WO2021074944A1

WO2021074944A1 - Defect inspection method and defect inspection device

Info

Publication number: WO2021074944A1
Application number: PCT/JP2019/040364
Authority: WO
Inventors: 馨梁; 淳二山本; 広井　高志
Original assignee: 株式会社日立ハイテク
Priority date: 2019-10-15
Filing date: 2019-10-15
Publication date: 2021-04-22

Abstract

This defect inspection device for detecting a defect in a semiconductor wafer having a plurality of dies formed therein is characterized in that: the device comprises an image acquisition subsystem for acquiring a swath image of the wafer and a computer subsystem for processing the acquired swath image and acquiring information about a defect candidate position; and the computer subsystem carries out threshold processing on differences obtained by comparing an inspection image and reference image and identifies a defect, said computer subsystem carrying out processing in which the swath image is divided into frame images which are further divided into a given number of pixels, processing in which the frame images are compared with a reference pattern and amounts of positional deviation are calculated, processing in which corrected position information is calculated for the frame images on the basis of the amounts of positional deviation, processing in which brightness information for the pixels is allocated to finer imaginary blocks on the basis of the position information and one item of brightness information is calculated for each imaginary block, processing in which a reference image is generated on the basis of the brightness information for each block, and processing in which the reference image and an inspection image are compared.

Description

Defect inspection method and defect inspection equipment

The present invention relates to a defect inspection device that inspects foreign substances and defects on a semiconductor wafer.

In recent years, with the increasing integration and capacity of large-scale integrated circuits (LSIs), the circuit line width required for semiconductor elements has become narrower and narrower. For example, with the miniaturization of the LSI pattern size formed on the semiconductor wafer, the defect size that must be detected as a pattern defect is also extremely small.

The patterned wafer defect inspection device is used to detect defects that cause device defects in a patterned wafer on which a circuit pattern of a semiconductor device is formed. Due to the above-mentioned miniaturization, even finer pattern defects and very small defects are directly linked to deterioration of yield, and the inspection device also detects fine defects with higher sensitivity than before. There is a need to. As an inspection method of a patterned wafer inspection device, for example, a wafer held on a stage is irradiated with light or an electron beam by XY scanning, and based on the light reflected or scattered from a sample, or the reflected or secondarily generated electron beam. There is a method of generating an image, dividing the image into predetermined regions such as cell units or die units, comparing the image with a reference image, performing threshold determination on the difference, and detecting defects. As the pattern line width of the semiconductor becomes finer, it is required that the positions of the images match with higher accuracy in the defect detection and in the superimposition of the images at the time of comparison. This is because finer defects can be detected without omission by superimposing and comparing images with high accuracy.

In Patent Document 1, the wafer is irradiated with light from a light source, the light reflected or scattered from the wafer is detected by the detection optical system, and the amount of misalignment between the inspection image captured by the detection optical system and the reference image. Is calculated, and a method of determining a defect by performing threshold processing on the difference image between the inspection image and the reference image corrected based on the amount of misalignment is disclosed. In Patent Document 1, it is written that in the die comparison type optical defect inspection apparatus, a positional shift occurs between the inspection image to be compared and the reference image due to vibration of the stage or the like. If the line width of the wiring pattern to be inspected or the fatal defect size itself becomes finer, even if this misalignment is slight, the image superposition accuracy is lowered and the inspection result is greatly affected. On the other hand, Patent Document 1 discloses that the misalignment information with the adjacent die image is obtained, and the misalignment information is corrected so as to match the design and adjustment conditions of the detection optical system.

Japanese Patent Application Laid-Open No. 2008-268199

However, there is room for further improvement in the method of obtaining the amount of misalignment in Patent Document 1. In fact, as miniaturization progresses, more strict misalignment correction is required.

Patent Document 1 discloses that scattered light from a repeating pattern is blocked by a spatial filter. In a repeating pattern such as a cell area of a memory, the light-shielding effect of the spatial filter is high. However, in the edge or the pattern of the logic area having no repeatability, which is not disclosed in Patent Document 1, there are many edges or patterns that cannot be shielded by the spatial filter, and the effect of the spatial filter is lower than that of the memory cell inspection. Is assumed. In an optical defect inspection device, light that cannot be completely blocked appears bright on the image. For example, in an edge or a pattern region with low repeatability, the detected image appears bright due to the influence of light from the edge or pattern. Then, since the shooting result fluctuates depending on the position of the fine pattern or edge relative to the pixel, it becomes difficult for the image processing unit to distinguish between light and dark, and the image processing unit uses the perturbation difference on the image for comparison. Can be considered. Since the defect size becomes very small, there is a problem that defects may be missed in this perturbation difference. As the pattern becomes finer in this way, the variation of pixel expression due to the change in the relative position of the pattern or edge and the pixel increases, and as a result of using the perturbation difference to correspond to the variation, the image comparison method such as die unit There is a problem that defect detection omission is likely to occur in inspection.

The present invention solves the above-mentioned problems by providing an image acquisition subsystem that acquires a swath image of the wafer and the acquired swath image in a defect inspection device that detects defects in a semiconductor wafer on which a plurality of dies are formed. A computer subsystem that processes and acquires information on the candidate position of the defect is provided, and the computer subsystem performs threshold processing on the difference obtained by comparing the inspection image and the reference image to determine the defect. In the system, the swath image is divided into an arbitrary frame image such as a die unit and further divided into an arbitrary number of pixels, and the frame image is collated with a reference pattern to calculate the amount of misalignment. A process of allocating the brightness information of the pixel to virtual blocks having a finer unit than the pixel based on the amount of misalignment and calculating one brightness information for each virtual block, and a reference image based on the brightness information of each block. Provided is a defect inspection apparatus characterized by executing a process of generating and a process of comparing the reference image with the inspection image.

According to the present invention, in an inspection method for detecting defects by superimposing and comparing an inspection image and a reference image on a die To die, a cell To cell, or the like, it is possible to improve the image alignment accuracy and the defect detection sensitivity. it can. In particular, it is possible to improve the defect detection sensitivity in the logic portion where the repeatability of edges and patterns is low. A high-resolution reference image can be generated, the threshold value for defect determination can be set with a stricter value, and defect determination with higher sensitivity becomes possible.

It is a block diagram of the defect inspection apparatus which concerns on Embodiment 1. FIG. It is a schematic diagram which shows the relationship between a wafer and a die, the relationship between a die and swath, and the relationship between swath and swath channel. It is a flow diagram of defect inspection which divides into sub-pixels and reconstructs a reference image. Die To Die It is a figure explaining the interpolation processing in the inspection die image and the reference image to compare. It is a schematic diagram which shows the concept of the coordinates in a die. It is a figure which showed typically the relationship between the 1st reference pattern and the deviation amount. It is a schematic diagram which shows the relationship between the coordinates in a die and the deviation amount distribution. It is a figure which shows the relationship between a die image, an arbitrary pixel, and an arbitrary sub-pixel. It is a figure explaining the detail of the 1st example of the process of dividing into sub-pixels. It is a figure explaining the detail of the 2nd example of the process of dividing into sub-pixels. It is a figure explaining the detail of the 2nd example of the process of dividing into sub-pixels. It is a conceptual diagram which generates a high-resolution reference image. It is a figure which shows the example which compares an inspection die and a super-resolution reference image.

<Embodiment 1>
FIG. 1 shows a configuration diagram of a defect inspection device 100 according to the first embodiment. The defect inspection device 100 is a device that detects defects (or locations that may be defects) existing in the pattern formed on the wafer 103 and outputs the inspection results. The defect inspection device 100 shown in FIG. 1 is composed of an image acquisition subsystem 101 that acquires an image of a wafer 103, a computer subsystem 102 that processes acquired image data and extracts position information of defect candidate locations, and the like. ..

The image acquisition subsystem 101 uses a light source 105 that irradiates the wafer 103 with light, a detection optical system 106 that detects scattered light or reflected light generated from the wafer by irradiating the light, and light detected by the detection optical system 106. A sensor 107 that converts an electric signal, an AD converter 108 that converts an analog electric signal output by the sensor 107 into a digital signal, and a plurality of signal channels 114 that transmit light detected by the sensor 107 to the AD converter 108. It is composed of a stage 104, a control unit 109, and the like on which the wafer 103 is placed and an arbitrary position on the wafer 103 is moved to the light irradiation position of the light source 105. The control unit 109 controls the operation of each component of the image acquisition subsystem 101 or the overall operation of the defect inspection device 100.

Here, when the defect inspection device 100 is a dark field optical inspection, the light source is a laser, and the detection optical system 106 mainly detects scattered light among the light generated from the wafer. When the defect inspection device 100 is a bright field inspection device, the light source is a broadband light source or the like, and mainly detects reflected light among the light generated from the wafer.

Further, the defect inspection device 100 uses an electron source instead of the light source 105, irradiates the wafer with an electron beam, and detects secondary electrons or backscattered electrons generated from the wafer by a detection system instead of the detection optical system. It may be a type inspection device. As described above, the image acquisition subsystem may be applied to any of a dark field inspection device, a bright field inspection device, and an electron beam inspection device. In this embodiment, a defect inspection device using a dark field inspection device will be described as an example.

The computer subsystem 102 is composed of an image processing unit 110 and a control PC 111. The image processing unit 110 can receive the pixel signal output by the AD converter 108 via the fiber 115. The image processing unit 110 is a unit that uses the output signal of the AD converter 108 to generate a swath image of the wafer 103 and executes information processing for misalignment correction, which will be described later. Here, the "swath image" is an image acquired by irradiating the wafer 103 with the light source 105 while continuously moving the stage 104 on which the wafer 103 is placed in the uniaxial direction, and is elongated in the moving direction of the stage 104. It is rectangular image data. Further, when the term "swath" is simply used, it means an elongated region scanned by stage scanning and illumination from the light source 105. In the case of an electron beam inspection device, the light source should be read as an electron source.

A processor 112 and a storage device 113 are provided to perform the above-mentioned information processing. The storage device 113 is a device that stores various data used by the control PC 111 and the image processing unit 110, and is composed of a large-capacity storage means such as a magnetic disk. The storage device 113 stores a recipe that is data for setting conditions such as a program to be used for various processes, an image acquisition method, and an image method. In order to process a large amount of image data, the image processing unit 110 is often configured by using a parallel computer that operates a plurality of processors in parallel. The control PC 111 functions as a user interface for setting various conditions for information processing of misalignment correction and defect inspection, which will be described later.

FIG. 2 is a schematic diagram showing the relationship between the wafer and the die, the relationship between the die and the swath, and the relationship between the swath and the swath channel. The figure shown in the upper part of FIG. 2 is a schematic view of a wafer in which 37 dies arranged in 7 rows × 7 columns are formed. Each die has coordinates and is specified by rows and columns of the matrix. In the stage scanning in the X direction, the movement from left to right in the figure is defined as the forward direction, and the movement from right to left is defined as the reverse direction.

The stage scan starts from the die at the bottom left corner of the wafer and proceeds in the direction indicated by the reference number 201 to acquire a swath image. If the light irradiation position for imaging has reached the rightmost die and the light irradiation position has not reached the uppermost die of the wafer, the stage is moved by one swath in the Y direction indicated by reference numeral 202. This time, the stage scan is performed in the reverse direction indicated by the reference number 201', and the swath image is acquired. When the light irradiation position reaches the leftmost die, the stage is scanned in the direction of reference number 202', and the stage is scanned in the X direction in the forward direction in the same manner. As described above, in order to acquire the swath images for one row of dies, it is necessary to repeat the X-direction stage scanning in the forward direction and the reverse direction several times to perform imaging. The upper view of FIG. 2 shows the wafer in which the stage scanning is started from the lower left die on the wafer of FIG. 2 and the stage scanning is completed until the middle of the die in the fourth row. Since the line indicating the above is complicated, the dies in the second and subsequent rows from the bottom are represented by one arrow for each row of dies.

As described above, in order to acquire an image for one die, it is not enough to acquire one swath image, but it is necessary to acquire a plurality of swath images. The schematic diagram shown in the middle part of FIG. 2 shows the relationship between the swath image of the die in the fourth row from the bottom of the wafer in the upper part of FIG. 2 and the die. When acquiring an image of a row of dies of dies 204 to 206, a total of six swaths, swaths 208a to 208f, are required. The die 205 corresponds to the wafer center die.

Furthermore, one swath image is divided and detected in a plurality of signal channels. For example, the scattered light corresponding to the swath 208f is detected by the sensor 210, but the output signal is collected for each signal channel and transmitted to the AD converter (108 in FIG. 1) at the subsequent stage. In the following examples, a swath image corresponding to data transmitted by one signal channel (that is, a swath image having a pixel width of 1024 pixels in the Y direction) will be referred to as a “swath channel image”. Further, the acquisition area of one swath channel image will be referred to as "swath channel". Further, an image obtained by cutting out a swath image in die units is referred to as a swath die image, and an image obtained by cutting out a swath channel image in die units is referred to as a swath channel die image. The schematic diagram in the lower part of FIG. 2 shows how the sensor 210 detects scattered light from the swath acquisition region composed of a plurality of swath channels (corresponding image is swath channel image 209) and outputs it as an analog output signal. ing. Since the image data that can be recognized by the image processing unit 110 and the control PC 111 constituting the computer subsystem 102 is digital data after passing through the AD converter 108, the swath channel image or the swath image will be AD-converted in the following description. It is assumed that the pixel data is quantized by the device 108. It should be noted that the meaning of the swath die image is also a die image of the same size as the die on the wafer, which can be combined with a plurality of swath die images.

FIG. 3 shows a flow diagram of defect inspection in which the reference image is reconstructed by dividing it into subpixels. First, S001 is the step of acquiring the swath image described above. In S002, a swath die image is generated from the swath image. For example, there are seven dies in the center row of wafers in FIG. In this case, the image processing unit 110 generates seven swath die images from one swath image. It is assumed that the swath die image is composed of n pixels in the scanning direction (S003).

Then, in S004, the alignment of each die image is performed. The unique pattern in the area to be aligned is registered as a template as a reference pattern. Template matching is performed on each die image, the coordinates of the reference pattern are collated with the coordinates of the reference pattern equivalent pattern in each die image, and the difference amount with respect to the coordinates of the reference pattern is calculated as the displacement amount. The image processing unit 110 stores this misalignment amount (correction amount) in the memory.

In S005, the coordinates including the amount of misalignment obtained in S004 of the pixels constituting the die image are rounded in units of 1 / m subpixels of the pixels. In S006, each pixel having the same coordinates is grouped in subpixel units. The brightness information of the sub-pixels is calculated based on the brightness information of the grouped pixels. The brightness value of the sub-pixel in which the corresponding pixel does not exist is obtained by an estimation calculation from the calculated brightness information of the sub-pixel. Based on this calculation result, the image processing unit 110 constitutes a reference image having an m-fold resolution (S007).

In S008, the image processing unit 110 reconstructs a reference image having the same resolution as the die image having the amount of misalignment that matches the amount of misalignment of the inspection die from the reference image having m times resolution, and the reconstructed reference image. The inspection die image is compared with the die To die. Then, the difference signal or the difference image is subjected to threshold processing, and those having a threshold value or more are determined to be defects. The image processing unit 110 outputs a signal determined to be defective together with the coordinate information thereof as an inspection result via an output unit (not shown).

Here are some examples of image blurring mechanisms in inspections that overlay images, such as die-to-die comparison. First, when the image processing unit 110 divides the swath image into die images, the starting point of the die on the actual wafer and the starting point at which the image processing unit 110 divides the image may be slightly deviated from each other. A swath image or swath channel image is composed of pixels, and each pixel has discrete coordinates in pixel units. On the other hand, the die width is not always an integral multiple of the pixels, and as a result, an error occurs between the line recognized as the start point of the die and the start point of the die on the actual wafer. In addition, the subtle vibration of the stage also causes an error. In order to allow this deviation, there is a method of using a perturbation difference that allows an error when calculating the difference. When the defect inspection device 100 takes a perturbation difference in advance, the defect inspection device 100 stores the allowable deviation amount in advance. Then, the image processing unit 110 performs processing on the assumption that there is no deviation as long as it is within the permissible amount. In the comparison using the perturbation difference, it may lead to omission of detection of defects existing in the region corresponding to the amount of deviation allowed in advance.

As mentioned above, there is a minute error between each die image, and interpolation processing is required when comparing die To dies. FIG. 4 is a diagram illustrating interpolation processing in the inspection die image and the reference image for which the die To die is compared. For example, when comparing the inspection die image 301 with the die To die using the image of the adjacent die as the reference image 302A, the inspection die image 301 and the reference image 302A acquired from the actual adjacent die have a deviation width w1. It is assumed that there is a bright spot as shown on the die. The brightness of each pixel changes depending on the position of imaging, and when one cell in FIG. 4 is regarded as one pixel, it is assumed that the brightness values of the inspection die 301 are 0, 40, 40, and 0 from the left pixel. When the brightness of the adjacent die used as the reference image is viewed in pixel units, the imaging position changes by the deviation width w1, so that the brightness of each pixel is 0, 6, 80, 6, 0 from the left, for example. To do. However, when the reference image is averaged as an interpolation process for superimposing the images at the same position, the reference image 302B is generated. If the amount of deviation is 0, the brightness that should be 0, 0, 40, 40, 0, 0 from the left pixel is 0, 3, 43, 43, 3, 0 from the left pixel as a result of interpolation processing. It becomes the reference image 302B of the brightness. As a result of generating a reference image through interpolation processing such as averaging with respect to the brightness of the adjacent die image actually captured, the resolution of the reference image is lowered and the value may be different from the original brightness. In this way, in order to compare the reference image with reduced resolution and the inspection die image, it is necessary to use a perturbation difference that allows deviation, and as a result, when the resolution on the reference image side decreases, the inspection die image and the reference image die Defects and noise are more difficult to separate on the To die.

In the inspection method for detecting defects by die-to-die comparison, if the defect size to be detected becomes smaller due to miniaturization, when the inspection die image and the reference image are superimposed in the die-to-die comparison, the image is blurred due to this slight deviation on the image. May lead to omission of defect detection.

In order to solve such a problem, a high-resolution reference image is generated and compared with the inspection die image and the die To die. The high-resolution reference image generation will be described in detail below. The process of generating a die image from the swath image and dividing the die image into n pixels is as described in FIG. The swath image is divided into die images, and the die image is further divided into n pixels.

Next, an example of the image misalignment amount calculation process performed by the image processing unit 110 using the reference pattern will be described. In this process, not only the amount of misalignment is obtained, but also the position correction process is finally performed based on the amount of misalignment. Since this is just an example, the alignment process is not limited to this method. First, in the image alignment process, the reference die is scanned in advance to obtain a reference pattern. In the following description, an example of obtaining a reference pattern with the center die will be described. The center die is used because the position error is the smallest near the center of the wafer. However, it may be other than the central die.

Next, a predetermined arithmetic process is performed on each of the swath channel images constituting the swath image to extract the first reference pattern, and further, the coordinates in the die of the first reference pattern are obtained. Here, the in-die coordinates are the internal coordinates for describing the pattern in the die set for each die. FIG. 5 is a schematic diagram showing the concept of coordinates in the die. The dies 204 to 206 shown in FIG. 5 are the same as the schematic diagram of the die row shown in the middle of FIG. 2, and are the views showing only the swath 208b extracted. The solid black line indicated by the reference number 401 is one of the swath channels constituting the swath 208b. In-die coordinates with the lower left corner as the origin (0,0) are set for each die, and the positions of the pixels constituting the swath image or swath channel image can all be expressed using the in-die coordinates.

The first reference pattern used for alignment will be described. The first reference pattern is a pattern for calibrating the misalignment of the swath image acquired by each die on the wafer in units of swath channel images according to the swath image acquired by a specific die. As described above, in this embodiment, this pattern is extracted from the swath image acquired by the wafer center die. Since it is used for misalignment correction, the first reference pattern needs to be a unique pattern (a pattern in which the same pattern does not exist elsewhere on the same swath channel die image). The first reference pattern requires at least one per swath channel die image. Further, in order to improve the accuracy of the misalignment correction, it is desirable to set a plurality of reference patterns (around the swath channel die image). The pattern may be visually confirmed and manually selected, but when the number of reference patterns is large, the manual selection is complicated. Therefore, the defect inspection device 100 may automatically set the reference pattern by finding the edge of the pattern on the two-dimensional pixel array by a pattern edge extraction method using the wavelet transform or the like.

The pattern shape and coordinates of the extracted first reference pattern (coordinates of the cutout image described later) are registered in association with the information of the corresponding swath channel.

As described above, the image processing unit 110 compares the die image obtained by cutting out the swath image in die units with the first reference pattern by template matching. By this comparison, the coordinates in the die of the pattern corresponding to the first reference pattern in the swath channel die image are obtained for each die. Further, the amount of deviation between the X-direction and the Y-direction of the swath channel die image of the predetermined die (ΔX and ΔY, which will be described later) can be obtained from the comparison between the obtained in-die coordinates and the in-die coordinates of the first reference pattern. The obtained deviation amount is stored in the memory in the image processing unit 110. The software for executing template matching is also stored in the storage device 113, and is called by the processor 112 in the image processing unit 110 at the time of execution, and then stored in the memory in the image processing unit 110 and executed. ..

FIG. 6A schematically shows the relationship between the first reference pattern and the amount of deviation. The die 205 is a wafer center die, the die 206 is a die to the right of the die, and the swath 208b is formed so as to straddle these two dies. The swath channel 401 is one of the 1024 pixel swath channels constituting the swath 208b, and the first reference pattern 402 is extracted from the swath channel die image of the wafer center die 205. Since the adjacent die 206 has the same pattern as the wafer center die 205, the swath channel die image of the die 206 has the same pattern 402'as the first reference pattern 402. The pattern 402'exists at a position slightly deviated from the first reference pattern on the image of the swath channel 401 due to the mechanical accuracy of the above. The lower part of FIG. 6A also shows an enlarged view of the first reference pattern 402 and the pattern 402'. In these enlarged views, the first reference pattern is shown as the pattern 502 present in the cropped image 601. Similarly, the pattern 402'is shown as a pattern 502' at a position deviated from the center of the cropped image 602. The pattern shown by the dotted line indicates the first reference pattern 502 registered in the recipe, and the pattern 502'is deviated from the pattern 502 by ΔX and ΔY. That is, if the coordinates in the die of the first reference pattern are expressed as (X, Y), the coordinates in the pattern die in the die 206 of the pattern 402'are expressed as (X + ΔX, Y + ΔY).

Since ΔX and ΔY take different values for each die, ΔX and ΔY take a distribution amount on the swath channel image. FIG. 6B conceptually shows the deviation amount distribution in the X direction with respect to the swath channel 401. The vertical axis of the figure is the deviation amount ΔX in the X direction, the horizontal axis is the coordinates WX in the X direction of the swath channel image, and the coordinates WX_204, WX_205 and WX_206 are the X coordinates of the first reference pattern in the coordinates in the die of each die. These are the coordinates obtained by reprinting the same position into the coordinates of the swath channel image. At the coordinates X_204, that is, the in-die coordinates of the first reference pattern on the die _204, a positional shift of ΔX 204 occurs in the X direction, and at the coordinates X_206, that is, the in-die coordinates of the first reference pattern on the die 206, X There is a misalignment of ΔX _{206 in the direction.} The coordinates X_205 are the coordinates of the first reference pattern on the wafer center die 205, and since they are the positions where the first reference pattern is extracted in the first place, no misalignment occurs. Since the deviation amount (ΔX, ΔY) is a discrete quantity obtained by each die, it is represented by a point on the “X-direction deviation amount distribution ΔX (WX)”. By interpolating the points of the coordinates X_204, the coordinates X_205, and the coordinates X_206, the solid line on FIG. 6B can be obtained, and the deviation amount ΔX with respect to an arbitrary WX can be read.

In this way, the first reference pattern is read out, the first reference pattern is used as a template, template matching is executed on the cut out image, and ΔX and ΔY described in FIG. 6A are obtained. Specifically, the first reference pattern corresponding to the acquired swath channel image is read out, and an image having the same size as the above-mentioned cropped image is cut out from the acquired swath channel image. Then, the cut-out image is subjected to template matching using the first reference pattern as a template, and ΔX and ΔY described in FIG. 6A are obtained. Next, misalignment correction for the swath image is performed. That is, for each swath channel image constituting the swath image, the processor 112 executes a difference calculation between the coordinates in the die of the pixels constituting the swath channel image and the ΔX and ΔY to correct the misalignment. Then, for each swath channel image, the processor 112 executes a difference calculation between the in-die coordinates of the pixels constituting the swath channel image and the ΔX and ΔY to correct the misalignment. For the parts of each die image that do not correspond to the template image, a complementary operation is performed based on the amount of misalignment obtained by comparing the reference pattern and the pattern corresponding to the reference pattern. As a result, the coordinates of the portion that does not match the reference pattern in each die image are also corrected. This is executed on all the die images to be inspected to correct the misalignment. Since the amount of deviation used in the above misalignment correction is also used in the generation of a high-resolution reference image described later, the image processing unit 110 stores this in the memory.

As another example of the alignment process, there is also a method of using design information. For example, a method of comparing the die layout with the die layout on the image acquired by the image processing unit 110 and calculating the amount of misalignment from the difference, or the image processing unit 110 uses an arbitrary reference on the wafer, for example, an arbitrary pattern. A method of obtaining the amount of misalignment by comparing the image acquired by the image acquisition subsystem 101 with the coordinates of the pattern or mark on the design data at the coordinates of the alignment mark or the alignment mark is also conceivable. Even when the design information is not used, even if the image whose coordinates are known in advance is compared with the image of the corresponding portion acquired by the image acquisition subsystem 101 on the inspection wafer, the amount of misalignment can be obtained. good. Various other methods for obtaining the amount of misalignment can be considered, and the method is not limited to the method described here.

FIG. 7 is a diagram showing the relationship between the die image, an arbitrary pixel, and an arbitrary sub-pixel. For example, FIG. 7 shows an example of dividing into m sub-pixels. One swath image includes k dies with

numbers

1, 2, 3 to k, each die contains n pixels, and each pixel is divided into m subpixels. For convenience of explanation, the n pixels of the first die are 11, 12, 13 to 1n, the n pixels of the second die are 21, 22, 23 to 2n, ..., The n pixels of the kth die are k1, Numbers such as k2, k3 to kn are attached. In the divided sub-pixels, the first pixel 11 of the first die is 11_1, 11_2, 11_3 to 11_m, the first pixel 21 of the second die is 21_1, 21_2, 21_3 to 21_m, and the k-th die. The first pixel k1 of the above is numbered such as k1_1, k1_2, k1_3 to k1_m.

8 and 9A and 9B are diagrams for explaining the details of the process of dividing into sub-pixels. First, a method of preparing virtual blocks in advance for the number of m divided into sub-pixels will be described. When dividing into m sub-pixels, the image processing unit 110 virtually prepares m blocks. Since the original pixel is set to 1 and divided into m pieces, one scale is 1 / m, and the increments of 1 / m are the virtual blocks referred to here.

The position information and brightness of each pixel are plotted on this virtual block based on the amount of misalignment calculated in FIGS. 5, 6A and 6B. The position of each pixel of each die is corrected based on the calculated misalignment amount, or the correct position information to be corrected is known. Based on this position information, the pixels at the same position in each die are distributed to m blocks. Hereinafter, the determination as to whether or not they belong to the "same position" is made in units of 1 / m of this block unit.

As the first means of plotting, there is an example in which the amount of misalignment obtained in FIGS. 5, 6A and 6B is grouped in units of virtual blocks for each of the n pixels constituting the die. The image processing unit 110 obtains the correct position information of the pixels of each die based on the calculated position shift amount. As a result, as shown in FIG. 8, it is assumed that some of the

pixels

11, 21 to k1 are located within the range of 0 to 1 / m. Here, the amount of deviation of each pixel is quantized in units of 1 / m. In this case, the pixels are grouped as pixels at the same position. At the time of grouping processing, arithmetic processing is also executed so that the luminance information possessed by each of the pixels becomes one luminance G1-1 in one group by averaging or the like. The brightness G1_1 is plotted on a block having a displacement amount of 0. Similarly for the other pixels of 11, 21 to k1, the image processing unit 110 obtains the correct position information of each pixel based on the amount of misalignment, groups each pixel at the same position in the die, and each pixel group exists. Brightness is distributed to blocks of 1 / m, 2 / m, 3 / m to (m-1) / m according to the position to be processed. For example, when represented by a graph in which the vertical axis is the brightness and the horizontal axis is the position, the scale on the horizontal axis indicating the position is 1 / m, 2 / m, 3 / m to (m-1) / m in block units. Become. As a result of grouping all the pixels and allocating each group to each block, no pixel group may be assigned depending on the block. Blocks that cannot be sorted by any group have no brightness information. In this case, the image processing unit 110 performs interpolation processing. For example, extrapolation interpolation or interpolation interpolation using a spline function is performed so that one luminance information is always plotted in each block. The

pixels

2, 3 to n are also processed in the same manner to obtain subpixel values of G2_1 to Gn_m.

As a second example of plotting means, for each of the n pixels constituting the die, the value of each pixel is directly plotted on each block with respect to the amount of misalignment obtained in FIGS. 5, 6A and 6B. .. The image processing unit 110, which is the same as the plotting means 1, obtains the correct position information of the pixels of each die based on the calculated displacement amount. Then, as shown in FIGS. 9A and 9B, the values of 11, 21, 31 to k1 are directly plotted at the corresponding positions. In that case, each pixel belongs to any block (0, 1 / m, 2 / m, 3 / m to (m-1) / m) according to the position information. As a result of this direct plot, for example, the brightness G1-1'is obtained by averaging all the brightness of the plots (

pixels

11 and 31 in the lower part of FIG. 9A and the upper part of FIG. 9B) included in the section of 0 to 1 / m. Be done. The brightness G1_1'is plotted on a block having a displacement amount of 0. Similarly, for the other blocks, the corresponding brightness can be obtained by averaging all the brightness of the plots included in each section. Similar to the plotting means 1, the luminance information of the block in which the luminance information does not exist is obtained by interpolation. As a result, a super-resolution image of pixel 1 can be obtained. The

pixels

2, 3 to n are also processed in the same manner to obtain subpixel values of G2_1'to Gn_m'.

The image processing unit 110 performs the processing of FIGS. 8 or 9A and B in all the swath images to be inspected. By obtaining G1_1'to Gn_m' for all the pixels of the die, the high-resolution reference image 302C shown in FIG. 10 is created. When the number of virtual blocks with the number of subpixels to be divided is m, the reference image with high resolution becomes a reference image with m times higher resolution than the die image which is not subjected to any processing in the original swath image.

The image processing unit 110 compares the high-resolution reference image created as described above with the inspection die image by die-to-die, executes threshold processing on the difference signal or difference image, and performs defect determination processing. The image processing unit 110 sequentially executes a comparison process of each inspection die image and a high-resolution reference image of the swath image. The image processing unit 110 calculates the amount of misalignment and corrects each die image as described above. That is, the position is adjusted with high accuracy. Based on this, the defect on the die image to be inspected is determined by calculating the difference between the high-resolution reference image and the signal at the same position. At this time, in each pixel, the resolution is different between the high-resolution reference image and the inspection die image. Since comparisons cannot be made with different resolutions, the resolution adjustment process shown in FIG. 11 is performed. First, the averaging process is performed only in the X direction, and the resolution is adjusted. The high-resolution reference image 302C is composed of 1 / m sub-pixels G1_1 to Gn_m. In consideration of the amount of misalignment Δ of the inspection die image, the brightness values of the pixels 1'to the pixels n'that constitute the low-resolution reference image 302D are obtained by the averaging process of the sub-pixels included in each pixel. For example, in FIG. 11, since the brightness value of pixel 1'includes pixels 12 to 21 of 302C, it is obtained by averaging the brightness values of these pixels. Similarly, the brightness values of the pixels 2'to the pixels n'are obtained. Then, the brightness of each pixel of the inspection die image and the reference image 302D is compared. The image processing unit 110 sets a threshold value for the difference obtained by this die-To-die comparison, and determines that a signal higher than the threshold value is a defect and a signal lower than the threshold value is noise. Then, the signal determined to be defective is output together with the coordinates as a detection result via the display unit or the output unit (not shown).

The above is a description of a defect inspection device that reconstructs a reference image by dividing it into subpixels and determines defects by comparing the reference image with an inspection die image and its method. The present invention is not limited to the above embodiment, and includes various modifications. The above-described embodiment has been described in detail in order to explain the present invention in an easy-to-understand manner, and is not necessarily limited to the one including all the described configurations.

Although the die-to-die comparison method has been described in the above embodiment, the present invention can also be applied to the cell-to-cell method and the image comparison method of any unit. In that case, for example, in the cell To cell method, the die is read as a cell. In the above embodiment, the die image has been described, but an image of an arbitrary unit according to the unit of the image to be superimposed on the die image, the cell image, and the inspection image is generically referred to as a “frame image”. The image processing unit 110 can divide the frame image into arbitrary pixels and execute the same processing.

100: Defect inspection device, 101: Image acquisition subsystem, 102: Computer subsystem, 103: Wafer, 104: Stage, 105: Light source, 106: Detection optical system, 107: Sensor, 108: AD converter, 109: Control Unit, 110: Image processing unit, 111: Control PC, 112: Processor, 113: Storage device, 207: Stage scanning direction at the time of acquisition to

Swath

208f, 208a to 208f: Swath image, 209: Swath channel image, 210: Sensor , 301: Inspection die image, 302A to C: Reference die image

Claims

In a defect inspection device that detects defects in a semiconductor wafer on which multiple dies are formed,
An image acquisition subsystem that acquires a swath image of the wafer,
A computer subsystem that processes the acquired swath image and acquires information on the candidate position of the defect is provided.
The computer subsystem is
It is a system that determines defects by performing threshold processing on the difference obtained by comparing the inspection image and the reference image.
The process of dividing the swath image into frame images and further dividing them into an arbitrary number of pixels,
The process of calculating the amount of misalignment by collating with the reference pattern in the frame image,
The process of calculating the corrected position information in the frame image based on the amount of misalignment, and
A process of allocating the luminance information of the pixel to finer virtual blocks based on the positional information and calculating one luminance information for each virtual block.
The process of generating a reference image based on the brightness information for each block, and
A defect inspection apparatus for performing a process of comparing the reference image with the inspection image.
In the defect inspection apparatus according to claim 1,
A defect inspection device characterized in that the frame image is a die image.
In the defect inspection apparatus according to claim 2,
A defect inspection apparatus for generating a reference image having a resolution of m times that of an inspection image, wherein the number of the virtual blocks is m.
In the defect inspection apparatus according to claim 2,
The computer subsystem is
In the process of calculating the amount of misalignment,
A step of acquiring a first swath image of a die formed in a region including the center of the wafer, and
In the first swath image, the pixel signal read from one of the plurality of signal channels (hereinafter referred to as swath channel image) is subjected to a predetermined arithmetic process to perform the first swath image. A step of setting at least one first reference pattern per swath channel die image cut out in die units from the swath channel image constituting the swath image, and a step of setting.
The step of acquiring the second swath image of the wafer and
In the swath channel image constituting the second swath image, the pattern at the same position as the first reference pattern and the coordinates in the die and the first reference pattern are the same included in the second swath image. A step of obtaining the distribution of the amount of misalignment between the first swath image and the second swath image by comparing at least two places on the swath channel image, and
A step of correcting the misalignment of the second swath image with respect to the first swath image using the obtained distribution of the misalignment amount, and
A defect inspection apparatus, characterized in that a step of obtaining a defect candidate position of the wafer is executed by performing a comparative inspection using a swath image in which the misalignment has been corrected.
In the defect inspection apparatus according to claim 1,
The computer subsystem is
A step of obtaining the position information of each pixel based on the amount of misalignment, and
A step of grouping each pixel determined to be the same position in the frame image based on the position information in the virtual block unit.
A defect inspection apparatus comprising executing a step of calculating brightness so that each group has one brightness information.
In the defect inspection apparatus according to claim 1,
The computer subsystem is
A step of obtaining the position information of each pixel based on the amount of misalignment, and
Based on the position information, the step of allocating to the virtual block at the position where each pixel exists, and
A defect inspection device that executes a step of calculating one luminance information by interpolating a block in which none of the pixels is distributed in each virtual block based on the position information and the luminance information of adjacent blocks.
In the defect inspection apparatus according to claim 5 or 6.
The computer subsystem is
In the process of comparing the inspection image with the reference image,
A defect inspection apparatus characterized in that the brightness of each virtual block constituting the reference image is averaged.
In the defect inspection apparatus according to claim 1,
The image acquisition subsystem is
A light source that irradiates the wafer with light,
A detection optical system that detects scattered light generated from the wafer by irradiating the light,
A sensor that converts the light detected by the detection optical system into an electrical signal,
A defect inspection device including a stage on which the wafer is placed and an arbitrary position of the wafer is moved to a light irradiation position of the light source.
For semiconductor wafers held on a stage and having multiple dies formed
A step of shining light from a light source, moving the stage in the XY direction, and acquiring a swath image based on a signal from a sensor that received the light from the wafer.
The step of dividing the swath image into the image of the die unit, and
The step of further dividing the die image into an arbitrary number of pixels,
In the die image, a step of collating with the coordinate information of the reference pattern to obtain the misalignment amount, and a step of collating the frame image with the reference pattern to calculate the misalignment amount.
A step of calculating the corrected position information in the frame image based on the amount of misalignment, and
A step of allocating the luminance information of the pixel to finer virtual blocks based on the positional information and calculating one luminance information for each virtual block.
A step of generating a reference image based on the brightness information for each virtual block, and
An optical defect inspection method including a step of determining a defect by threshold processing with respect to a difference obtained by comparing the reference image and the inspection die image.