Feature calculation device, identity identification system, feature calculation method, and feature calculation program

Abstract

This invention provides a feature calculation device, an identity identification system, a feature calculation method, and a feature calculation program that can identify whether or not items are identical with high accuracy. [Solution] In the identity identification system, the reference product feature quantity calculation device 11-1 and the judgment target product feature quantity calculation device 11-2, which calculate feature quantities for identifying identity of a semiconductor device that is the target product, each comprises an image data acquisition unit 15-1, 15-2 that acquires image data showing the identification layer 4 of the target product 1, and a feature quantity calculation unit 16-1, 16-2 that calculates the feature quantities of the target product based on the image data. The identification layer contains dispersed pattern-forming particles made of an inorganic material with a refractive index n of 3 or more, the feature quantities are feature quantities related to the pattern of the pattern-forming particles, and the image data is data represented in the HSV color space.

Description

Technical Field

[0001] The present invention relates to a feature amount calculation device, an identity identification system, a feature amount calculation method, and a feature amount calculation program.

Background Art

[0002] There is a need for a technology to identify the identity of an article. For example, in the field of semiconductor devices, the distribution of counterfeit products has become a problem. Therefore, a technology for identifying whether a distributed product is the same as an authentic product is required. As a technology for identifying identity, a technology that extracts feature amounts of a target article by image analysis and performs identification using the extracted feature amounts is known.

[0003] As a technology related to the above, the technology described in Patent Document 1 (Japanese Patent Application Laid-Open No. 2013-97658) can be cited. Patent Document 1 discloses "an individual identification device for identifying an article in which tags are randomly arranged on a substrate, the device comprising: feature point extraction means for extracting feature points from an image obtained by photographing the substrate surface to which the tags are attached; feature amount calculation point specifying means for specifying one feature amount calculation point at a time from among the plurality of feature points extracted by the feature point extraction means, reference point specifying means for specifying a reference point from among the feature points excluding the feature amount calculation point, reference point specifying means for specifying one or more reference points from among the feature points located at a distance of a predetermined distance or more from the feature amount calculation point, and feature amount calculation means for calculating a feature amount for each feature amount calculation point using the three points of the feature amount calculation point, the reference point, and the reference point; identification means for comparing the reference feature amount data, which is the feature amount of a reference article calculated by the feature amount calculation means, with the target feature amount data, which is the feature amount of a target article calculated by the feature amount calculation means, to identify whether the reference article and the target article are the same individual." According to the description of Patent Document 1, even when the position of the feature point specified as the reference point is shifted, the influence on the feature amount can be reduced by the individual identification device.

Prior Art Documents

Patent Documents

[0004] [Patent Document 1] Japanese Patent Publication No. 2013-97658 [Overview of the project] [Problems that the invention aims to solve]

[0005] When identifying articles, higher identification accuracy is desirable. Therefore, the object of the present invention is to provide a technology that can identify whether or not articles are identical with high accuracy. [Means for solving the problem]

[0006] In one embodiment, the present invention relates to a feature quantity calculation device. This feature quantity calculation device is a device for calculating feature quantities for identifying a target product. This feature quantity calculation device comprises an image data acquisition unit that acquires image data showing the identification layer of the target product, and a feature quantity calculation unit that calculates feature quantities of the target product based on the image data. The identification layer contains dispersed pattern-forming particles made of an inorganic material with a refractive index n of 3 or more. The feature quantities are feature quantities relating to the pattern of the pattern-forming particles. The image data is data represented in the HSV color space.

[0007] In one embodiment, the present invention relates to an identity identification system. This identity identification system comprises the feature quantity calculation device and the identity determination device. The feature quantity calculation device calculates the feature quantities of a reference product as the target product. The identity determination device compares the calculated feature quantities of the reference product with the feature quantities of the product to be determined and determines whether or not the product to be determined is identical to the reference product.

[0008] In one embodiment, the present invention relates to a reference product feature quantity calculation device. This reference product feature quantity calculation device is a feature quantity calculation device described above, and calculates the feature quantity of a reference product for comparison with the feature quantity of a product to be judged.

[0009] In another embodiment, the present invention relates to an identity identification system. This identity identification system comprises the feature quantity calculation device described above, a reference product data storage device for storing reference product data, and a determination device. The feature quantity calculation device calculates the feature quantities of the product to be determined, which is the target product. The determination device determines whether the product to be determined is identical to the reference product based on the reference product data and the feature quantities of the product to be determined.

[0010] In one embodiment, the present invention relates to a determination device. This determination device determines whether or not an item to be determined is identical to a standard item based on standard item data stored in a standard item data storage device and the characteristic quantities of the item to be determined. The characteristic quantities of the item to be determined are calculated by the characteristic quantity calculation device described above.

[0011] In one embodiment, the present invention relates to a feature calculation method. This feature calculation method is a method for calculating feature quantities for identifying the identity of a target product. This feature calculation method comprises an image data acquisition step of acquiring image data showing the identification layer of a target product, and a feature calculation step of calculating feature quantities for identifying the identity of a target product based on the image data. The identification layer contains dispersed pattern-forming particles made of an inorganic material with a refractive index n of 3 or more. The feature quantities are feature quantities relating to the pattern of the pattern-forming particles. The image data is data represented in the HSV color space.

[0012] In one embodiment, the present invention relates to a feature calculation program. This feature calculation program is a program that causes a computer to execute a feature calculation method for calculating feature quantities for identifying the identity of a target product. The feature calculation method comprises an image data acquisition step of acquiring image data showing the identification layer of a target product, and a feature calculation step of calculating feature quantities for identifying the identity of a target product based on the image data. The identification layer contains dispersed pattern-forming particles made of an inorganic material with a refractive index n of 3 or more. The feature quantities are feature quantities relating to the pattern of the pattern-forming particles. The image data is data represented in the HSV color space. [Effects of the Invention]

[0013] According to the present invention, a technique is provided that can identify with high accuracy whether articles are the same or not.

Brief Description of the Drawings

[0014] [Figure 1] FIG. 1 is a block diagram showing the overall configuration of the identity identification system according to the embodiment. [Figure 2] FIG. 2 is a plan view showing an example of the identification layer. [Figure 3] FIG. 3 is a flowchart showing an identity identification method using the identity identification system according to the embodiment. [Figure 4] FIG. 4 is a flowchart showing a preferred embodiment of the method for calculating feature amounts. [Figure 5] FIG. 5 is a conceptual diagram showing a preferred example of the color separation process. [Figure 6] FIG. 6 is a conceptual diagram showing an example of the method for calculating feature amounts. [Figure 7] FIG. 7 is a schematic diagram showing a modified example of the identity identification system. [Figure 8] FIG. 8 is a schematic cross-sectional view showing an example of the target article. [Figure 9] FIG. 9 is a graph showing the measurement results of the particle size distribution of the Si nanofiller used in the examples.

Embodiments for Carrying Out the Invention

[0015] Hereinafter, the identity identification system according to the embodiment of the present invention will be described.

[0016] (1) Outline FIG. 1 is a block diagram showing the overall configuration of the identity identification system 10 according to the present embodiment.

[0017] The identity identification system 10 according to this embodiment is used to identify whether the reference product 1-1 and the determination target product 1-2 are the same. The reference product 1-1 is, for example, a product at the time of manufacture and is a genuine product. On the other hand, the determination target product 1-2 is, for example, a product after distribution. The identity identification system 10 according to this embodiment is used, for example, by the purchaser of the determination target product 1-2 to confirm whether the purchased determination target product 1-2 is genuine. In this specification, the reference product 1-1 and the determination target product 1-2 are collectively referred to as "target product 1".

[0018] An identification layer 4 is provided on the target product 1. The identification layer 4 is used to identify the target product 1 and has a unique pattern.

[0019] FIG. 2 is a plan view showing an example of the identification layer 4. Pattern forming particles 5 are randomly dispersed in the identification layer 4. Here, the pattern forming particles 5 are formed of an inorganic material having a refractive index n of 3 or more.

[0020] The pattern forming particles 5 formed of an inorganic material having a refractive index n of 3 or more generate scattered light based on a principle called Mie resonance. Mie resonance means that when light with a wavelength λ (nm) is incident on a substance (refractive index n), the effective wavelength in the substance becomes λ / n (nm), and a standing wave is formed when the effective wavelength of the light λ / n (nm) is equal to the diameter of the particle, and the lowest-order Mie resonance, that is, the electrical and magnetic dipole resonances appear in the optical region (see, for example, Japanese Patent No. 7277923). That is, particles using an inorganic material having a refractive index n of 3 or more selectively and strongly scatter light of a specific wavelength due to Mie resonance. That is, even such particles of the same type emit colors different depending on the particle size. The identification layer 4 contains pattern forming particles 5 having different particle sizes so as to emit colors different from each other. In this embodiment, it is preferable that the pattern forming particles 5 contained in the identification layer 4 have a particle size distribution such that they emit colors in various colors.

[0021] As described above, pattern-forming particles 5 that emit different colors are randomly dispersed in the identification layer 4. The pattern (position and color) of these pattern-forming particles 5 is information unique to the target product 1. Therefore, the identity identification system 10 according to this embodiment is configured to identify whether or not the reference product 1-1 and the product to be judged 1-2 are identical, using the pattern of the pattern-forming particles 5.

[0022] Here, the inventors have found that when analyzing the image of the identification layer 4 containing the specific pattern-forming particles 5 described above, identity can be accurately identified by calculating feature quantities using image data represented in the HSV color space (hereinafter sometimes referred to as HSV data). Image data obtained by a typical digital imaging device is in RGB format. According to the inventors' findings, by calculating feature quantities from HSV data rather than from typical RGB format image data, the hue is represented in 360°, making it easier to intuitively grasp colors such as "red," "yellow," and "green," and thus enabling accurate identity identification.

[0023] In other words, the identity identification system 10 according to this embodiment is configured to identify identity by calculating feature quantities from HSV data for each of the reference product 1-1 and the product to be judged 1-2, and comparing the calculated feature quantities.

[0024] The above is an overview of this embodiment. Next, the identity identification system 10 and the identity identification method using the system according to this embodiment will be described in detail.

[0025] (2) Identity identification system As shown in Figure 1, the identity identification system 10 comprises a reference product feature quantity calculation device 11-1, a product to be judged feature quantity calculation device 11-2, and an identity determination device 12. The reference product feature quantity calculation device 11-1 has an image data acquisition unit 15-1 and a feature quantity calculation unit 16-1. The product to be judged feature quantity calculation device 11-2 has an image data acquisition unit 15-2 and a feature quantity calculation unit 16-2.

[0026] Furthermore, the standard product feature quantity calculation device 11-1 and the judgment target product feature quantity calculation device 11-2 have essentially the same function, differing only in the data they process. Therefore, in this specification, when there is no need to distinguish between the two, each device may simply be referred to as "feature quantity calculation device 11".

[0027] Furthermore, each device included in the identity identification system 10 is implemented by a computer. That is, each device is implemented by a computer through the execution of a program stored in a memory device (ROM, etc.) by an arithmetic unit such as a CPU. These devices may be implemented by different computers or by the same computer.

[0028] (3) Identity identification method Figure 3 is a flowchart illustrating the identity identification method using the identity identification system 10 according to this embodiment. The functions of each component included in the identity identification system 10 according to this embodiment and the identity identification method will be described in detail below with reference to Figures 1 and 3.

[0029] (Step S1-1) Imaging of the reference sample First, the identification layer 4 of reference sample 1-1 is imaged. The identification layer 4 is imaged using, for example, a digital imaging device. Specifically, the identification layer 4 is imaged using an imaging device equipped with an optical microscope.

[0030] During imaging, illumination light is shone onto the identification layer. The mode of illumination is not particularly limited. The illumination light may be side illumination, direct illumination, or a combination of both. However, preferably, the illumination light used has an average intensity in the 380-490 nm wavelength range in its spectral distribution that is 50% or less of the average intensity in the 500-780 nm wavelength range. Using such illumination light reduces the reflection of light from the underlying layer, making the pattern formed by the pattern-forming particles 5 easier to recognize. In particular, when the underlying layer of the identification layer 4 is the back surface resin layer of a semiconductor device (details will be described later), the reflection of light from the underlying layer is more easily reduced, and noise can be reduced. Furthermore, when silicon particles, described later, are used as the pattern-forming particles 5, light below 490 nm tends not to be scattered easily by the silicon particles. Therefore, even if the intensity in the wavelength range below 490 nm is reduced, the color development of the pattern-forming particles 5 is not easily hindered. Therefore, it is possible to reduce the effect of reducing light reflection in the underlying layer while minimizing the impact on the ease of recognition of the pattern formed by the pattern-forming particles 5. Such specific illumination light may be obtained, for example, by using a light-cutting filter, or by using a light source that combines only red LEDs and green LEDs.

[0031] The magnification during imaging of the discrimination layer 4 is not particularly limited, but is, for example, 300 to 3000 times.

[0032] Furthermore, the size of the image used in the image processing described below is not particularly limited. For example, an image of an area with sides of 10 μm or more is used. Preferably, an image of a rectangular area with sides of 10 μm to 1000 μm is used. Therefore, at least an area of ​​this size should be captured during imaging.

[0033] (Step S2-1) Acquisition of HSV data The image obtained by imaging is input as data to the reference product feature quantity calculation device 11-1. In the reference product feature quantity calculation device 11-1, the image data acquisition unit 15-1 acquires the image data of the discrimination layer 4. At this time, the image data acquisition unit 15-1 acquires the image data as HSV data. As described above, HSV data is data that represents the image using the HSV color space. That is, it is data that represents the image using the elements of H value (hue), S value (saturation), and V value (brightness).

[0034] In digital imaging devices, image data is often generated in a format other than HSV. Generally, as described above, image data is generated as RGB format data. Therefore, if the input image data is in RGB format, the image data acquisition unit 15-1 also includes a data conversion unit (not shown) to perform data conversion and acquire HSV data. In this way, the image data acquisition unit 15-1 may acquire HSV format image data indirectly via HSV conversion rather than directly from the identification layer. However, if the input data is HSV data, the image data acquisition unit 15-1 only needs to acquire that HSV data.

[0035] (Step S3-1) Calculation of characteristic quantities of the standard product After acquiring HSV data, the feature calculation unit 16-1 of the reference product feature calculation device 11-1 calculates the feature quantities of reference product 1-1 from the HSV data. The feature quantities calculated here are the feature quantities related to the pattern of the pattern-forming particles 5. The calculated feature quantities of the reference product are stored, for example, in a server (not shown).

[0036] Figure 4 is a flowchart showing a preferred embodiment of the feature calculation method. In the example shown in Figure 4, first, color separation is performed (step S3-A). Next, the positions of pattern-forming particles are detected based on the data after color separation (step S3-B). Then, feature quantities are calculated based on the detected positions of pattern-forming particles (step S3-C). Each step will be described in detail below.

[0037] (Step S3-A) Color Separation First, color separation is performed on the HSV data. Specifically, data within a certain range of the HSV color space is extracted as color separation data.

[0038] The range extracted as color separation data may be single or multiple. Preferably, color separation data is extracted from each of multiple ranges in the HSV color space. Figure 5 is a conceptual diagram showing a preferred example of the color separation process. Figure 5(A) shows the image data before color separation. On the other hand, Figures 5(B) and 5(C) show the data after color separation, respectively. In the example shown in Figure 5, data within the first range in the HSV color space is extracted as the first color separation data (see Figure 5(B)). Also, data within the second range in the HSV color space is extracted as the second color separation data (see Figure 5(C)). Each color separation data contains multiple bright spots caused by the pattern-forming particles 5.

[0039] As described above, by extracting color separation data from each of multiple ranges, it becomes possible to identify identity with greater accuracy.

[0040] The extraction range for color separation data may be a portion of the H value, a portion of the S value, a portion of the V value, or a combination of these.

[0041] Preferably, data with H values ​​(hue) within a certain range in the HSV color space is extracted as color separation data. As described above, pattern-forming particles 5 emit different colors depending on their particle size. In this case, by extracting data with H values ​​within a certain range as color separation data, it is possible to extract intuitively understandable color regions such as "red" and "green". Then, in the next step S3-B (location detection of pattern-forming particles), detecting pattern-forming particles in a specific color region significantly improves the accuracy of identity determination compared to detecting pattern-forming particles in grayscale. It is also possible to selectively extract colors that pattern-forming particles 5 are likely to generate.

[0042] The specific color separation process involves the feature calculation unit 16 first displaying the HSV data as an image on a display device (not shown). The feature calculation unit 16 then accepts input from the user for the extraction range of the color separation data. Once the user inputs the extraction range, the feature calculation unit 16 displays the extracted color separation data on the display device. The user can then review the color separation data to confirm whether the extracted data is suitable for detecting pattern-forming particles. If the user finds no problems, they input a confirmation command. Upon receiving the confirmation command, the feature calculation unit 16 finalizes the color separation data as the data to be used in the next step. This method allows for the extraction of color separation data within an optimal range.

[0043] Furthermore, when the extraction range of color separation data is specified by the user, it is preferable to extract data where the H value falls within a certain range as color separation data. When color separation data with extracted H values ​​is displayed as image data, the user can understand the detection of pattern-forming particles for each intuitively recognizable color, such as "red" or "green." Therefore, it becomes easier to detect pattern-forming particles from multiple color regions, and identity can be determined with greater accuracy.

[0044] In a more preferred embodiment, when extracting color separation data, a range of hues (H values) from red to yellow is selected as the first range, and a range of hues (H values) of green is selected as the second range. That is, data from red to yellow is extracted as the first color separation data. In addition, data for green is extracted as the second color separation data. On the other hand, it is preferable that the range of blue to purple is not extracted. When using the pattern-forming particles 5 (particles of inorganic material with a refractive index n of 3 or more) specified in this embodiment, the pattern-forming particles 5 tend to produce red to yellow and green colors easily, but not blue to purple colors easily. Therefore, by extracting color separation data within the above range, it becomes possible to detect the pattern-forming particles 5 with greater accuracy.

[0045] When extracting red to yellow data as color separation data, and the H value is expressed as 0 to 360°, the lower limit of the extraction range (lower limit of the H value) is, for example, -76 to -20°. The upper limit of the extraction range is, for example, 46 to 75°. However, if the angle is a negative value, that angle is represented as θ, and the angle is expressed as 360° + θ.

[0046] Furthermore, when extracting green data as color separation data, and the H value is expressed as 0 to 360°, the lower limit of the extraction range is, for example, 63 to 89°. The upper limit of the extraction range is, for example, 150 to 173°. Also, if the pattern-forming particles 5 are capable of producing blue to purple colors, blue to purple data may be extracted as color separation data, and when the H value is expressed as 0 to 360°, the lower limit of the extraction range is, for example, 157 to 184°. The upper limit of the extraction range is, for example, 258 to 289°.

[0047] For example, in color separation processing, binarization processing may be performed. If data within a certain range of H values ​​is extracted during color separation processing, binarization processing is performed based on both or one of the remaining S and V values.

[0048] (Step S3-B) Detection of the position of pattern-forming particles As shown in Figure 4, after color separation, the feature calculation unit 16 detects the positions of the pattern-forming particles. That is, the positions of the pattern-forming particles are detected from the extracted color separation data. If multiple color separation data sets are extracted, the positions of the pattern-forming particles are detected from each of these color separation data sets. That is, if first color separation data and second color separation data sets are extracted, the positions of the pattern-forming particles are detected from both the first and second color separation data sets.

[0049] The feature calculation unit 16 detects the position of pattern-forming particles based on the image data after binarization. More specifically, the feature calculation unit 16 extracts, for example, the contour lines of each pattern-forming particle (lines that define the region where pixels determined to constitute bright spots by binarization are clustered) from the image data after binarization, quantifies the region constituting each contour line (the region where each pattern-forming particle exists), calculates the coordinates of the center (centroid) of each quantified region (centroid coordinates), and detects the calculated centroid coordinates as the position of the pattern-forming particle.

[0050] (Step S3-C) Calculation of features Next, the feature calculation unit 16 calculates feature quantities based on the positions of the detected pattern-forming particles. If multiple color separation data sets are extracted, feature quantities are calculated from each of the color separation data sets. That is, if the first color separation data and the second color separation data sets are extracted, the first feature quantity is calculated from the first color separation data set, and the second feature quantity is calculated from the second color separation data set.

[0051] The feature quantities can be, for example, the distance between different pattern-forming particles 5, the angle formed by the straight lines connecting the different pattern-forming particles 5, or a combination thereof. In particular, the feature quantity calculation unit 16 of this embodiment calculates a combination of the distance between different pattern-forming particles 5 and the angle formed by the straight lines connecting these pattern-forming particles 5 as feature quantities. More specifically, the feature quantity calculation unit 16 calculates a combination of the distance between the position of the first pattern-forming particle 5 and the position of the second pattern-forming particle 5, and the angle formed by the straight line connecting the position of the third pattern-forming particle 5 and the position of the first pattern-forming particle 5, and the straight line connecting the position of the third pattern-forming particle 5 and the position of the second pattern-forming particle 5 as feature quantities. Furthermore, in this embodiment, from the viewpoint of obtaining suitable feature quantity calculation results, a preferred origin and coordinate system for representing the position (centroid) of each pattern-forming particle 5 can be appropriately set.

[0052] Figure 6 is a conceptual diagram showing an example of a feature calculation method. In the example shown in Figure 6, the feature calculation unit 16 first recognizes the center O of the image. Next, the feature calculation unit 16 recognizes the particle A closest to the center O and sets the position of particle A as a new origin A. Furthermore, the feature calculation unit 16 recognizes the particle B closest to the origin A. Then, using polar coordinates (r, θ) around the origin A, the feature calculation unit 16 calculates the vectors of each particle as features, with the position of particle B as the reference point B. For example, for each particle C, D, E... on the image, the scalar quantity r that constitutes each vector is calculated as the length of line segment BC, the length of line segment BD, the length of line segment BE... respectively, and the angle θ is calculated as the magnitude of ∠ABC, ∠ABD, ∠ABE... In other words, in this case, for example, the position of particle B is set to the position of the first pattern-forming particle 5 described above, the position of particle C is set to the position of the second pattern-forming particle 5 described above, and the position of particle A is set to the position of the first pattern-forming particle 5 described above and the origin, and the polar coordinate system is selected as the coordinate system to calculate the feature quantities. Feature quantity extraction is performed for all particles present in the image of reference product 1-1, except for particles A and B.

[0053] According to the method described above, particle A is set as a new origin, so high robustness can be obtained even if there is some positional shift during imaging. Furthermore, rotational shifts can be accommodated by using polar coordinates.

[0054] Furthermore, by using vectors of distance and angle as features, it becomes possible to perform accurate authentication even for points far from the reference point.

[0055] (Steps S1-2 to S3-2) Calculation of feature quantities for the product to be judged Please refer again to Figures 1 and 3. When determining whether the item to be judged 1-2 is identical to the reference item, the feature quantities for the item to be judged 1-2 are calculated using the same procedure as for the reference item 1-1. That is, the identification layer 4 of the item to be judged 1-2 is imaged (S1-2). The obtained image data is input to the item to be judged feature quantity calculation device 11-2 in the identity identification system 10. In the item to be judged feature quantity calculation device 11-2, the image data acquisition unit 15-2 acquires the image data as HSV data. Based on the HSV data, the feature quantity calculation unit 16-2 calculates the feature quantities of the item to be judged 1-2 (steps S2-2 and S3-2). When adopting the example of feature calculation method shown in Figure 6, feature quantities are calculated by assuming that multiple particles (20 particles in the example described later) that are closest in distance from the origin O are particles that match particle A of reference product 1-1 (particle A'). In determining identity, the feature quantity calculated using particle A' that shows the comparison result where the similarity between the product to be judged 1-2 and reference product 1-1 is evaluated most highly is used. In other words, when using an error evaluation function as the comparison method, the feature quantity calculated using particle A' that minimizes the error is used.

[0056] (Step S4) Determination of Identity Subsequently, the identity determination device 12 compares the feature quantities of the reference product calculated in step S3-1 with the feature quantities of the product to be determined calculated in step S3-2. Based on the comparison result, it determines whether the product to be determined 1-2 is identical to the reference product 1-1. The identity determination device 12 outputs the determination result via an output device (not shown). The identity determination device 12 may output only the comparison result from the output device, and the determination of identity may be made by an operator based on the comparison result. In other words, the identity determination device 12 may be a device that provides only the information necessary for determining identity.

[0057] The method of comparing features performed by the identity determination device 12 is not particularly limited. For example, the identity determination device 12 calculates the degree of agreement using an error evaluation function from the features of the reference product 1-1 and the features of the product to be determined 1-2. Any evaluation function can be used as the error evaluation function. For example, MAE, MSE, RMSE, and R-squared can be used as evaluation functions. Then, by comparing the calculation result of the degree of agreement with a preset threshold, it is determined whether or not the reference product is identical.

[0058] Furthermore, if multiple types of features are calculated, the degree of agreement is calculated for each type of feature. Based on the degree of agreement for each feature, it is determined whether or not the reference products are identical. For example, if a scalar feature (r expressed in polar coordinates in the example of the feature calculation method shown in Figure 6) and an angular feature (θ expressed in polar coordinates in the example of the feature calculation method shown in Figure 6) are calculated, the degree of agreement is calculated for each particle, for both the scalar feature and the angular feature. Then, for each particle, the sum of the degree of agreement for the scalar feature and the degree of agreement for the angular feature (the sum of the degree of agreement for r expressed in polar coordinates and the degree of agreement for θ expressed in polar coordinates in the example of the feature calculation method shown in Figure 6) is calculated, and based on the calculated value, it is determined whether or not the reference products are identical.

[0059] The degree of agreement is usually calculated for multiple particles. In this case, we will explain an example of how to calculate the degree of agreement using RMSE (Root Mean Square Error) as the evaluation function, as shown in Figure 6, which is an example of the feature calculation method. For example, the polar coordinate of particle C should be compared with the polar coordinate of particle C' which corresponds to particle C in the product being judged to calculate the error. However, it is not specified which particle corresponds to particle C' in the image of the product being judged. Therefore, for all particles C, D, E, etc. included in the image of the reference product, the RMSE is calculated for all possible combinations with the particles included in the image of the product being judged. The combination with the smallest RMSE is considered the true combination and is used as the degree of agreement for judgment. More specifically, for each combination of particles in the reference product and particles in the product being judged, the RMSE(r) of r expressed in polar coordinates and the RMSE(θ) of θ expressed in polar coordinates are calculated, and the polar coordinate RMSE of the combination with the smallest value of the sum of these, polar coordinate RMSE(RMSE(r)+RMSE(θ)), is used as the degree of agreement.

[0060] If features are calculated from multiple color separation data, the degree of agreement is calculated for each feature calculated from the respective color separation data. Based on the degree of agreement for each feature, it is determined whether the reference products are identical. For example, if a first feature is calculated from the first color separation data and a second feature is calculated from the second color separation data, the degree of agreement is calculated for each of the first and second features. The sum of the degree of agreement for the first feature and the degree of agreement for the second feature (in the example of feature calculation method shown in Figure 6, where the degree of agreement is calculated using polar coordinate RMSE, this is the sum of the first polar coordinate RMSE calculated from the first color separation data and extracted as the degree of agreement, and the second polar coordinate RMSE calculated from the second color separation data and extracted as the degree of agreement) is calculated, and based on the calculated value, it is determined whether the reference products are identical.

[0061] This embodiment has been described above. As described above, according to this embodiment, identity is identified using features calculated using HSV data. This makes it possible to identify identity with higher accuracy compared to when features are calculated using other data formats, such as RGB data.

[0062] (4) Modified versions of the identity identification system As described above, each device included in the identity identification system 10 according to this embodiment may be implemented by the same computer or by separate computers. Furthermore, the devices included in the identity identification system 10 may be managed by different businesses.

[0063] Therefore, the identity identification system 10 according to this embodiment can also be considered as a modified version as described below. Figure 7 is a schematic diagram showing a modified version of the identity identification system 10. In this modified version, the identity identification system 10 comprises a reference product data storage device 13, a product feature quantity calculation device 11-2, and a determination device 14.

[0064] The standard product data storage device 13 is a server managed, for example, by the manufacturer of the product in question. The standard product data storage device 13 already stores data related to the standard product as standard product data. Here, the standard product data is image data that shows the identification layer 4 of the standard product.

[0065] The product feature calculation device 11-2 and the determination device 14 are managed, for example, by the purchaser of the product.

[0066] The item feature quantity calculation device 11-2 is the same as that described in the previously stated embodiment. That is, the item feature quantity calculation device 11-2 is configured to calculate features from the image data of the item to be determined.

[0067] On the other hand, the determination device 14 identifies whether the product to be determined is identical to the standard product based on the standard product data stored in the standard product data storage device 13 and the features of the product to be determined calculated by the product feature calculation device 11-2. Specifically, the determination device 14 includes a standard product feature calculation device 11-1 and an identity determination device 12. The functions of these devices are the same as in the embodiments described above. That is, in the determination device 14, the standard product feature calculation device 11-1 acquires the standard product data stored in the standard product data storage device and calculates the features of the standard product based on the acquired standard product data (image data). The identity determination device 12 also acquires the features of the product to be determined from the product feature calculation device 11-2. The identity determination device 12 then compares the features of the product to be determined with those of the standard product. Based on this comparison result, it is identified whether the product to be determined is identical to the standard product.

[0068] The above describes the configuration of the identity identification system 10 according to this modified example. In this modified example, the case in which image data is stored as reference product data in a reference product data storage device 13 managed by the manufacturer has been described. However, the calculated feature quantities of the reference product may also be stored as reference product data. That is, the manufacturer of the product in question may perform the calculation of the feature quantities of the reference product in advance, and the calculated feature quantities may be stored as reference product data in the reference product data storage device 13. In this case, the reference product feature quantity calculation device 11-1 is not required in the determination device 14. In the determination device 14, the feature quantities of the product to be determined are simply compared with the feature quantities of the reference product by the identity determination device 12.

[0069] (5) Target items Next, we will describe the target item 1 whose identity is identified. The specific type of target item 1 is not particularly limited, but in a preferred embodiment, the target item is a semiconductor device. As mentioned above, the circulation of counterfeit semiconductor devices is a problem. By using the technology according to this embodiment, it is possible to accurately determine the authenticity of semiconductor devices. Furthermore, there is a demand for miniaturization in semiconductor devices. According to the technology according to this embodiment, identification can be performed by fine patterns, making it suitable for semiconductor devices where miniaturization is required.

[0070] Figure 8 is a schematic cross-sectional view showing an example of product 1. In this example, product 1 is a semiconductor device. Product 1, being a semiconductor device, has a semiconductor substrate 2, a backside resin layer 3, and an identification layer 4.

[0071] The semiconductor substrate 2 is a substrate on which the integrated circuit 7 is formed. For example, the semiconductor substrate 2 is a semiconductor chip or a semiconductor wafer. For example, the semiconductor substrate 2 is made of silicon.

[0072] The back resin layer 3 is provided for purposes such as protecting the semiconductor substrate 2. In the example shown in Figure 8, the back resin layer 3 is provided on the back surface of the semiconductor substrate 2 (the surface opposite to the surface on which the integrated circuit 7 is formed).

[0073] The identification layer 4 is provided on the back surface resin layer 3.

[0074] (6) Discrimination layer Next, we will describe the identification layer 4 in detail. As previously mentioned, the identification layer 4 contains pattern-forming particles 5.

[0075] The thickness of the identification layer 4 is, for example, 0.1 to 20 μm, preferably 0.2 to 10 μm, more preferably The size is approximately 0.5 to 5 μm.

[0076] The identification layer 4 is formed, for example, by an identification layer composition. The identification layer composition may be provided in the form of a film, or, for example, in the form of a liquid containing a solvent, or in the form of a solvent-free liquid curable composition obtained by fluidizing a low molecular weight polymerizable monomer. The identification layer composition is preferably a resin composition. That is, the identification layer 4 is a cured product of the resin composition. It is preferable that the cured product of this resin composition is as transparent as possible. The identification layer 4 can be formed by supplying a curable identification layer composition onto the back resin layer 3 and curing it.

[0077] The following describes an example of the configuration of the identification layer 4, while explaining the composition of the identification layer composition.

[0078] (Pattern-forming particles) As previously described, the identification layer 4 contains pattern-forming particles 5.

[0079] The refractive index n of the inorganic material forming the pattern-forming particles should be 3 or higher. When the refractive index n is 3 or higher, light of specific wavelengths is selectively scattered by Mie resonance. This makes bright spots formed by individual particles more easily recognizable. The refractive index n is preferably 3.5 or higher, and more preferably 4.0 or higher. A higher refractive index n results in higher reflectivity and higher color development. As a result, the identification pattern becomes more easily recognizable.

[0080] As previously described, the constituent material of the pattern-forming particles 5 is an inorganic material. Using inorganic materials can increase the durability of the pattern. For example, when patterns are formed using pigments and dyes made from organic materials, the patterns tend to fade easily. In contrast, using inorganic materials allows for greater durability than when using organic materials.

[0081] The inorganic material forming the pattern-forming particles is preferably not a pure metal. If a pure metal is used, alteration due to oxidation or other factors may occur. Furthermore, metal elements may migrate from the pattern-forming particles to the surrounding environment. As a result, changes in the pattern may occur. Using an inorganic material other than a pure metal can suppress changes in the pattern due to alteration and migration.

[0082] Specific examples of inorganic materials that form pattern-forming particles include crystals of metal compounds, metalloids, and metalloid compounds. Examples of metal compounds include compounds formed from a metal and at least one element selected from the group consisting of oxygen, nitrogen, phosphorus, arsenic, and sulfur. Examples of metalloids include boron, silicon, germanium, arsenic, antimony, tellurium, selenium, polonium, and astatine. Examples of metalloid compounds include compounds formed from a metalloid and at least one element selected from the group consisting of oxygen, nitrogen, phosphorus, arsenic, and sulfur.

[0083] Specific examples of inorganic materials that form patterned particles include silicon (refractive index 4.32), GaAs (refractive index 4.27), GaP (refractive index 3.6), and InP (refractive index 3.0).

[0084] In a particularly preferred embodiment, the pattern-forming particles are silicon particles, and more preferably crystalline silicon particles. Silicon particles selectively and strongly scatter light of specific wavelengths. Furthermore, patterns formed using silicon particles have very high durability. They also have high heat resistance. Therefore, they are particularly suitable as a constituent material for pattern-forming particles.

[0085] The average particle size of the pattern-forming particles is, for example, 0.09 to 5.0 μm, preferably 0.1 to 2.0 μm. Within this range, a single particle generates scattered light with high intensity in the visible region (blue to red) or the near-infrared region (800 to 1200 nm). Therefore, the patterns formed by the pattern-forming particles become easily recognizable. The average particle size referred to here means the median diameter D50 in the particle size distribution measured using a laser diffraction particle size distribution analyzer.

[0086] The particle size distribution of the pattern-forming particles contained in the identification layer 4 preferably has multiple peaks. As described above, each pattern-forming particle emits a color corresponding to its particle size, and if the pattern-forming particles include multiple particles with different particle sizes, each particle generates a bright spot of a color corresponding to its respective particle size. If the particle size distribution of the pattern-forming particles has multiple peaks, it becomes easier to obtain a pattern of bright spots with different colors. Therefore, the amount of information used for authenticity determination can be increased, making it possible to determine authenticity more reliably.

[0087] In one preferred embodiment, the pattern-forming particles are silicon particles, and the pattern-forming particles contain silicon particles having a particle size of 90 to 300 nm. More preferably, the particle size distribution of the pattern-forming particles has a cumulative value of 20% or more of the presence in the 90 to 300 nm range, and even more preferably, 25% or more. With respect to the particle size distribution of the pattern-forming particles, the upper limit of the cumulative value of the presence in the 90 to 300 nm range may be 100% or less, or 70% or less. In such a particle size range, individual particles tend to develop a color corresponding to their particle size, making it easy to increase the density of bright spots in the identification pattern.

[0088] The content of pattern-forming particles is, for example, 0.01 to 5% by mass, preferably 0.05 to 1% by mass, based on the total mass of the identification layer composition excluding the solvent. Within this range, the pattern-forming particles are dispersed in a suitable dispersion state for use as an identification pattern.

[0089] Furthermore, while it is possible to determine the nature of the pattern-forming particles within the identification layer by directly analyzing the identification layer, it is also possible to determine them by extracting the pattern-forming particles from the identification layer and observing them. For example, if the identification layer is a resin composition, the pattern-forming particles can be selectively extracted by dissolving the resin components with a solvent. Alternatively, the identification layer can be fired to remove components other than the pattern-forming particles that function as a binder, thereby extracting only the pattern-forming particles.

[0090] (Curable resin composition) As described above, the identification layer composition is preferably a curable resin composition. The identification layer composition may be a thermosetting composition, an energy ray curable composition, or both a thermosetting composition and an energy ray curable composition.

[0091] As an energy-ray curable identification layer composition, for example, a composition comprising an energy-ray curable compound (e.g., monomer or oligomer) and a photopolymerization initiator can be used. Examples of energy-ray curable compounds include compounds having unsaturated hydrocarbon groups that can be polymerized by irradiation with energy rays, such as polyacrylate compounds.

[0092] Examples of photopolymerization initiators include photoradical polymerization initiators.

[0093] Furthermore, the identification layer composition may contain components other than energy ray-curable compounds and photopolymerization initiators, such as silane coupling agents.

[0094] (7) Resin layer on the back As previously described, the back resin layer 3 is a layer intended for the protection of the semiconductor substrate 2. In the example shown in Figure 8, the back resin layer 3 is a back surface coating of the semiconductor substrate 2. More preferably, the back resin layer 3 is a back surface coating provided on the back surface of a semiconductor substrate 2 (e.g., a semiconductor chip) that is mounted in a face-down manner. However, the back resin layer 3 does not necessarily have to be a back surface coating. The back resin layer 3 is a layer intended for the protection of the semiconductor substrate 2 and can serve as a base for the identification layer 4. For example, the back resin layer 3 may be a sealing material layer that seals the semiconductor substrate 2. That is, it may be a layer formed of a so-called sealing resin.

[0095] The back surface resin layer 3 is formed from a back surface resin layer forming composition. The back surface resin layer forming composition is preferably a curable composition. For example, the back surface resin layer 3 is formed by supplying the back surface resin layer forming composition onto a semiconductor substrate 2 and curing the supplied back surface resin layer forming composition. The back surface resin layer forming composition may be a thermosetting composition or an energy ray curable composition. It may also be a composition having both thermosetting and energy ray curing properties.

[0096] When the back surface resin layer 3 is a back surface coating, the back surface resin layer forming composition is preferably provided in film form. For example, as the back surface resin layer forming film, a film is prepared in which the back surface resin layer forming composition is supported on a support film. Then, the back surface resin layer forming composition is transferred onto the semiconductor substrate 2 using the back surface resin layer forming film. Subsequently, the back surface resin layer 3 is formed by curing the transferred back surface resin layer forming composition.

[0097] On the other hand, if the back surface resin layer 3 is a sealing material layer, the back surface resin layer 3 can also be formed by molding a back surface resin layer formation composition on the semiconductor substrate 2 using a mold.

[0098] However, in a preferred embodiment, the back surface resin layer 3 is a back surface coating formed using a back surface resin layer forming film. When the back surface resin layer 3 is a sealing layer and a mold is used for its molding, a release agent is usually used. As a result, components of the release agent may transfer to the surface of the back surface resin layer 3. The transferred components of the release agent may reduce the adhesion of the identification layer 4. In contrast, when forming a back surface coating using a back surface resin layer forming film, the release film used as a support film for the back surface resin layer forming film is relatively resistant to release agent transfer. Therefore, from the viewpoint of adhesion, it is preferable to use a back surface coating formed using a back surface resin layer forming film as the back surface resin layer 3. That is, it is preferable to use the back surface coating formed in this way as the base layer for the identification layer 4.

[0099] Furthermore, the surface of the back-side coating formed by transfer from the back-side resin layer forming film tends to be smoother than the surface of the sealing material layer obtained using a mold. A smoother surface on the underlying back-side resin layer 3 makes it easier to obtain good contrast when imaging the identification layer 4. As a result, the authenticity determination pattern becomes easier to recognize accurately. From this viewpoint as well, it is preferable to use a back-side coating formed using a back-side resin layer forming film as the base layer for the identification layer 4 (i.e., as the back-side resin layer 3).

[0100] When the back surface resin layer 3 is a back surface coating, the thickness of the back surface resin layer 3 is, for example, 1 to 100 μm, preferably 5 to 50 μm.

[0101] As previously described, the back surface resin layer 3 is formed by a back surface resin layer forming composition. As previously described, the back surface resin layer forming composition may be provided in the form of a film, but may also be provided in the form of a liquid containing a solvent, for example, or in the form of a solvent-free liquid curable composition that is fluidized by incorporating low molecular weight polymerizable monomers. The composition of the back surface resin layer 3 will be described in more detail below by describing the composition of the back surface resin layer forming composition.

[0102] (Composition for forming a thermosetting resin layer on the back surface) If the composition is thermosetting, the composition for forming the back resin layer includes, for example, a thermosetting component (A), a colorant (B), etc. If the composition for forming the back resin layer is in the form of a back resin layer forming film, it is also preferable to include a polymer component (C).

[0103] The content of polymer component (A) is, for example, 5 to 50% by mass, preferably 10 to 40% by mass, and more preferably 15 to 35% by mass, based on the total mass of the back surface resin layer forming composition excluding the solvent.

[0104] (A) Thermosetting component Examples of the thermosetting component (B) include epoxy-based thermosetting resins, thermosetting polyimides, thermosetting polyurethanes, unsaturated polyesters, and silicone rubbers. Preferably, it is an epoxy-based thermosetting resin.

[0105] (B) Colorants As the coloring agent (E), inorganic pigments, organic pigments, and organic dyes can be used. Among these, inorganic pigments are preferred because they do not fade easily.

[0106] (C) Polymer components Examples of polymer components (C) include acrylic resins, polyesters, urethane resins, acrylic urethane resins, silicone resins, rubber resins, and phenoxy resins.

[0107] (Energy ray curable composition for forming a backside resin layer) As previously mentioned, the composition for forming the back surface resin layer may also be an energy-curable composition. In this case, the composition for forming the back surface resin layer can be realized, for example, by a composition containing the energy-curable compound and photopolymerization initiator described above for the "identification layer".

[0108] Furthermore, if the curing type of the back surface resin layer forming composition is the same as that of the identification layer composition, the back surface resin layer forming composition and the identification layer composition can be cured together.

[0109] (Filler) The composition for forming the backside resin layer may contain a filler. By using a filler, the physical properties of the backside resin layer can be adjusted. For example, the physical properties of the backside resin layer can be made closer to those of the semiconductor substrate 2.

[0110] The filler content in the back surface resin layer forming composition (i.e., the filler content in the back surface resin layer) is preferably 30 to 70% by mass of the total mass of the back surface resin layer forming composition (calculated on the basis of active ingredients excluding the solvent). More preferably, the filler content is 35 to 65% by mass of the total mass of the back surface resin layer.

[0111] The material of the filler is not particularly limited. For example, the filler may be an inorganic filler or an organic filler. Preferably, it is an inorganic filler. Examples of inorganic fillers include at least one selected from the group consisting of silica, alumina, talc, calcium carbonate, titanium white, red iron oxide, silicon carbide, and boron nitride. Preferably, the filler is silica.

[0112] The size of the filler is not particularly limited. The average particle size of the filler is, for example, 0.05 to 5 μm, preferably 0.1 to 1.0 μm, and more preferably 0.1 to 0.5 μm. The average particle size referred to here is the value measured using a laser diffraction particle size analyzer.

[0113] Preferably, the filler is spherical. Using a spherical filler makes it easier to obtain the same pattern regardless of the lighting environment. That is, when observing the identification layer, illumination light is incident on the back resin layer 3. The incident illumination light is reflected by the filler. In this case, if the filler is spherical, the intensity and direction of the reflected light from the filler are less dependent on the direction of incidence of the illumination light. Therefore, it is easier to obtain the same pattern even if the lighting environment is different. Consequently, it becomes possible to accurately identify identical layers.

[0114] The spherical filler may be either oblate or perfectly spherical. Preferably, the spherical filler is perfectly spherical.

[0115] Preferably, 80% by mass or more of the total filler is spherical filler. More preferably, 90% by mass or more of the total filler is spherical filler, and even more preferably, 95% by mass or more of the total filler is spherical filler. Most preferably, substantially all of the filler is spherical filler.

[0116] (8) Manufacturing method Next, we will explain an example of a manufacturing method for a semiconductor device, assuming that the target item 1 is a semiconductor device.

[0117] First, a semiconductor substrate 2 is prepared, and a backside resin layer 3 is formed on the semiconductor substrate 2. Specifically, a protective layer-forming composition is supplied onto the semiconductor substrate 2. Then, the backside resin layer-forming composition is cured. This forms the backside resin layer 3. Next, an identification layer composition is supplied onto the backside resin layer 3. Then, the supplied identification layer composition is cured. This forms the identification layer 4. This results in a semiconductor device. Note that the curing of the backside resin layer-forming composition and the curing of the identification layer composition may be carried out simultaneously.

[0118] Semiconductor devices may be manufactured by a so-called wafer-level packaging process or by a panel-level packaging process. Examples of wafer-level packaging processes include wafer-level chip-size packaging (WLCSP) and fan-out wafer-level packaging. Examples of panel-level packaging processes include fan-out panel-level packaging. In these packaging processes, a backside resin layer 3 is formed on multiple semiconductor devices (typically multiple semiconductor chips) at once. Subsequently, the multiple semiconductor devices are separated into individual components. Here, the identification layer 4 can be formed all at once after the formation of the backside resin layer 3 and before the separation of the components. Since the identification layer 4 can be formed on multiple semiconductor devices at once, it is efficient.

[0119] In a preferred example, the semiconductor substrate 2 is a face-down type semiconductor chip. During manufacturing, the back surface resin layer 3 is formed on the back surface of the wafer for the face-down type semiconductor chip as a back surface coating. [Examples]

[0120] The present invention will now be described in more detail by the inventors, based on examples implemented by the inventors. However, the present invention should not be construed as being limited to the following examples.

[0121] (Prototype A) <Manufacturing of film for forming a resin layer on the back surface> (Preparation of composition for forming the resin layer on the back surface) A composition for forming a back surface resin layer was prepared by mixing the following raw materials with a methyl ethyl ketone solvent and stirring at 23°C for 60 minutes, with an active ingredient (components other than the solvent) content of 52% by mass. In the following description, the amounts of each component refer to the amount of the active ingredient.

[0122] (1) Acrylic polymer (20 parts by mass): Acrylic polymer obtained by copolymerizing 15 parts by mass of n-butyl acrylate, 10 parts by mass of methyl methacrylate, 60 parts by mass of methyl acrylate, and 15 parts by mass of 2-hydroxyethyl acrylate (weight-average molecular weight: 600,000) (2) Bisphenol A type liquid epoxy resin (15 parts by mass): Nippon Shokubai Co., Ltd. "BPA328" (3) Bisphenol A type epoxy resin (1.8 parts by mass): "jER1055" manufactured by Mitsubishi Chemical Corporation (4) Dicyandiamide (0.45 parts by mass): Thermally activated latent epoxy resin curing agent, ADEKA "ADEKA Hardener EH-3636AS" (5) 2-phenyl-4,5-dihydroxymethylimidazole (0.45 parts by mass), manufactured by Shikoku Chemicals Co., Ltd., "Curesol 2PHZ" (6) Spherical silica filler (60 parts by mass): Admatex Co., Ltd. "SC105G-MMQ: Spherical Silica" (average particle size 0.3 μm) (7) Silane coupling agent: 3-Glycidoxypropyltrimethoxysilane (3-Glycidyloxypropyltrimethoxysilane) (0.4 parts by mass), Shin-Etsu Chemical Co., Ltd. "KBM403" (8) Coloring agent: Carbon black (1.9 parts by mass), Mitsubishi Chemical Corporation "MA600B" (average particle size 28 nm)

[0123] (Formation of film for forming the resin layer on the back surface) A release film having a release surface was prepared. The backside resin layer forming composition prepared above was then coated onto the release surface using a knife coater. After coating, it was dried at 110°C for 2 minutes. The thickness of the backside resin layer forming composition after drying was 25 μm. Furthermore, the release surface of another release film was bonded onto this backside resin layer forming composition to create a backside resin layer forming film sandwiched between two release films.

[0124] (Manufacturing of silicone chips with a resin layer on the back) A 6-inch silicon wafer (100 μm thick) with a #2000 polished surface was prepared as a semiconductor substrate. One release film was peeled off from the film for forming the backside resin layer, and the exposed backside resin layer formation composition was applied to the silicon wafer. The other release film remaining on the backside resin layer formation film was then peeled off and removed. Next, the silicon wafer was heat-treated at 130°C for 2 hours to cure the backside resin layer formation composition and form a backside resin layer on the silicon wafer.

[0125] (Formation of the discrimination layer) On the other hand, a composition for the identification layer was prepared. Specifically, 91 parts by mass of polyester acrylate (Arkema Corporation CN2270NS), 0.1 parts by mass of Si (silicon) nanofiller (average particle size 0.8 μm), 4.6 parts by mass of photoinitiator (IGM Resins, Omnirad 127D), and 4.6 parts by mass of silane coupling agent (Shin-Etsu Chemical Co., Ltd. KBM-1083) were mixed to prepare the composition for the identification layer.

[0126] Furthermore, the particle size distribution of Si (silicon) nanofillers was measured using a laser diffraction particle size distribution analyzer (Mastersizer 3000 (Malvern Panalytical)), and the graph shown in Figure 9 was obtained. In the graph shown in Figure 9, the horizontal axis represents the diameter of the filler (μm), and the vertical axis represents the abundance (%). As shown in Figure 9, a particle size distribution with two peaks was obtained. The integrated abundance in the 90-300 nm range was 34%. The integrated abundance in the 800-1200 nm range was 27.8%.

[0127] The prepared identification layer composition was applied to the surface of the backside resin layer formed on a silicon wafer using a YOSHIMITSU SEIKI YBA-type baker applicator. The composition was then cured by UV irradiation under the following conditions to obtain an identification layer with a thickness of 5 μm. Equipment: UV irradiation machine (GS Yuasa Corporation, CSN2-40) Conditions: N2 purge present, UV irradiance 360 ​​mJ / cm² 2 Irradiation speed 7 m / min, oxygen concentration 0.3% or less.

[0128] Following the above procedure, a silicon wafer for prototype A was obtained. This was attached to Lintec's Adwill D-485H adhesive sheet for semiconductor processing, and then cut into 10mm x 10mm squares using a dicer (DISCO DFD6362) to obtain individual pieces, resulting in a semiconductor chip with an identification layer that mimics a semiconductor device, for prototype A.

[0129] (Prototypes B, C, D) By extracting chips different from prototype A from the silicon wafer used to obtain prototype A, semiconductor chips with identification layers related to prototypes B, C, and D were obtained.

[0130] (Acquisition of characteristic features of standard products) The identification layer of prototype A was imaged using a digital microscope (Keyence Digital Microscope VHX-7000). Imaging was performed under omnidirectional illumination with side illumination (all ring illuminations on). The optical magnification was 2500x. In the image, it was observed that multiple bright spots emitted different colors.

[0131] In this embodiment, a program was created using a programming language embedded in MATLAB®, an integrated development environment, to analyze an image based on an example of the algorithm described with reference to Figure 6. This program analyzed the obtained image and calculated the total RMSE.

[0132] Specifically, the obtained image data (RGB format) was converted to HSV format image data. Next, color separation was performed on the HSV format image data, and color separation data was extracted. The color separation data was decomposed into two ranges: a first range where the H value is within the red to yellow region (H value -27.72 (332.28) to 66.24°) and a second range where the H value is within the green region (H value 69.84 to 160.56°). Then, binarization was performed on each range using S and V values ​​to obtain the first and second color separation data, respectively. From the obtained color separation data, the centroid coordinates of the pattern-forming particles were calculated. Then, following the method shown in Figure 6, the origin A and reference point B were determined, and the vector (polar coordinates) of each pattern-forming particle was calculated from reference point B as a feature quantity of the reference product. That is, the first feature quantity was calculated from the first color separation data (red to yellow region), and the second feature quantity was calculated from the second color separation data (green region).

[0133] (Example 1) For the discrimination layer of prototype A, an area 30 pixels to the left of the area used to acquire the features of the reference product was imaged. Then, using the same method as when acquiring the features of the reference product, the features related to Example 1 (first feature and second feature) were calculated.

[0134] Next, the feature quantities of the reference product were compared with the feature quantities related to Example 1. Specifically, the polar coordinate RMSE, as described in the embodiment, was calculated for each of the first and second feature quantities. Then, the sum of the first polar coordinate RMSE of the first feature quantity and the second polar coordinate RMSE of the second feature quantity was calculated as the total polar coordinate RMSE (HSV).

[0135] In addition to the total RMSE mentioned above, the RMSE for grayscale image processing was also calculated. Specifically, the polar coordinate RMSE extracted as the degree of agreement when RGB image data obtained simply by imaging, without converting to HSV format, was binarized using grayscale was calculated as the polar coordinate RMSE (grayscale). Adaptive binarization was used for this binarization process.

[0136] (Example 2) For the discrimination layer of prototype A, an area 70 pixels to the left of the area where the feature quantities of the reference product were acquired was imaged. Other aspects were the same as in Example 1, and the total polar coordinate RMSE(HSV) and polar coordinate RMSE(grayscale) for Example 2 were determined.

[0137] (Example 3) For the discrimination layer of prototype A, an area 80 pixels to the left of the area where the feature quantities of the reference product were acquired was imaged. Other aspects were the same as in Example 1, and the total polar coordinate RMSE(HSV) and polar coordinate RMSE(grayscale) in Example 3 were determined.

[0138] (Example 4) For the identification layer of prototype A, imaging was performed with the image rotated 90 degrees relative to the time when the feature quantities of the reference product were acquired. Other aspects were the same as in Example 1, and the total polar coordinate RMSE(HSV) and polar coordinate RMSE(grayscale) for Example 4 were determined.

[0139] (Example 5) For the identification layer of prototype A, imaging was performed with the image rotated 180° compared to when the feature quantities of the reference product were acquired. Other aspects were the same as in Example 1, and the total polar coordinate RMSE(HSV) and polar coordinate RMSE(grayscale) for Example 5 were determined.

[0140] (Example 6) For the discrimination layer of prototype B, the sum polar coordinate RMSE(HSV) and polar coordinate RMSE(grayscale) were determined using the same method as when acquiring the features of the reference product.

[0141] (Example 7) For the discrimination layer of prototype C, the sum polar coordinate RMSE(HSV) and polar coordinate RMSE(grayscale) were determined using the same method as when acquiring the features of the reference product.

[0142] (Example 8) For the discrimination layer of prototype D, the sum polar coordinate RMSE(HSV) and polar coordinate RMSE(grayscale) were determined using the same method as when acquiring the features of the reference product.

[0143] (Results and Discussion) Table 1 shows the results. Examples 1-5 all had significantly smaller total RMSE (HSV) values ​​compared to Examples 6-8. Specifically, there was a difference of more than 13 orders of magnitude. On the other hand, the difference in total RMSE (grayscale) between Examples 1-5 and Examples 1-8 was only a few times, and not as significant as the difference in total RMSE (HSV). From this, it can be understood that using total RMSE (HSV) makes it easy to identify whether or not a product is identical to a reference product. In other words, it can be understood that identity can be identified with high accuracy by extracting features from HSV data.

[0144] [Table 1]

[0145] [Note] The scope of matters described in the original specification of this application includes the following I to III:

[0146] [I] A computer (10) programmed to calculate feature quantities (r,θ) for identifying the identity of the target items (1-1,1-2), Image data obtained by imaging the identification layer 4 of the target product (1-1, 1-2) under the illumination of a specific illumination light is acquired. The program is designed to calculate the feature quantities (r,θ) of the target items (1-1,1-2) based on the image data (S3-1,S3-2). The identification layer 4 contains dispersed pattern-forming particles 5 made of an inorganic material with a refractive index n of 3 or higher. The image data is represented in the HSV color space. The feature quantity (r,θ) is a quantity that suggests the pattern (position and color) of the pattern-forming particles 5 in the discrimination layer 4, computer (10).

[0147] [II] A computer (10) programmed to determine the identity between reference product 1-1 and product 1-2, Image data obtained by imaging the identification layer 4 of standard product 1-1 under the illumination of a specific illumination light is acquired. Based on the image data, the feature quantities (r,θ) of standard product 1-1 are calculated (S3-1), The feature quantities of the items to be judged 1-2 stored in a designated memory area are read out. The program is programmed to compare the calculated feature quantities (r,θ) of reference product 1-1 with the read-out feature quantities of the product to be judged 1-2 (S4). The identification layer 4 contains dispersed pattern-forming particles 5 made of an inorganic material with a refractive index n of 3 or higher. The image data is represented in the HSV color space. The feature quantity (r,θ) is a quantity that suggests the pattern (position and color) of the pattern-forming particles 5 in the discrimination layer 4, computer (10).

[0148] [III] A computer (10) programmed to determine the identity between standard item 1-1 and item 1-2, Image data obtained by imaging the identification layer 4 of the items to be judged 1-2 under the illumination of a specific light source is acquired. Based on the image data, the feature quantities (r,θ) of the target item 1-2 are calculated (S3-2), The reference product data (in particular, the feature quantities of the reference product) stored in the predetermined memory area (13) is read out. The program is designed to compare the calculated feature quantities (r,θ) of the target product 1-2 with the retrieved reference product data (especially the feature quantities of the reference product). The identification layer 4 contains dispersed pattern-forming particles 5 made of an inorganic material with a refractive index n of 3 or higher. The image data is represented in the HSV color space. The feature quantity (r,θ) is a quantity that suggests the pattern (position and color) of the pattern-forming particles 5 in the discrimination layer 4, computer (10). [Explanation of symbols]

[0149] 1. Target product, 2. Semiconductor substrate, 3. Backside resin layer, 4. Identification layer, 5. Pattern-forming particles, 6. Integrated circuit, 10. Identity identification system, 11-1. Reference product feature quantity calculation device, 11-2. Judgment target product feature quantity calculation device, 12. Identity determination device, 13. Reference product data storage device, 14. Determination device, 15 (15-1~15-2) Image data acquisition unit, 16 (16-1~16-2) Feature quantity calculation unit

Claims

1. A feature calculation device that calculates feature quantities for identifying the identity of a target product, An image data acquisition unit that acquires image data showing the identification layer of the target product, A feature quantity calculation unit calculates the feature quantities of the target product based on the image data, Equipped with, The identification layer contains dispersed pattern-forming particles made of an inorganic material with a refractive index n of 3 or more. The aforementioned feature quantities are feature quantities relating to the pattern of the pattern-forming particles, The aforementioned image data is data represented in the HSV color space. Feature extraction device.

2. A feature quantity calculation device according to claim 1, The feature calculation unit, From the aforementioned image data, a portion of the data in the HSV color space is extracted as color separation data. The feature quantities are calculated from the aforementioned color separation data. Feature extraction device.

3. A feature calculation device according to claim 2, The feature calculation unit, As the aforementioned color separation data, at least first color separation data, which is data within a first range in the HSV color space, and second color separation data, which is data within a second range in the HSV color space, are extracted. As the aforementioned features, a first feature is calculated from the first color separation data, and a second feature is calculated from the second color separation data. Feature extraction device.

4. A feature calculation device according to claim 2, The aforementioned color separation data is data where the H value in the HSV color space falls within a certain range. Feature extraction device.

5. A feature calculation device according to claim 2, The feature calculation unit, Based on the aforementioned color separation data, the position of the pattern-forming particles is detected. Based on the positions of the detected pattern-forming particles, the feature quantities are calculated. Feature extraction device.

6. A feature quantity calculation device according to claim 1 or 2, The aforementioned feature quantity is either the distance between the pattern-forming particles, the angle formed by the straight lines connecting the pattern-forming particles, or a combination thereof. Feature extraction device.

7. A feature quantity calculation device according to claim 1 or 2, The image data is data of an image obtained by imaging the identification layer under the irradiation of a specific illumination light. The aforementioned specific illumination light is light in which, in the spectral spectrum, the average intensity in the wavelength region of 380 to 490 nm is 50% or less of the average intensity in the wavelength region of 500 to 780 nm. Feature extraction device.

8. A feature quantity calculation device according to claim 1 or 2, The aforementioned product is a semiconductor device. Feature extraction device.

9. A feature quantity calculation device according to claim 8, The semiconductor device includes a back surface resin layer, The identification layer is formed on the back surface resin layer, Feature extraction device.

10. Identity identification system comprising the feature quantity calculation device described in claim 1 or 2 and an identity determination device, The feature quantity calculation device calculates the feature quantities of a reference product as the target product, The identity determination device compares the calculated characteristic quantities of the reference product with the characteristic quantities of the product to be determined. Identity identification system.

11. A reference product feature quantity calculation device as the feature quantity calculation device described in claim 1 or 2, Calculate the feature quantities of a reference product to compare with the feature quantities of the product being evaluated. A device for calculating characteristic features of a standard product.

12. Identity identification system comprising the feature quantity calculation device described in claim 1 or 2, a reference product data storage device for storing reference product data, and a determination device, The feature quantity calculation device calculates the feature quantities of the target product to be judged, The determination device compares the standard product data with the characteristic quantities of the product to be determined. Identity identification system.

13. A determination device that compares standard product data stored in a standard product data storage device with the characteristic quantities of a product to be judged, The characteristic quantities of the product to be judged are calculated by the characteristic quantity calculation device described in claim 1 or 2. Judgment device.

14. A feature calculation method for calculating feature quantities to identify the identity of target items, Image data acquisition step to obtain image data showing the identification layer of the target product, The system includes a feature calculation step of calculating feature quantities for identifying the identity of the target item based on the image data, The identification layer contains dispersed pattern-forming particles made of an inorganic material with a refractive index n of 3 or more. The aforementioned feature quantities are feature quantities relating to the pattern of the pattern-forming particles, The aforementioned image data is data represented in the HSV color space. Feature extraction method.

15. A feature calculation program that causes a computer to execute a feature calculation method for calculating feature quantities to identify the identity of target items, The feature calculation method described above is: Image data acquisition step to obtain image data showing the identification layer of the target product, A feature calculation step, which involves calculating feature quantities for identifying the identity of the target item based on the aforementioned image data, Equipped with, The identification layer contains dispersed pattern-forming particles made of an inorganic material with a refractive index n of 3 or more. The aforementioned feature quantities are feature quantities relating to the pattern of the pattern-forming particles, The aforementioned image data is data represented in the HSV color space. Feature extraction program.

Images