection target items, the verification data set (or software) The method comprising the steps of: first searching for a feature of a plurality of images of the plurality of images, discriminating the plurality of images by the feature, comparing the plurality of images with the classification result by the storage list, determining a tool in the current state as the inspection tool, If the predetermined criterion is not satisfied, the process of searching for a new feature again while modifying the basic tool through the algorithm modification is repeated until the predetermined criterion is satisfied, or until the predetermined criterion is satisfied, Can be determined by the inspection tool.

Further, in the semiconductor device inspection method according to an embodiment of the present invention, the tool creates one individual tool for one parameter and forms a set of individual tools for all necessary items (parameters) for inspection in this manner Or the like.

In the method for inspecting a semiconductor device according to an embodiment of the present invention, when determining a tool for inspection for a plurality of items among the inspection target items in the determination step, the basic tool is subjected to deep running To determine a plurality of individual tools for the plurality of items,

In the inspection step, the plurality of individual tools may be applied to the inspection target image to find out what category the inspection target image belongs to for each item constituting the plurality of items.

In the method of inspecting a semiconductor device according to an embodiment of the present invention, the inspection target item or parameter may be a hole of a semiconductor device, a through silicon via (TSV), a groove, The bottom width, the bottom width, the depth, the height, and the inclination angle of each of the LINE patterns, and the category may be a numerical range to which the items belong. In the case of hole or through silicon vias (TSV), the diameter may be the diameter of an inlet of a hole or a via, a diameter of a bottom, a depth, and the like.

When the shape and size of the unknown semiconductor device portion are determined through the present invention, the defects are determined by comparing the shape and the size of the unknown semiconductor device portion with a normal pattern, and the wafer or die having the detected defect can be reworked or discarded. The item itself is abnormal and the category can be an abnormal or normal decision value.

In the present invention, when storing items, categories, and images in a storage list (database) with respect to a semiconductor device portion (object to be inspected), the database can be maintained in an open state so as to be constantly supplemented.

For example, if a defect is determined through the method of the present invention with respect to an unknown semiconductor device portion, another test such as SEM is applied to it to determine a specific item and category of defect, and the result is added to the save list. (Including the database). By using this tool, a new unknown semiconductor device portion can be inspected and the inspection result can be obtained.

According to the TSOM image acquisition method of the present invention, it is possible to obtain a TSOM image with high accuracy for a target object while taking less time and effort compared with the conventional technique, thereby enabling efficient and economical inspection of a semiconductor device.

According to the TSOM image acquisition method of the present invention, when acquiring an image with various focus depths, a sufficient number of different focus depths to obtain an accurate TSOM image without changing the optical setting or the position of the focal distance direction of the object largely or frequently In the case of applying the present invention to a semiconductor device inspection method of the present invention, that is, a semiconductor device inspection method using deep drawing and a TSOM image, It becomes easy to secure a large number of TSOM images that know the inspection target item and the category of each item with a small amount of time, effort, and cost, thereby contributing to real inspection.

According to the present invention, it is possible to overcome the limitation of the inspection method of the semiconductor device by the existing TSOM such as the use of the MSD by using the deep learning, and to realize the semiconductor device more realistically and more efficiently in terms of the time and accuracy required for the inspection It is possible to provide a semiconductor device inspection method which can be used for inspection.

According to the present invention, by combining the deep running concept with the inspection of a semiconductor device using the TSOM technology, the semiconductor device can be efficiently tested at a low cost while taking advantage of the advantages of the TSOM.

According to the semiconductor device inspection method of the present invention, in the semiconductor device inspection step, the image to be inspected is expressed by a probability distribution method for all categories of the inspection target items, and the representative value of each category is multiplied by the probability to be included in each category It is possible to minimize the inspection error by determining the numerical determination value for the inspection target item of the inspection target image by using the interpolation method of adding all.

In addition, according to the semiconductor device inspection method of the present invention, it is possible to obtain an image that is focused on an object, an image that is out of focus with a shorter distance from the lens to the imaging surface than the actual focal distance, It is unnecessary to form a TSOM image by using a multi-channel image including an unfocused image having a longer distance to the imaging surface as a verification data set for deep learning and an input image for inspection, And the image acquisition time can be shortened.

1 is a flow chart showing important steps of a semiconductor device inspection method according to an embodiment of the present invention;
2 is a flow chart illustrating one embodiment of a TSOM image acquisition method of the present invention.
FIG. 3 is a conceptual diagram showing a concept of a method of obtaining image data of various different focal positions for an inspection object or a verification data set through an optical system analysis using an FMM, and obtaining a TSOM image therefrom.
FIGS. 4 to 6 are charts showing table form when three categories of line width, height, line width, and height are categorized according to categories, as TSOM images for multiple line patterns as data for deep running.
7 is a conceptual diagram conceptually showing an example of a model for determining a tool for inspection through deep learning;
FIGS. 8 and 9 are diagrams for explaining a method of inputting new test set data into an inspection tool determined through deep learning using a verification data set according to an exemplary embodiment of the present invention, Diagrams showing an example of the results obtained.
10 is a flowchart of a semiconductor device inspection method according to another embodiment of the present invention.
11 is a conceptual diagram of a semiconductor device inspection method according to an embodiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a flowchart showing important steps of a semiconductor device inspection method according to an embodiment of the present invention.

According to this embodiment, first, a TSOM image is obtained for a plurality of semiconductor device parts, e.g., lines having various top widths, heights, and tilt angles (S10).

Since the TSOM image for many objects is required at this time, the TSOM image acquisition method here uses the characteristic TSOM image acquisition method of the present invention as shown in the flow chart of FIG.

Unlike the conventional TSOM image acquisition method of acquiring images at a plurality of different focal positions in an optical system for TSOM for each object to be actually inspected and then combining and processing them in terms of light quantity or color value, For example, three or five focal positions, and obtains the focal position information at which the images and the respective images are obtained (S110)

Subsequently, a plurality of virtual images having different focal positions from the actual images and their focal positions are obtained based on the degrees of deviation of the actual images and the actual images from each other (S120). Finally, And the virtual images are combined and processed in terms of the light quantity and the color value as in the conventional manner, thereby obtaining the TSOM image of the object (S130).

At this time, in step S120 of acquiring a plurality of virtual images and their focal positions, when the actual images of a plurality of different focal positions and the focal positions of the respective real images are obtained, interpolation based on these data is used A plurality of virtual images having different focus positions from the actual image and a focus position thereof may be acquired.

Interpolation is a well-known method in mathematics, science, and engineering because it can reasonably deduce the size of a system of coordinates between two or more known coordinates based on the size or boundary condition of the system. Therefore, Or a weighted average method can be used.

On the other hand, in the step of acquiring a plurality of virtual images and their focal positions, an analysis for a given optical system and a conversion formula (conversion program) for representing the analysis well can be used.

For example, as shown in FIG. 3, a method of analyzing an optical system including illumination, an object to be inspected, and an imaging plane using FMM (Fourier Modal Method) may be used in the first step. In this case, an FMM image (image) detected at an image-capturing plane through which light of a planar light source having a certain wavelength is input and reflected on the surface of the object and reflected may be considered as an output.

Once the FMM image is obtained, it can be used as an input to obtain an output image at a distance d2 from the lens through the action of an optical element called a lens at a distance d1 from the image plane. At this time, it is possible to obtain the output image at the position where the focal distance of the lens is changed or the distance is shortened or the distance is elapsed while changing d2.

It is possible to calculate which image is to be obtained on the imaging plane at a certain distance from the lens with respect to a certain type of inspection object by obtaining an accurate optical analysis of the obtained optical system and an optical element of the FMM image and a conversion formula corresponding thereto, Is compared with the calculated image, it is possible to know whether the optical analysis or conversion formula is valid or not.

For example, first, an optical system is analyzed using the FMM (Fourier Modal Method), and a plurality of virtual images having the same focal position and the actual images are acquired using the setting (characteristics) of the optical system obtained through the optical system analysis, The process of correcting the analysis of the optical system to obtain an appropriate analysis and an acceptable conversion formula therefor, and acquiring a plurality of virtual images having different focus positions from the actual images through the acceptable conversion formula. In the embodiment of the present invention, the optical system is analyzed using the FMM, but the present invention is not limited to this, and the optical system may be analyzed using another type of simulator.

In the step of acquiring a plurality of virtual images and their focal positions, an interpolation method and a conversion formula can be used together. In reality, it is difficult to sufficiently interpret the actual optical system having a plurality of optical elements so as to clearly express the characteristics through about two to three actual images. Therefore, And acquiring an additional virtual image based on the real image and the virtual image by an interpolation method.

On the other hand, in the concrete application of the interpolation method, it is assumed that the focal position is assumed to have a Gaussian distribution centering on the distance at which the virtual image is focused, and when the focal position for the virtual image is selected, A weighted arrangement in which the focal point position is made narrower and the focal point position deviates much from the focal point is arranged. That is, a more reliable TSOM image can be obtained by weighting the focus position around the focal distance to obtain an out-of-focus image, and finally obtaining a TSOM image reflecting the weight.

In the case of using the interpolation method, it is assumed that the virtual images obtained through the above are assumed to have the Gaussian distribution of the focal position around the distance to which the focal point is focused, and when the focal position for the virtual image is selected, A weighted arrangement in which the focal point position is made narrower and the focal point position deviates much from the focal point is arranged. That is, we obtain a more reliable TSOM image by weighting the focal position around the focal length to obtain an out-of-focus image, and finally obtain a TSOM image that reflects the weight.

Then, they are prepared as a verification data set for deep learning (S10). As one of the preparations, each TSOM image, the items to be inspected, and the category category of the items are stored in a database form.

In order to form a validation data set, the target (parameters) such as the height, top width, and tilt angle of the TSOM image and target lines thereof, and the size thereof It is possible to make the TSOM images evenly distributed over each test object and the numerical range (category) in which it is located.

For example, a verification data set is first prepared in order to form a program in a concrete form or an algorithm suitable for inspection through deep learning in a tool for inspection of a semiconductor device. This verification data set is, 5 from the viewpoint of the height of the light emitting diode. Here, the TSOM images constituting the verification data set can be divided into classes of 1 to 5 according to the numerical range of the upper width as shown in FIG. 4, and divided into classes of A to E according to the numerical range of the height as shown in FIG. Also, as shown in FIG. 6, the entire TSOM image can be divided into 25 sub-classes according to numerical ranges in terms of top width and height. Each subclass, for example, has 100 TSOM images, and the whole has 2500 TSOM images. Of course, each TSOM image is identified by its unique number.

In practice, we make line patterns with various line widths, heights, inclination angles, and target numerical categories, check the objects and categories through the actual inspection, obtain TSOM images for them, and prepare verification data sets However, since it is difficult to actually form various reticles and form a semiconductor device pattern having various items and categories with expensive semiconductor equipment, it is difficult to form a few semiconductor device patterns which are already known and classified It is a realistic way to obtain a TSOM image for it and use simulation to synthesize a TSOM image with a numeric category between them to prepare a validation data set.

In order to increase the fitness of the test, the storage list (database) as the data for the deep learning is managed as an open type so that the TSOM image which is the item to be inspected and the category can be continuously supplemented, The determination step of the tool can be operated to newly determine the new inspection tool when the increase in the number of data in the storage list exceeds a predetermined value.

Simulation of multiple TSOM images of different numerical categories is now known to some extent possible via TSOM technology and is based on the fact that it is possible to physically produce an object or categories of TSOM images that have been synthesized inversely, It is possible to synthesize the TSOM image so that there is no significant difference from the synthesized image by finding the corresponding TSOM image, although this is a limited category.

Next, a basic tool is prepared (S20) and input into the basic tool for examining the TSOM image of the primary database data as deep learning data or input. A tool is a combination of a computer and a program executed in the computer. The program can use the same program as the existing shape recognition program while changing the detailed settings. The program has a neural network structure design that enables recognition of features for certain target items through the TSOM image, and an algorithm that enables neural network learning. As the learning progresses, the basic algorithm settings are changed to make concrete algorithms specifically . For example, weights and functions can be activated for learned nodes as learning is done. The concepts related to deep running are already known, so a detailed description thereof will be omitted.

In the present invention, the inspection tool compares characteristics of the input data with respect to the inspection target item. In the deep learning process, we first look at the features or specific patterns of multiple TSOM images through algorithms (software, programs) included in the basic tools, identify the TSOM images by their characteristics, (The database data created in the previous step). If not, the tool will be modified through its own algorithm, and the activity of identifying the TSOM image again with the characteristic or specific pattern will continue to be performed over a certain probability It is determined that the tool for inspection by deep running has been determined when the predetermined criteria are met (S30)

It is desirable to simplify the items to be inspected and reduce the number of categories in order to facilitate the deep running and efficiently in a short time when adapting and specifying the program of the tool for inspection through the deep learning. For example, the tool is firstly divided into five categories for line widths as shown in the table of FIG. 4 to determine the line width inspection tool, and then the line height is divided into five categories as shown in the table of FIG. 5 Thereby determining a height inspection tool. It can be considered that a whole inspection tool is obtained by combining inspection tools for each item.

For example, if you want to find TSOM images that fall within a specific line width category and a specific height category among the TSOM images for the line, for example, by locating and inspecting a tool for inspection, you can find the TSOM images in the specific line width category, Once you find the TSOM images you want to add, you can find the intersection of these TSOM images.

In this case, it is possible to reduce the time required for deep running and increase the fitness. In addition, it is possible to easily supplement the inspection tool by deep-running on the target item added to arbitrarily expand the inspection target item. On the contrary, when two or more inspection items are divided by deep running, the number of target categories becomes 25, and the time required for deep running increases sharply. When the time is limited, the accuracy of the inspection tool is low Can be seen.

Once the tools for inspection are determined through this process, it is possible to check the accuracy and reliability of the inspection tool through additional verification data sets or test sets.

FIG. 7 is a conceptual block diagram conceptually showing a model for determining an inspection tool through such deep learning. In this model, the input data set appears as output data through the hidden layer corresponding to the inspection tool (feedforward direction). If the input data set is all the TSOM images that make up the verification data set, then the output data shows in which category each TSOM image of these TSOM image sets falls. If this output data matches all of the inspection items and categories associated with the corresponding TSOM image according to the previously acquired database, the inspection tool is determined. If they do not match, the current inspection tool again performs its own algorithm or program modification (backpropagation). Then, through the modified algorithm, the above classification is performed, and this process can be repeated.

It is assumed that one generation epoch has elapsed in deep running every time this correction is performed, and an error rate equal to or less than an error rate (the ratio of the number of misclassified data divided by the total number of data) Or when the designer reaches the number of households designated arbitrarily, the inspection tool can be determined as well.

When the line width inspection tool and the line height inspection tool are determined through the above process, the linewidth and the height of the line to be inspected which does not know the numeric category are inspected based on this. As input data for the inspection tool, a TSOM image for an unknown line is prepared first, and a prepared TSOM image is input to a tool for inspection to obtain inspection results (S40). The TSOM image for the unknown line is also obtained based on the TSOM image acquisition method shown in Fig.

The output of the test on the object is indicated by the line in which the target line is included in a certain range of numerical values (category) and the height of the item is included in the numerical range. If the numerical range is a criterion for determining the object defect, the subsequent step may be followed by a step of disposing or reworking the semiconductor device at the process step having the object line in accordance with the criterion.

On the other hand, the results of the test are not a form of judging whether a specific category of a target item simply enters or exits, but rather a probability distribution of probability that falls into several categories in the form of a few percent probability .

The reason for this is that the target item does not show a clear actual shape and size in the TSOM image but rather a somewhat abstracted pattern.

Normally, the representative value of the category with the highest probability of distribution for this probability distribution can be defined as the result value for the test item to be inspected, but the representative value is usually the middle value of the corresponding category. The accuracy is lowered.

To complement this, we can use the fact that the test results have a distribution probability for each numerical category, so that the test results have more accurate numerical values. In other words, as mentioned above, since the target item is represented by a somewhat abstraction pattern rather than a definite actual shape and size in the TSOM image, this type of probability distribution implicating this ambiguity is paradoxically referred to as " And can be used to determine more accurate figures. In other words, it is possible to estimate that the test subject has a certain numerical value in the numerical category at the same time as it enters the numerical category of the target item.

For this, the determination value of the item to be inspected can be derived based on the representative value of each category and the probability of belonging to the category (S45), and interpolation or weighted average can be applied in the derivation process.

FIGS. 8 and 9 show results obtained by inputting data of a test set into a tool determined through deep learning using a verification data set, and deducing the results of the line width and height in the form of a probability distribution belonging to each numerical category of items . In this case, a TSOM image is obtained for a line having a width of 51 nm, a height of 50 nm, and a bottom inclination angle of 89 nm as a test material, inputting the TSOM image into a tool determined in relation to the linewidth, and a probability of satisfying the class 1 corresponding to a representative value of 47.5 in the range of 45 to 50 nm is 24.05% , The probability that it corresponds to the class 2 corresponding to the representative value 53.5 in the range of 51 ~ 56nm is 75.37%, the probability that it corresponds to the class 3 corresponding to the representative value 60 in the range of 57 ~ 63nm is 0.58%, the range of 64 ~ 69nm corresponds to the 66.5 , And the probability that it would be against class 5 corresponding to 72.5 in the range of 70 to 75 nm and 0% in the range of 70 to 75 nm.

In addition, input into the tool determined in relation to the height item has a probability of 0.03% corresponding to the class A corresponding to the 48nm representative value, 54.657% corresponding to the class B corresponding to the 49nm representative value, and 50% The probability that it will be equivalent to 35.31%, the probability that it corresponds to the class D corresponding to the 51nm reference value is 9.99%, and the probability that it corresponds to the class E corresponding to the 52nm reference value is 0%.

For this result, the line width is usually 53.5, the representative value of class 2 is 53.5, and the height is 49 nm. However, in the present invention, the weighted average obtained by multiplying the representative value of each category by the probability belonging thereto is calculated by interpolation. nm and a height of 49.5525 nm. This is a value much closer to the measured value line width 51 nm and height 50 nm of the previously known line pattern. That is, a value much closer to the actual value can be obtained through the interpolation method.

As a result of the measurement obtained in this interpolation method, in comparison with the measurement error 3.1197 and the standard deviation 2.3937 in the case of the conventional MSD method, the measurement error is 3.093 and the standard deviation is 2.0552, As the number of TSOM images constituting the database increases, the time required for the computation in the conventional MSD method increases proportionally to 783.2 milliseconds for the 155 database numbers, (Ms), 5032.2 ms for 1000, 10065.7 ms for 2000, and 25161.3 ms for 5000, but according to the present invention, 21.1 ms for 155 data and 21 But it does not increase proportionally. Even when the number of data is large, it takes very little time to obtain the test result value of the measurement object Could know.

FIG. 10 is a flowchart illustrating an important step of a semiconductor device inspection method according to another embodiment of the present invention, and FIG. 11 is a conceptual diagram of a semiconductor device inspection method according to an embodiment of the present invention.

Referring to FIG. 10, a semiconductor device inspection method according to another embodiment of the present invention is a method for inspecting a semiconductor device including at least one inspection target item (parameter) and a plurality of semiconductor device parts Storing the inspection target item, the category and the image in a storage list (database) as a verification data set for deep learning in an associated state (step S200), combining the computer hardware and software (Step S202), and a deep run is performed on at least one of the inspection target items based on the plurality of images of the storage list so that each of the images Category, and when the classification result is compared with the classification result by the above-mentioned storage list, (Step S204) of determining a tool having a software satisfactorily compliant with the specification as an inspection tool suitable for inspection, and a determination step (Step S206, S208) of obtaining a multi-channel image of the inspection target image and determining which category the inspection target image belongs to by using the inspection tool determined through the deep learning.

In the inspection step, the category of the inspection target item belongs to the category of the inspection target item is not a simple availability determination method but is represented by a probability distribution method for all categories of the inspection target item, and the category is expressed by a range of numerical values S206), and the numerical value determination value for the inspection target item of the inspection target image is determined by adding all of the representative values of each category multiplied by the probabilities to be included in each category (step S208).

The semiconductor device inspection method according to another embodiment of the present invention shown in FIG. 10 is similar to the semiconductor device inspection method according to an embodiment of the present invention shown in FIG.

In the semiconductor device inspection method (300 of FIG. 11) according to an embodiment of the present invention shown in FIG. 1, three actual images are acquired for an object, virtual images are formed based thereon, To form a TSOM image 304, and then the semiconductor device is inspected using the TSOM image as a verification data set for deep running and as an input image for inspection.

However, in the semiconductor device inspection method (302 in FIG. 11) according to another embodiment of the present invention shown in FIG. 10, the TSOM image is not formed, the image 308 focused on the object, A multi-channel image 306 including an unfocused image 310 having a shorter distance from the lens to the imaging plane and an unfocused image 312 having a longer distance from the lens to the imaging plane than the actual focal distance, Are directly used as a verification data set for deep running and as an input image for inspection to inspect the semiconductor device.

In the case of through silicon vias (TSVs) formed in semiconductor devices, due to their structural characteristics, the sizes of the holes vary depending on the depth, and the noise components are larger than those of the FinFET structure due to the influence of illumination. Thus, it is not easy to accurately register multiple images that are out of focus with the focused image in order to form a TSOM image under the influence of each different size of the hall image and noise. Therefore, there is a problem in the accuracy of the inspection according to the image registration.

In order to improve this, in the semiconductor device inspection method (302 of FIG. 11) according to another embodiment of the present invention shown in FIG. 10, a multi-channel image 306 is directly formed as a verification data set And as an input image for inspection. Therefore, there is no need for an image registration process to form a TSOM image, which can improve the inspection accuracy and shorten the image acquisition time.

In the semiconductor device inspection method (302 of FIG. 11) according to another embodiment of the present invention shown in FIG. 10, an image 308 focused on the object and a shorter distance from the lens to the imaging surface Although we have used a total of three multichannel images 306 including an unfocused image 310 and an unfocused image 312 with a longer distance from the lens to the imaging plane relative to the actual focal length, The present invention is not limited to this, and may be applied to a case where an image that is focused on an object, M images whose focal length is shorter than the actual focal distance by a distance from the lens to the imaging surface, (M + N + 1) multichannel images that include N images that are out of focus and have a longer distance from the original image.

In the above, M and N may be arbitrary integers of 1 or more and 4 or less, and M and N may be the same or may not be the same.

While the present invention has been described with reference to exemplary embodiments, it is to be understood that the invention is not limited to the disclosed exemplary embodiments. Accordingly, those skilled in the art will appreciate that various modifications, additions and substitutions are possible, without departing from the scope and spirit of the invention as disclosed in the accompanying claims.

304: TSOM image
306: Multi-channel image
308: Focused video
310, 312: out-of-focus video

Claims (20)

Acquiring actual images of a plurality of different focus positions and an out-of-focus distance (distance) of each actual image through an optical means with respect to the object to be inspected;
Acquiring a plurality of virtual images and their focal positions that differ in focal position from the actual image based on the degree of deviation of the actual images and the actual images from the focal point; And
And obtaining a TSOM image for the object to be inspected using the real image and the virtual image,
In order to obtain a focal position of each of the plurality of virtual images and the plurality of virtual images,
The optical system is analyzed using FMM (Fourier Modal Method)
Acquiring a plurality of virtual images having the same focal position and actual images using the setting (characteristics) of the optical system obtained through the optical system analysis, comparing the virtual images with the actual images,
Based on the comparison, a more suitable analysis of the optical system and a conversion formula (conversion program) for the appropriate analysis are obtained,
And acquiring a focal position of each of the plurality of virtual images and the plurality of virtual images having different focal positions from the actual images through the conversion formula. The method according to claim 1,
In the acquiring of the plurality of virtual images and their focal positions,
A plurality of virtual images obtained by obtaining the actual images with respect to a plurality of different focal positions and using an interpolation method based on the focal position and the information about the actual image, Wherein the step of acquiring the focus position is performed. 3. The method of claim 2,
The virtual image obtained by using the interpolation method assumes that the focal position is in a Gaussian distribution centered on the distance to which the focal point is focused,
Wherein a weighted placement method is employed in which a focus position near the focus position where the focus is focused when the focus position for the virtual image is selected, and a focus position that is less than the focus position when the focus position is out of focus. The method according to claim 1,
The actual images include an image that is in focus with respect to the object to be examined, an image in which the distance from the lens to the image sensing plane is shorter than the actual focal distance, and a distance from the lens to the image sensing plane And the three images including the non-focused images. delete The method according to claim 1,
Wherein the light source for irradiating the object to be inspected for obtaining the actual image is a planar light source having a single wavelength for the optical system analysis through the FMM. Acquiring an image of a plurality of semiconductor device parts that know at least one item to be inspected (parameter) and a category (class) in the item, and performing a deep learning ) In a storage list (database) as a verification data set for the verification data set;
(TOOL) for inspection in the form of combining computer hardware and software and performing a deep run on at least one of the items to be inspected based on the plurality of images in the storage list, When the result of discrimination is compared with the result of classification by the above-mentioned storage list, deep running is performed until the predetermined standard is satisfied, and the state of having the software satisfying the condition Determining a tool as an inspection tool suitable for inspection; And
And an inspection step of obtaining an inspection object image for an unknown semiconductor device part and using the inspection tool determined through the deep learning to find out which category the inspection object image belongs to which inspection item,
An image for a plurality of semiconductor device portions that know the at least one item to be inspected (parameter) and a category (class) in the item and an image to be inspected for the unknown semiconductor device portion are passed through a through-focus scanning optical microscope ) Image,
Wherein the TSOM image for a plurality of semiconductor device parts that know the at least one item to be inspected (parameter) and the category (class) within the item, and the TSOM image to be inspected for the unknown semiconductor device part,
Acquiring actual images of a plurality of different focal positions and an out-of-focus degree (distance) of each actual image through an optical means for an object to be inspected for the plurality of semiconductor device parts or the unknown semiconductor device part;
Acquiring a plurality of virtual images and their focal positions that differ in focal position from the actual image based on the degree of deviation of the actual images and the actual images from the focal point; And
A TSOM image for a plurality of semiconductor device parts that know the at least one inspected item (parameter) and the category (class) within the item using the real image and the virtual image, and a test for the unknown semiconductor device part Obtaining a target TSOM image, wherein the target TSOM image is obtained using a TSOM image acquisition method,
In order to obtain a focal position of each of the plurality of virtual images and the plurality of virtual images,
The optical system is analyzed using FMM (Fourier Modal Method)
Acquiring a plurality of virtual images having the same focal position and actual images using the setting (characteristics) of the optical system obtained through the optical system analysis, comparing the virtual images with the actual images,
Based on the comparison, a more suitable analysis of the optical system and a conversion formula (conversion program) for the appropriate analysis are obtained,
And acquiring a focal position of each of the plurality of virtual images and the plurality of virtual images having different focal positions from the actual images through the conversion formula. 8. The method of claim 7,
In the inspection step, the category of the inspection target item belongs to the category of the inspection target item is not a simple availability determination method but is expressed by a probability distribution method for all categories of the inspection target item,
The categories are expressed as a range of values,
Wherein the numerical determination value for the inspection target item of the inspection target image is determined by adding all of the representative value of each category multiplied by the probability to be included in each category. delete 8. The method of claim 7,
In the acquiring of the plurality of virtual images and their focal positions,
A plurality of virtual images obtained by obtaining the actual images with respect to a plurality of different focal positions and using an interpolation method based on the focal position and the information about the actual image, And a step of acquiring a focus position is performed. 11. The method of claim 10,
The virtual image obtained by using the interpolation method assumes that the focal position is in a Gaussian distribution centered on the distance to which the focal point is focused,
Wherein a weighted placement method is employed in which a focus position near the focus position where the focus is focused when the focus position is selected for the virtual image, and a focus position that is less than the focus position is placed less. 8. The method of claim 7,
Wherein the actual images include an image that is focused on the object to be inspected for the plurality of semiconductor device portions or the unknown semiconductor device portion and a non-focused image having a shorter distance from the lens to the imaging surface than the actual focal distance, Wherein the distance between the lens and the imaging surface is longer than the actual focal distance. delete 8. The method of claim 7,
Wherein the light source for irradiating the object to be inspected for obtaining the actual image is a planar light source having a single wavelength for the optical system analysis through the FMM. 8. The method of claim 7,
Wherein the image for a plurality of semiconductor device parts that know the at least one item to be inspected (parameter) and the category (class) in the item, and the image to be inspected for the unknown semiconductor device part,
Wherein an image focused on a plurality of semiconductor device parts or an unknown semiconductor device part that knows the at least one inspection target item (parameter) and the category (class) in the item is focused and compared with an actual focal distance A multi-channel image including M images whose focal length is shorter than the distance from the lens to the imaging surface, and N images whose focal length is longer than the actual focal distance,
Wherein M and N are arbitrary integers of 1 or more and 4 or less. 16. The method of claim 15,
Wherein M and N are the same. 17. The method of claim 16,
Wherein the M and N are 1. 8. The method of claim 7,
Wherein when deep running is performed on at least one of the inspection target items, a feature of a plurality of images of the verification data set is first searched through an algorithm (software, program) included in the basic tool, The tool is compared with the classification result by the storage list, and if the predetermined criteria are met, the current tool is determined to be the tool for inspection, and if the predetermined criteria are not met, And repeating the process of searching for a new feature until the predetermined criterion is satisfied or repeating until the predetermined number of times is satisfied, and determining the repeated current tool as a tool for inspection. 8. The method of claim 7,
Wherein when determining a tool for inspection of a plurality of items among the items to be inspected in the determining step, deep-learning is applied to one of the plurality of items to the basic tool to determine individual tools, Determine the tool,
Wherein in the inspecting step, the plurality of individual tools are applied to the inspection target image to find out what category the inspection target image belongs to for each item constituting the plurality of items. 8. The method of claim 7,
The item to be inspected may be one of an upper width, a lower width, a depth, a height, and an inclination angle of each of a hole of a semiconductor device, a through silicon via (TSV), a groove, One,
Wherein the category comprises a numerical range within which the items can belong.

Images