JP2005061853A - Surface inspection system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a surface inspection system capable of appropriately classifying a plurality of optimal candidates of system conditions decided based on images of an object (substrate) to be inspected. <P>SOLUTION: The images of the object to be inspected are captured under a plurality of system conditions respectively, and the plurality of optimal candidates of system condition are decided from the plurality of system conditions, based on these images. Then, two or more kinds of feature quantity of the images defined beforehand are extracted from respective images corresponding to the plurality of candidates being obtained (S12), and the two or more kinds of feature quantity are processed statistically, and one or more optimal kinds are selected from the two or more kinds (S14), and the optimal kinds of the feature quantity are processed statistically, thereby classifying the plurality of candidates (S15). <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a surface inspection apparatus for inspecting the surface of a substrate in a manufacturing process of a semiconductor circuit element or a liquid crystal display element.
[0002]
[Prior art]
Conventionally, in a manufacturing process of a semiconductor circuit element or a liquid crystal display element, a defect inspection of a repetitive pattern formed on the surface of a substrate (wafer or plate) has been performed. An automated surface inspection system irradiates the surface of the substrate with illumination light for inspection, captures the image of the substrate based on the diffracted light generated from the repetitive pattern on the substrate, and the contrast difference between the images. To identify the defective part of the repeated pattern.
[0003]
Further, in a conventional surface inspection apparatus, an image of a substrate is captured while automatically changing the apparatus conditions (for example, the tilt angle of the substrate), and optimum apparatus conditions are determined based on the obtained images (for example, (See Patent Document 1). Then, a predetermined inspection is performed under the optimum apparatus conditions.
[Patent Document 1]
JP 2002-162368 A
[0004]
[Problems to be solved by the invention]
However, in the conventional surface inspection apparatus described above, a number of optimum apparatus conditions can be obtained according to the number of types of repetitive patterns on the substrate. When inspection is performed sequentially under these optimum apparatus conditions, the entire process is performed. The time will be longer. In order to shorten the processing time, it is necessary to narrow down a number of optimum apparatus conditions and select one suitable for inspection. Further, as the pre-processing, it is necessary to classify a number of optimum apparatus conditions, but an appropriate classification method has not yet been proposed. In the following description, a number of optimum apparatus conditions determined based on the substrate image as described above are referred to as “optimum apparatus condition candidates”.
[0005]
An object of the present invention is to provide a surface inspection apparatus capable of appropriately classifying a number of optimum apparatus condition candidates determined based on an image of a test object (substrate).
[0006]
[Means for Solving the Problems]
The surface inspection apparatus according to claim 1, wherein the plurality of images are acquired based on an image capturing unit that captures an image of a test object under each of a plurality of apparatus conditions, and the plurality of images captured by the image capturing unit. Determining means for determining a plurality of optimum apparatus condition candidates from among the apparatus conditions, and a plurality of predetermined types of the image for each of the images corresponding to each of the plurality of candidates determined by the determining means Extraction means for extracting the feature quantity, a selection means for statistically processing the plurality of types of feature quantities and selecting one or more optimum types from the plurality of types, and And classifying means for statistically processing feature quantities and classifying the plurality of candidates.
[0007]
According to a second aspect of the present invention, in the surface inspection apparatus according to the first aspect, the selection unit compares the variance value or the standard deviation value of the feature amount in each of the plurality of types, The most appropriate type is selected.
According to a third aspect of the present invention, in the surface inspection apparatus according to the first aspect, the selection unit compares the correlation values of the feature quantities in any two types of the plurality of types by comparing the magnitudes thereof. The optimum type is selected.
[0008]
According to a fourth aspect of the present invention, in the surface inspection apparatus according to the first aspect, the selection means compares the variance value or the standard deviation value of the feature amount in each of the plurality of types, and the plurality of the plurality of types. The optimum type is selected by comparing the correlation values of the feature quantities in any two of the types.
[0009]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
[0010]
This embodiment corresponds to claims 1 to 4.
As shown in FIG. 1, the surface inspection apparatus 10 of the present embodiment includes a holder 12 that supports a wafer 11 that is an object to be inspected, an illumination system 13 that irradiates the surface of the wafer 11 with illumination light L <b> 1, and the wafer 11. The light receiving system 14 that receives the diffracted light L <b> 2 from the surface, the image processing device 15, and the control device 16 are configured.
[0011]
The surface inspection apparatus 10 is an apparatus that automatically inspects the surface of the wafer 11 in the manufacturing process of semiconductor circuit elements. On the surface of the wafer 11, as shown in FIG. 2, a plurality of chips 20 are arranged in the xy direction. Each chip 20 has a plurality of types of repeated patterns. The repetitive pattern is a circuit pattern (resist pattern) having a line array shape that is periodically repeated. The arrangement of the repeated patterns in each chip 20 is the same.
[0012]
Generally, in the surface inspection apparatus 10, before performing inspection of the surface of the wafer 11 (defect pattern defect detection), optimum apparatus conditions (for example, a tilt angle, illumination light amount, and illumination wavelength described later) based on the density information of the wafer image. Etc.) are determined, and the obtained many candidates are classified by clustering processing. Then, in consideration of the classification result, an optimum apparatus condition is set and an inspection is performed. The reason why a large number of candidates are obtained is that a plurality of types of repetitive patterns exist on the wafer 11.
[0013]
Here, each part of the surface inspection apparatus 10 will be described.
The holder 12 places the wafer 11 transported by a transport device (not shown) on the upper surface, and fixes and holds the wafer 11 by, for example, vacuum suction. Further, the holder 12 can be tilted within a predetermined angle range about an axis Ax1 passing through the surface of the wafer 11 by a tilt mechanism (not shown). The tilt angle T of the holder 12 is one of the device conditions that can be arbitrarily set by the control device 16.
[0014]
1 corresponds to the normal line of the holder 12. The axis Ax2 in FIG. Here, a direction parallel to the axis Ax2 (reference normal) in a state where the holder 12 is kept horizontal is defined as a Z direction.
A direction parallel to the axis Ax1 is defined as an X direction. Further, a direction orthogonal to the X direction and the Z direction is defined as a Y direction.
The illumination system 13 is a decentered optical system that includes a light source 21, a light guide 22, and a concave reflecting mirror 23. The light guide 22 transmits light from the light source 21 and emits the light from the end face 22a toward the concave reflecting mirror 23. The end surface 22 a is disposed at the front focal position of the concave reflecting mirror 23.
[0015]
The concave reflecting mirror 23 is a reflecting mirror having a spherical inner surface as a reflecting surface, and is disposed obliquely above the holder 12. That is, an axis (optical axis O1) passing through the center of the concave reflecting mirror 23 and the center of the holder 12 is inclined by a predetermined angle θi with respect to the Z direction. θi is a fixed value.
The concave reflecting mirror 23 is disposed so that the optical axis O1 is orthogonal to the axis Ax1 (X direction) of the holder 12 and the rear focal plane substantially coincides with the surface of the wafer 11. The illumination system 13 of the surface inspection apparatus 10 is an optical system that is telecentric with respect to the wafer 11 side.
[0016]
In the illumination system 13, the light from the light source 21 is applied to the entire surface of the wafer 11 through the light guide 22 and the concave reflecting mirror 23 (illumination light L1). The illumination light L1 is a light beam in which the center line of the light beam reaching an arbitrary point on the wafer 11 is substantially parallel to the optical axis O1. The incident angle (θi−T) of the illumination light L1 corresponds to the angle between the axis Ax2 perpendicular to the surface of the wafer 11 and the optical axis O1, and changes according to the tilt angle T of the holder 12.
[0017]
Further, the wavelength λ and the amount of light of the illumination light L1 change according to the wavelength and the amount of light emitted from the light source 21. For example, when a wavelength selection filter and a neutral density (ND) filter are provided in the light source 21, the wavelength λ and the light amount of the illumination light L1 can be changed by controlling each filter. Similarly to the tilt angle T, the wavelength λ and the light amount of the illumination light L1 are one of the apparatus conditions that can be arbitrarily set by the control device 16.
[0018]
When the illumination light L1 is irradiated in this way, diffracted light L2 is generated from the repetitive pattern formed on the surface of the wafer 11 according to the diffraction conditions described later. The diffraction angle (θr + T) of the diffracted light L2 corresponds to the angle between the axis Ax2 perpendicular to the surface of the wafer 11 and the traveling direction of the diffracted light L2, and changes according to the tilt angle T. θr represents an angle between the traveling direction of the diffracted light L2 and the Z direction. The linear direction of the repeated pattern that generates the diffracted light L2 is substantially parallel to the axis Ax1 of the holder 12.
[0019]
Here, the diffraction condition is expressed by the following equation (1) using the wavelength λ and the incident angle (θi−T) of the illumination light L1, the diffraction angle (θr + T) and the diffraction order n of the diffracted light L2, and the pitch p of the repetitive pattern. ).
sin (θi−T) −sin (θr + T) = nλ / p (1)
In Expression (1), the incident angle (θi−T) and the diffraction angle (θr + T) are positive for the angle direction expected on the incident side with reference to the axis Ax2 perpendicular to the surface of the wafer 11, and are negative for the angle direction expected on the reflection side. And With respect to the diffraction order n, the angle direction expected on the incident side with respect to 0th order diffracted light (regular reflection light) of n = 0 is positive, and the angle direction expected on the reflection side is negative.
[0020]
In Expression (1), θi represents an angle (fixed value) between the Z direction and the optical axis O1, and θr represents an angle between the Z direction and the traveling direction of the diffracted light L2. The tilt angle T is T = 0 when the holder 12 is kept in a horizontal state, the angle direction toward the incident side is plus, and the angle direction toward the opposite side is minus. θi corresponds to the incident angle of the illumination light L1 when the tilt angle T = 0. θr corresponds to the diffraction angle of the diffracted light L2 when the tilt angle T = 0.
[0021]
As can be seen from the equation (1), by changing the tilt angle T, the incident angle (θi−T) of the illumination light L1 can be changed according to the tilt angle T. As a result, the rotation of the diffracted light L2 can be changed. The folding angle (θr + T) can also be changed.
The light receiving system 14 is an optical system that receives the diffracted light L <b> 2, and is an eccentric optical system that includes a concave reflecting mirror 27 and a CCD camera 28.
[0022]
The concave reflecting mirror 27 is a reflecting mirror similar to the concave reflecting mirror 23 described above, and is disposed obliquely above the holder 12. That is, the axis (optical axis O2) passing through the center of the concave reflecting mirror 27 and the center of the holder 12 is arranged so as to be inclined by a predetermined angle θd with respect to the reference normal line (Z direction). θd is a fixed value. Hereinafter, an angle (θd + T) between the axis Ax2 perpendicular to the surface of the wafer 11 and the optical axis O2 is referred to as a light receiving angle. The light receiving angle (θd + T) also changes in accordance with the tilt angle T, similar to the incident angle (θi−T).
[0023]
The CCD camera 28 is arranged so that its imaging surface substantially coincides with the focal plane of the concave reflecting mirror 23. A plurality of pixels are two-dimensionally arranged on the imaging surface of the CCD camera 28.
In the light receiving system 14, the diffracted light L <b> 2 generated from the repetitive pattern on the surface of the wafer 11 is collected via the concave reflecting mirror 23 and reaches the imaging surface of the CCD camera 28. On the imaging surface of the CCD camera 28, an image of the wafer 11 (wafer diffraction image) is formed by the diffracted light L2. The CCD camera 28 captures a wafer diffraction image and outputs an image signal to the image processing device 15.
[0024]
As described above, since the illumination system 13 and the light receiving system 14 are fixed (θi and θd are fixed values), the combination of the tilt angle T of the holder 12 and the wavelength λ of the illumination light L1 is adjusted to produce diffracted light. If the diffraction angle (θr + T) of L2 is set so as to coincide with the light receiving angle (θd + T), the diffracted light L2 generated from the repetitive pattern on the surface of the wafer 11 can travel along the optical axis O2 of the light receiving system 14. it can.
[0025]
When the amount of the diffracted light L2 guided along the optical axis O2 of the light receiving system 14 is an appropriate amount, the CCD camera 28 can capture a wafer diffraction image with good contrast. The light amount of the diffracted light L2 can be set to an appropriate amount by adjusting the light amount of the illumination light L1 applied to the wafer 11.
The apparatus conditions to be adjusted in order to capture a wafer diffraction image with good contrast are the tilt angle T of the holder 12, the wavelength λ of the illumination light L1, and the amount of light. In addition, when the exposure time and electrical sensitivity of the CCD camera 28 can be adjusted, these may be included as device conditions. Further, when the holder 12 can rotate around the axis Ax2, the rotation angle may be included in the apparatus conditions. The controller 16 adjusts the apparatus conditions.
[0026]
Here, the intensity of the diffracted light L <b> 2 is different between a defective portion and a normal portion of the repeated pattern on the surface of the wafer 11. For this reason, in the wafer diffraction image formed on the imaging surface of the CCD camera 28, a light / dark difference (contrast difference) due to a defective portion and a normal portion of the repetitive pattern occurs.
The image processing device 15 converts the image signal of the wafer diffraction image obtained from the CCD camera 28 into a digital image of a predetermined bit (for example, 8 bits) and stores it in the memory as a wafer image. Furthermore, the apparatus conditions at that time (a combination of the tilt angle T of the holder 12 and the wavelength λ of the illumination light L1) are also stored in the memory.
[0027]
In addition, the image processing apparatus 15 cooperates with the control apparatus 16 before performing defect detection of the repetitive pattern formed on the surface of the wafer 11 to obtain optimum apparatus conditions (combination of tilt angle T, wavelength λ, etc.). A large number of candidates are determined (see FIG. 3). Then, the image processing apparatus 15 classifies the obtained many candidates by clustering processing (see FIG. 4). As described above, the reason why a large number of candidates are obtained is that there are a plurality of types of repeated patterns having different pitches, for example, on the wafer 11.
[0028]
Next, a process for determining candidates for optimal apparatus conditions (combination of tilt angle T, wavelength λ, etc.) will be described. This determination process is performed by the image processing device 15 and the control device 16 in accordance with the procedure of the flowchart shown in FIG.
When the wafer 11 is fixed on the holder 12, the control device 16 controls the holder 12, the light source 21, and the like to set the device conditions to, for example, initial values (step S 1). Then, the image processing device 15 captures the device conditions at that time from the control device 16, captures the wafer image from the CCD camera 28 (step S2), and stores the device conditions and the wafer image in association with each other (step S3). ).
[0029]
The processing in steps S1 to S3 is repeatedly performed until processing is completed under each of all the presumed apparatus conditions (combinations such as tilt angle T and wavelength λ) (that is, the determination in step S4 is Yes). . During this period, every time the apparatus condition is changed by the controller 16, the image processing apparatus 15 takes in the wafer image, and the memory stores information about the apparatus condition and the wafer image.
[0030]
When the processing is completed under all the apparatus conditions (Yes in step S4), the image processing apparatus 15 examines the luminance values of all the pixels for each wafer image stored in the memory to obtain the maximum luminance value. (Step S5). Further, the relationship between the apparatus condition associated with each wafer image and the maximum luminance value is examined, and a large number of apparatus conditions whose maximum luminance value has a maximum value are determined as “optimum apparatus condition candidates” (step S6). . The determined many candidates are associated with the wafer image in the memory. In the following description, the determined number of candidates is M.
[0031]
Next, a large number of candidate classification processes will be described. The classification processing is performed by the image processing apparatus 15 according to the procedure of the flowchart shown in FIG.
First, the image processing apparatus 15 obtains a chip average image of the wafer image associated with each candidate (step S11). That is, partial images (hereinafter referred to as “chip images”) corresponding to the plurality of chips 20 shown in FIG. 2 are extracted from each wafer image, and the density of the obtained plurality of chip images is averaged for each pixel. Generate an image.
[0032]
As described above, since the arrangement of the repeated patterns in each chip 20 in one wafer 11 is the same, the density distribution of each chip image is also the same. Therefore, it is considered that the chip average image obtained by the density average of a plurality of chip images shows the characteristics of the wafer image (the entire image of the wafer 11).
FIG. 5 shows an example of a chip average image. Each of the M (9 in FIG. 5) chip average images is generated from a wafer image associated with each of the M candidates. For this reason, the M chip average images are also associated with each candidate in the memory. Comparing the chip average images, it can be seen that the density distribution is slightly different according to the difference in the candidate apparatus conditions.
[0033]
Next, in step S12 of FIG. 4, the image processing apparatus 15 extracts a plurality of predetermined types (for example, 22 types) of feature amounts of the chip average image for each chip average image. Here, an example of the feature amount of the chip average image will be described.
(1) Two attention areas A and B as shown in FIG. 6A are designated in each chip average image, and the average density of each of the attention areas A and B is calculated. Then, the difference between the two average densities (area density difference or area density ratio) is calculated, and the feature value f ₁ , F ₂ And Further, the same calculation is performed for the attention areas A and B as shown in other FIGS. ₃ ~ F ₁₀ And
[0034]
(2) For each chip average image, the average density is calculated to calculate the feature value f. ₁₁ And calculate the standard deviation of the density and calculate the feature value f ₁₂ And In addition, the center of gravity coordinates are calculated using the density as a weight for the entire chip average image, and the feature value f ₁₃ , F ₁₄ And Further, the same calculation is performed on the high density portion (bright portion) of each chip average image (deflected barycentric coordinates), and the feature value f ₁₅ , F ₁₆ And As the high density part, a part having an average density or higher and a part having (average density + density standard deviation) or higher can be considered.
[0035]
(3) By calculating the moment of inertia around the center of the image of each chip average image, the feature value f ₁₇ And calculate the moment of inertia around the X-axis and Y-axis ₁₈ , F ₁₉ And Also, the feature amount f is calculated by calculating the moment of inertia around the deflected image center in the high density portion (bright portion) of each chip average image. ₈₀ And calculate the moment of inertia about the deflected X and Y axes to calculate the feature value f ₂₁ , F ₂₂ And
[0036]
The image processing apparatus 15 has obtained 22 types of feature values f for each of the chip average images (FIG. 5) associated with a large number of candidates by the processing so far. ₁ ~ F ₂₂ 4 is extracted, the process proceeds to step S13 in FIG. ₁ ~ F ₂₂ Is normalized. This is because 22 types of feature values f are used in the next selection process or clustering process. ₁ ~ F ₂₂ Is a process for making it possible to handle with the same scale.
[0037]
The image processing apparatus 15 then calculates the feature value f after normalization. ₁ ~ F ₂₂ Based on the above, the selection process of step S14 is performed. In the following description, the normalized feature quantity is simply referred to as “feature quantity”.
In step S14, the image processing apparatus 15 performs 22 types of feature values f. ₁ ~ F ₂₂ Are statistically processed (details will be described next), and, for example, five “optimal types” are selected from the 22 types. This selection process is a pre-process for improving the performance of the next clustering process, and is performed, for example, according to the following procedures [1], [2], [3],.
[0038]
[1] Feature amount f in each of 22 types (all types) ₁ ~ F ₂₂ The standard deviation value σ of is calculated. The standard deviation value σ is the feature value f ₁ ~ F ₂₂ It is a statistical value about the dispersion | variation in. An example of the calculation result is shown in FIG. “Ave.”, “Devi.”, “Cent. X”, “Cent. Y”, “C.X-up”, “C. Y-up”, “I-mmt.”, And “I-mmtX.” In FIG. , Feature amount f described above ₁₁ ~ F ₁₈ Represents the type. Each chip average image is associated with each of the M candidates. The numerical value of FIG. ₁ ~ F ₂₂ Is the value of
[0039]
[2] Arrange 22 types (all types) in descending order of standard deviation value σ (ranking). For example, in the calculation example of FIG. 7, “C.Y-up” → “Cent.Y” → “C.X-up” → “Cent.X” → “Ave.” → “I-mmt.” → “ Dev. "→" I-mmtX. " According to a statistical concept, a type with a small standard deviation value σ (for example, I-mmtX.) Has a characteristic amount (for example, a characteristic amount f of each of the chip average images No. 1, 2, 3,...). ₁₈ ) Has a similar value centered on the average value. For this reason, the amount of information is small. On the other hand, the type with a large standard deviation value σ (for example, CY-up) has a large variation in the feature amount and includes a large amount of information. Therefore, it is preferable to use a type having a large standard deviation value σ as an “optimal type” for the clustering process described later.
[0040]
[3] First, the type “CY-up” having the largest standard deviation value σ is selected as one of the “optimal types”.
[4] Next, attention is paid to the type “Cent. Y” having the second largest standard deviation value σ as a candidate of the “optimal type”, and the feature amount f of this type “Cent. Y”. ₁₄ And the feature value f of the type “C.Y-up” selected in [3] above ₁₆ Correlation value C _A Calculate And this correlation value C _A Is a predetermined threshold C _S (For example, 0.6), and the correlation value C _A Is threshold C _S In the following cases, the type “Cent. Y” that is currently focused on is also selected as one of the “optimal types”. Conversely, the correlation value C _A Is threshold C _S If it is larger, the focused type “Cent. Y” is excluded from the “optimal type”.
[0041]
According to a statistical concept, the feature quantity f of types “C.Y-up” and “Cent.Y” ₁₆ , F ₁₄ Inter-correlation value C _A When is high, these two types are similar feature amounts. For this reason, even if both types are used for the clustering process described later, the information is only redundant and useless. In contrast, the correlation value C _A When the value is low, the above two types have different tendencies, so it is useful for the clustering process to use both.
[0042]
Therefore, in the procedure [4], the correlation value C _A Is threshold C _S Only in the following cases, the focused type “Cent. Y” is selected as one of the “optimal types”. As a result, two types “C.Y-up” and “Cent.Y” are selected as the “optimal types”. When the noticed type “Cent.Y” is excluded from the “optimal type”, the “optimal type” at the current stage is only the type “C.Y-up” selected in the procedure [3].
[0043]
Here, the feature value f in any two of the 22 types (all types) ₁ ~ F ₂₂ FIG. 8 shows an example of the result (correlation coefficient matrix) obtained by calculating the correlation values between each other. “Ave.”, “Devi.”, “Cent. X”, “Cent. Y”, “C.X-up”, “C. Y-up”, and “I-mmt.” In FIG. Further, “Mmt-up”, “MmtX-up”, “MmtY-up”, “P1”, “P2”,... In FIG. ₈₀ , F ₂₁ , F ₂₂ , F ₁ , F ₂ Indicates the type of. The numerical values in FIG. 8 are correlation values.
[0044]
In the calculation example of FIG. 8, the feature amount f of the types “C.Y-up” and “Cent.Y” in the procedure [4]. ₁₆ , F ₁₄ Inter-correlation value C _A Is 0.483 and the threshold value C _S It is as follows. Therefore, in the procedure [4], the candidate type “Cent. Y” of interest is selected as one of the “optimal types”.
[0045]
[5] Next, attention is paid to the type “C.X-up” having the third largest standard deviation value σ as a candidate of the “optimal type”, and the feature amount f of this type “C.X-up”. ₁₅ And the same correlation calculation and comparison processing with the “optimum type” feature quantity already selected.
When the “optimum type” that has already been selected is only one type “C-Y-up”, the correlation calculation and comparison processing in the procedure [5] is the same as the above [4]. That is, the feature amount f of the noticed type “C.X-up” ₁₅ And “optimum type“ C-Y-up ”” feature amount f ₁₆ Correlation value C _B And the correlation value C _B Is threshold C _S Only in the following cases, the candidate type “C.X-up” to which attention is paid is selected as one of the “optimal types”.
[0046]
When the “optimum type” already selected is the two types “C.Y-up” and “Cent.Y”, the correlation calculation and the comparison process in the procedure [5] are as follows. That is, the feature amount f of the noticed type “C.X-up” ₁₅ And “optimum type“ C-Y-up ”” feature amount f ₁₆ Correlation value C _C , And the feature amount f of the noted type “C.X-up” ₁₅ And “optimum type“ Cent.Y ”” feature quantity f ₁₄ Correlation value C _D Calculate And two correlation values C _C , C _D Each with threshold C _S The two correlation values C _C , C _D Are both threshold C _S Only in the following cases, the candidate type “C.X-up” to which attention is paid is selected as one of the “optimal types”.
[0047]
In the calculation example of FIG. 8, the feature quantity f of the types “C.X-up” and “C.Y-up” in the procedure [5]. ₁₅ , F ₁₆ Inter-correlation value C _C Is 0.026, and the feature amount f of types “C.X-up” and “Cent.Y” ₁₅ , F ₁₄ Inter-correlation value C _D Is 0.246, both of which are threshold values C _S It is as follows. For this reason, in the procedure [5], the candidate type “C.X-up” of interest is selected as one of the “optimum types”.
[0048]
[6] Similarly, attention is paid to “optimum type” candidates in descending order of the standard deviation value σ, and “correlation between the feature type of the noticed type and the feature quantity of the selected“ optimal type ”. All values are threshold C _S Only in the following cases, the process of selecting the focused candidate type as one of “optimal types” is repeated. Then, when the number of selections of “optimum type” reaches 5, for example, the selection process in step S14 of FIG.
[0049]
As a result of this selection process, for example, five “optimum types” are selected from the 22 types. The five “optimum types” mean that each standard deviation value σ is large and the correlation value between the feature quantities is a threshold value C. _S It is as follows. That is, as a result of the selection process in step S14, five types that include a large amount of information and have different tendencies from each other are selected as “optimal types”.
[0050]
Next, in step S15, the image processing apparatus 15 statistically processes, for example, five “optimum types” feature amounts. That is, the clustering process is performed based on the “optimal type” feature amount. Then, a large number of candidates are classified.
Specifically, a five-dimensional Euclidean space (feature space) with five “optimal types” as axes is set, and a similarity measure is introduced into this feature space, and a large number of candidates are selected. Classify similar groups (ie clusters). In addition, if each axis of the feature amount space is normalized, it can be treated equivalently in terms of distance.
[0051]
The clustering process will be briefly described. First, the image processing apparatus 15 sets all of the M candidates as clusters. At this point, the number of clusters is M. The number of elements included in each cluster is one. Next, the distance between clusters (that is, the Euclidean distance between elements) is calculated for all combinations.
Then, a pair of clusters having the smallest distance between the clusters is selected and merged. By this merging, the pair of clusters becomes one cluster. This will include two elements. The number of clusters at this time is (M-1). Next, the distance between clusters is calculated similarly. However, when a plurality of elements are included in one cluster, the distance between the clusters is obtained by an average between groups.
[0052]
When the calculation of the distance is completed, the image processing apparatus 15 performs the merging process of the pair of clusters having the minimum distance as described above. Further, the calculation of the distance between the clusters and the merging process are repeated, and when the number of clusters reaches a predetermined number (for example, 5), the classification process of many candidates (FIG. 4) is terminated. Note that the final number of clusters can be changed according to the number of candidates subjected to classification processing.
[0053]
FIG. 9 shows an example of candidates classified into five clusters and chip average images associated therewith. As shown in FIG. 9, the M chip average images (FIG. 5) are classified into five clusters, and each cluster includes a chip average image with high similarity. Therefore, candidate apparatus conditions corresponding to the respective chip average images are also classified into the same cluster having high similarity.
[0054]
According to this embodiment, 22 types of feature quantities (f) are obtained from each chip average image associated with a large number of candidates. ₁ ~ F ₂₂ ) And the clustering process is performed based on, for example, five “optimum types” feature quantities selected from the 22 types. 5) clusters. That is, a large number of optimum apparatus condition candidates can be appropriately classified.
[0055]
In this embodiment, since the number of selections of “optimal types” of feature quantities is set to five, for example, the feature quantity space in the clustering process corresponds to the number of samples (number of candidates, M, for example, several tens). An appropriate number of dimensions can be set. As a result, the performance of the clustering process is improved.
Incidentally, if the number of “optimal types” selected (that is, the number of dimensions of the feature space) is small, space separation in the clustering process becomes insufficient. On the other hand, when the number of “optimal types” selected (that is, the number of dimensions) increases, the number of necessary samples increases exponentially, and the actual number of samples (several tens) becomes relatively short. For this reason, the performance of the clustering process can be improved by setting the number of “optimum types” to five, for example.
[0056]
After appropriately classifying a number of optimum apparatus condition candidates by clustering processing, the image processing apparatus 15 can select representative candidate apparatus conditions from each cluster and determine them as optimum conditions. In this case, the number of optimum conditions determined is the same as the number of clusters (for example, 5), which is smaller than the total number of candidates (M). Therefore, even if the wafer 11 is inspected once under each optimum condition, the entire processing time can be shortened as compared with the case where all candidates are set to the optimum condition and inspected once. Can do.
[0057]
Furthermore, it is possible to further select representative candidates selected from each cluster after classification, and select the most suitable candidate from among them as the optimum condition. In such a case, since the number of matrixes to be selected can be reduced, the optimum condition can be selected more easily.
Further, according to the present embodiment, the feature amount can be extracted efficiently by using the chip average image. Therefore, a large number of candidates can be classified accurately and easily.
[0058]
(Modification)
In the above-described embodiment, the standard deviation value σ is compared in size in the process of selecting the “optimal type” of the feature amount. Instead of the standard deviation value σ, the variance value σ ² May be compared. Variance σ ² Is also the feature amount f ₁ ~ F ₂₂ It is a statistical value about the dispersion | variation in.
Furthermore, in the above-described embodiment, the correlation value magnitude comparison is performed after the standard deviation value σ magnitude comparison in the “optimal type” selection process of the feature quantity, but the order may be reversed. That is, the standard deviation value σ may be compared after the correlation value is compared.
[0059]
As a selection process in this case, in the correlation coefficient matrix of FIG. _S It is conceivable to search for the above pair (a combination of two types), select one of them as the “optimal type”, and exclude the other. At this time, it is preferable to select a type having a higher standard deviation value σ and exclude a lower type. Further, after this selection, ranking may be performed based on the standard deviation value σ, and a predetermined number (for example, five) having a large standard deviation value σ may be finally selected.
[0060]
In the above-described embodiment, the standard deviation value σ (or the variance value σ ² ) And the correlation value, but the standard deviation value σ (or variance value σ ² ) May be used, or only the correlation value may be used.
Furthermore, in the above-described embodiment, the optimum condition candidate is determined based on the maximum luminance value of each wafer image (steps S5 and S6 in FIG. 3), but the present invention is not limited to this. Optimal condition candidates may be determined based on the average luminance value of each wafer image and other factors (luminance information).
[0061]
Further, in this embodiment, when a large number of candidates are classified, a chip average image is obtained from the wafer image and a feature amount is extracted from the chip average image. However, the present invention is not limited to this. (1) A feature amount may be extracted from a wafer image. This is effective when inspecting defects caused by unevenness of the thickness of a resist film applied to the surface of a substrate such as the wafer 11 or distortion of the substrate. (2) A shot average image may be obtained from the wafer 1 image, and a feature amount may be extracted from the shot average image.
This is effective even when the shot area is smaller than one chip 20.
[0062]
Furthermore, in this embodiment, when performing the clustering process, an example in which the Euclidean space is used as the feature amount space is shown, but other distances (for example, Mahalanobis distance) and spaces may be used.
Further, in this embodiment, an example in which the inspection is performed using the diffracted light L2 from the surface of the wafer 11 has been shown, but the inspection may be performed using the scattered light from the surface of the wafer 11, or the wafer. The inspection may be performed using both the diffracted light L2 from the surface 11 and the scattered light.
[0063]
Furthermore, the present invention may be applied to a surface inspection apparatus having a configuration different from that of the surface inspection apparatus 10 of the present embodiment. For example, the present invention may be applied to a surface inspection apparatus in which the illumination system and the light receiving system are movable with respect to a fixed holder, and the same effect as the tilt of the holder can be obtained by moving the illumination system and the light receiving system.
In this embodiment, a large number of candidate determination processes and classification processes are performed by the image processing apparatus 15 of the surface inspection apparatus 10, but the same applies even when an external computer connected to the surface inspection apparatus 10 is used. The effect of can be obtained.
[0064]
【The invention's effect】
As described above, according to the present invention, it is possible to appropriately classify a large number of optimum apparatus condition candidates determined based on an image of a test object (substrate).
[Brief description of the drawings]
FIG. 1 is a diagram showing an overall configuration of a surface inspection apparatus 10;
FIG. 2 is a schematic view showing the appearance of the surface of a wafer 11. FIG.
FIG. 3 is a flowchart showing a process for determining an optimum apparatus condition candidate.
FIG. 4 is a flowchart showing classification processing of a large number of candidates.
FIG. 5 is a diagram illustrating an example of a chip average image.
FIG. 6 is a diagram illustrating an example of attention areas A and B of a chip average image.
FIG. 7 is a diagram illustrating an example of a plurality of types of feature amounts extracted from each chip average image and a standard deviation value σ calculated for each type.
FIG. 8 is a diagram illustrating an example of a correlation coefficient matrix.
FIG. 9 is a diagram illustrating an example of a classification result by clustering.
[Explanation of symbols]
10 Surface inspection equipment
11 Wafer
12 Holder
13 Lighting system
14 Light receiving system
15 Image processing device
16 Control device
20 chips
21 Light source
22 Light Guide
23, 27 Concave reflector
28 CCD camera

Claims (4)

Image capturing means for capturing an image of a test object under each of a plurality of apparatus conditions;
Determining means for determining a plurality of optimum apparatus condition candidates from the plurality of apparatus conditions based on the plurality of images captured by the image capturing means;
Extraction means for extracting a plurality of predetermined types of feature amounts of the image for each of the images corresponding to each of the plurality of candidates determined by the determination means;
A selection means for statistically processing the plurality of types of feature values and selecting one or more optimum types from the plurality of types;
A surface inspection apparatus comprising: a classifying unit that statistically processes the optimum type of feature quantity and classifies the plurality of candidates. The surface inspection apparatus according to claim 1,
The surface inspection apparatus, wherein the selection unit selects the optimum type by comparing the variance value or standard deviation value of the feature amount in each of the plurality of types. The surface inspection apparatus according to claim 1,
The surface inspection apparatus, wherein the selection unit selects the optimum type by comparing the correlation values of the feature quantities of any two types among the plurality of types. The surface inspection apparatus according to claim 1,
The selection means compares the variance value or standard deviation value of the feature amount in each of the plurality of types, and compares the correlation value between the feature amounts in any two types of the plurality of types. Thus, the surface inspection apparatus is characterized in that the optimum type is selected.

Images