JP7453931B2 - Learning data generation device, anomaly detection device, anomaly detection system, and anomaly detection program - Google Patents

Description

The present invention relates to a learning data generation device, an anomaly detection device, an anomaly detection system, and an anomaly detection program.

2. Description of the Related Art In recent years, technology has been proposed for detecting abnormalities in equipment, products, etc. using AI (Artificial Intelligence) that performs machine learning in the inspection of equipment, products, etc.

For example, in an anomaly detection system equipped with a learning device that performs machine learning to detect abnormalities in equipment, if it is determined that the learning device has little learning data, a technology is proposed to generate learning data. (See Patent Document 1). In the technology described in Patent Document 1, when it is determined that the learning data for the learning device to learn is small, data is generated by simulation or learning data is generated by adding noise etc. to the actual operation data, and the learning data is generated. I am letting the device learn.

JP2019-133212A

However, in the technique described in Patent Document 1, learning is performed using learning data generated by simulation and learning data generated by adding noise or the like to actual operation data. Therefore, the generated learning data may not match the actual data, and it is not always possible to accurately detect abnormalities using the generated learning data.

The present disclosure has been made in view of the above circumstances, and provides a learning data generation device, an anomaly detection device, an anomaly detection system, and an anomaly detection system that can accurately detect anomalies even when using generated learning data. The purpose is to provide an anomaly detection program.

In order to achieve the above object, an abnormality detection device according to a first aspect includes an acquisition unit that acquires a measurement value obtained by measuring the elements for each element group in an element group including a plurality of elements on a wafer; Input the measurement values related to the element group into a learning device that has performed machine learning to identify whether the element group is an abnormal element group that contains abnormal elements, and determine whether the element group is an abnormal element group. and a detection section that detects an abnormal element from among the elements included in the abnormal element group.

Further, in the abnormality detection device according to the second aspect, the detection unit detects an abnormal element from the abnormal element group using the number of times the element is included in the abnormal element group, or measures the element related to the abnormal element group. An abnormal element is detected from the abnormal element group using the value.

Further, in the abnormality detection generation device according to the third aspect, the element group may include elements adjacent to each other vertically and horizontally on the wafer, elements adjacent to the wafer in a convex shape, elements adjacent to the wafer in a hook shape, or elements adjacent to the wafer in a hook shape. Any one of the elements located at the same distance from the center.

Further, in the abnormality detection device according to the fourth aspect, the abnormality detection device acquires the measured values of the elements related to the element group as learning data, and uses the acquired learning data to determine whether the elements are in the abnormal element group. The learning device further includes a learning unit that performs machine learning for determination on the learning device.

On the other hand, in order to achieve the above object, in a learning data generation device according to a fifth aspect, element group information, which is information of the element group, is acquired for each element group in an element group including a plurality of elements on a wafer. The device includes an element group information acquisition unit, and a learning data generation unit that generates learning data by increasing the amount by changing the arrangement of elements according to the acquired element group information.

Further, in the learning data generation device according to the sixth aspect, the learning data generation unit rotates or reverses the position of the element related to the learning data, changes the arrangement of the element related to the learning data, and increases the amount of learning data. do.

Meanwhile, in order to achieve the above object, in an anomaly detection system according to a seventh aspect, an abnormality detection device according to the first to fourth aspects and learning data according to the fifth aspect or the sixth aspect are provided. A generating device.

On the other hand, in order to achieve the above object, in the abnormality detection program according to the eighth aspect, an acquisition step of acquiring measured values of the elements for each element group in an element group including a plurality of elements on a wafer; Input the measurement values related to the element group into a learning device that performs machine learning to identify whether the element group is an abnormal element group that contains abnormal elements, and confirm that the element group is an abnormal element group. A computer is caused to execute a determination step of determining whether or not the abnormal element is present, and a detection step of detecting an abnormal element from among the elements included in the abnormal element group.

According to the anomaly detection device of the first aspect, the anomaly detection system of the seventh aspect, and the anomaly detection program of the eighth aspect, anomalies can be detected with high accuracy even when using generated learning data. .

Moreover, according to the abnormality detection device of the second aspect, since information of a plurality of elements is compared, an abnormal element can be detected with high accuracy.

Moreover, according to the abnormality detection apparatus of the third aspect, it is possible to more uniformly obtain the measured values of the elements in the element group, regardless of the arrangement of the elements on the wafer.

Moreover, according to the abnormality detection device of the fourth aspect, it is possible to cause the learning device to learn the acquired measurement values.

Further, according to the learning data generation device of the fifth aspect, it is possible to generate various learning data in accordance with measured values.

Further, according to the learning data generation device of the sixth aspect, it is possible to easily increase the amount of learning data based on the measured values of the elements.

FIG. 1 is a block diagram showing an example of the configuration of an abnormality detection system according to each embodiment. FIG. 1 is a block diagram showing an example of the configuration of an abnormality detection device and a learning data generation device according to each embodiment. FIG. 2 is a block diagram showing an example of a functional configuration of an abnormality detection device and a learning data generation device according to each embodiment. FIG. 2 is a block diagram showing an example of an element group according to each embodiment. FIG. 2 is a block diagram showing an example of an element group according to each embodiment. FIG. 3 is a block diagram showing an example of an element group for explaining position dependence according to each embodiment. FIG. 3 is a diagram showing an example of measured values of an element for explaining detection of an abnormal element according to each embodiment. FIG. 1 is a schematic diagram showing an example of a neural network according to each embodiment. FIG. 2 is a block diagram illustrating an example of an element group for explaining generation of learning data according to each embodiment. It is a flowchart which shows an example of abnormality detection processing concerning a 1st embodiment. FIG. 3 is a sequence diagram illustrating an example of learning processing according to each embodiment. FIG. 7 is a diagram illustrating an example of a list of abnormal element groups for explaining detection of abnormal elements according to the second embodiment. It is a flow chart which shows an example of abnormality detection processing concerning a 2nd embodiment.

[First embodiment]
Hereinafter, embodiments for implementing the technology of the present disclosure will be described in detail with reference to the drawings. Note that the present disclosure describes an abnormality detection system 1 that detects an abnormality in an element such as a semiconductor chip. However, it is not limited to this. The element may be a sensor, a device, etc., and the abnormality detection system 1 may be used to detect an abnormality in any device or equipment.

FIG. 1 is a block diagram showing an example of the configuration of an abnormality detection system 1 according to the present embodiment. As shown in FIG. 1, an anomaly detection system 1 includes an anomaly detection device 2 and a learning data generation device 3. Note that the abnormality detection device 2 and learning data generation device 3 according to the present embodiment will be described as a server such as a personal computer or a terminal. The abnormality detection device 2 and the learning data generation device 3 may be terminals. Further, a case will be described in which the abnormality detection device 2 and the learning data generation device 3 according to this embodiment are each separate devices. However, it is not limited to this. The learning data generation device 3 may be installed in the abnormality detection device 2, or the abnormality detection device 2 and the learning data generation device 3 may be installed in other devices.

The abnormality detection device 2 includes a learning model for determining whether or not an element on a wafer is a normal element, and detects an abnormal element by inputting a measured value of the element into the learning model. The learning data generation device 3 generates learning data for the learning model to perform machine learning. Here, the learning model is an example of a "learning device."

The abnormality detection device 2 and the learning data generation device 3 are connected to each other via a network N. The learning data generating device 3 acquires the measured values of the elements from the anomaly detecting device 2 and generates learning data using the measured values, and the anomaly detecting device 2 acquires the learning data from the learning data generating device 3 and generates learning data using the measured values. , perform machine learning of the learning model using the training data.

Next, with reference to FIG. 2, the configurations of the abnormality detection device 2 and the learning data generation device 3 will be described. FIG. 2 is a block diagram showing an example of the configuration of the abnormality detection device 2 and the learning data generation device 3. As shown in FIG. 2, the abnormality detection device 2 according to the present embodiment includes a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, a storage 14, an input unit 15, and a monitor. 16, and a communication interface (communication I/F) 17. Each of the CPU 11, ROM 12, RAM 13, storage 14, input section 15, monitor 16, and communication I/F 17 is interconnected by a bus 18.

The CPU 11 oversees and controls the entire abnormality detection device 2. The ROM 12 stores various programs and data, including an abnormality detection program and a learning program used in this embodiment. The RAM 13 is a memory used as a work area when various programs are executed. The CPU 11 performs processing for detecting abnormal elements by loading a program stored in the ROM 12 into the RAM 13 and executing it. The storage 14 is, for example, an HDD (Hard Disk Drive), an SSD (Solid State Drive), or a flash memory. Note that the storage 14 may store an abnormality detection program, a learning program, and the like. The input unit 15 is a mouse and keyboard for inputting characters and the like. The monitor 16 is a monitor that displays image data, characters, and the like. The communication I/F 17 transmits and receives data to and from the learning data generation device 3 via the network N.

Similarly, the learning data generation device 3 according to the present embodiment includes a CPU 21, a ROM 22, a RAM 23, a storage 24, an input section 25, a monitor 26, and a communication interface (communication I/F) 27. Each of the CPU 21, ROM 22, RAM 23, storage 24, input section 25, monitor 26, and communication I/F 27 is interconnected by a bus 28.

The CPU 21 oversees and controls the entire learning data generation device 3. The ROM 22 stores various programs and data, including the learning data generation program used in this embodiment. The RAM 23 is a memory used as a work area when various programs are executed. The CPU 21 performs a process of generating learning data by loading a program stored in the ROM 22 into the RAM 23 and executing it. The storage 24 is, for example, an HDD, SSD, or flash memory. Note that the storage 24 may store a learning data generation program and the like. The input unit 25 is a mouse and keyboard for inputting characters and the like. The monitor 26 is a monitor that displays image data, characters, and the like. The communication I/F 27 transmits and receives data to and from the abnormality detection device 2 via the network N.

Next, with reference to FIG. 3, the functional configurations of the abnormality detection device 2 and the learning data generation device 3 will be described. FIG. 3 is a block diagram showing an example of the functional configuration of the abnormality detection device 2 and the learning data generation device 3 according to the present embodiment.

As shown in FIG. 3, the abnormality detection device 2 includes an acquisition section 31, a determination section 32, a detection section 33, a storage section 34, and a learning section 35. The CPU 11 functions as an acquisition section 31, a determination section 32, a detection section 33, a storage section 34, and a learning section 35 by executing the abnormality detection program and the learning program. Further, the learning data generation device 3 includes an element group information acquisition section 36 and a learning data generation section 37. By executing the learning data generation program, the CPU 21 functions as the element group information acquisition section 36 and the learning data generation section 37.

The acquisition unit 31 sets a plurality of elements (hereinafter referred to as "element group") on one wafer as one set, and acquires the measured values of each element for each element group. Note that in this embodiment, a mode will be described in which the acquisition unit 31 acquires a value obtained by measuring leakage current of an element as a measurement value. However, it is not limited to this. The measured value may be a gate oxide film breakdown voltage, a threshold voltage, or the like. Further, in this embodiment, a mode will be described in which an abnormal element is detected using a single measurement value such as leakage current of the element. However, it is not limited to this. A plurality of measured values such as leakage current and threshold voltage of the element may be used to detect an abnormal element. For example, multiple measurements of device leakage current and threshold voltage may be used to detect abnormal devices at each measurement.

Further, in this embodiment, as shown in FIG. 4 as an example, a configuration will be described in which the element group 40 is composed of a plurality of adjacent elements uniformly arranged in the vertical and horizontal directions on the wafer 41. However, it is not limited to this. For example, as shown in FIG. 5, a plurality of elements adjacent in a convex shape are grouped into an element group 42, a plurality of elements adjacent in a hook shape (L-shape) are grouped into an element group 43, or a plurality of elements adjacent in an S-shape are grouped into an element group 43. may be used as the element group 44. Furthermore, as shown in FIG. 6 as an example, a plurality of elements located at the same distance from the center of the wafer may be used as an element group. Here, in FIG. 6, the same symbols are attached to elements located at the same distance in order of decreasing distance from the center. For example, in FIG. 6, the plurality of elements labeled with the symbol "A" indicate the element closest to the center of the wafer.

Regarding the elements of the element group, adjacent elements on the wafer have similar element structures (film configurations within the elements, etc.), so if the elements are normal, the measured values will be close. Therefore, by eliminating the influence of fluctuations in measured values related to the position of elements on the wafer for elements related to the element group, slight differences in measured values due to abnormalities can be extracted and abnormal elements can be detected. . Furthermore, measured values of elements on the wafer may have dependence on the position of the element on the wafer, such as, for example, threshold voltage. The threshold voltage depends on the oxide film capacitance of the device, and the oxide film capacitance depends on the oxide film thickness. The oxide film is formed by thermal oxidation, CVD (Chemical Vapor Deposition), or the like. The thickness of the oxide film formed by this method changes from the center of the wafer to the edges. Therefore, the oxide film capacitance changes from the center to the edge of the wafer. Therefore, due to variations in the process within the wafer surface (oxide film thickness, dose concentration, processed shape, etc.), characteristics such as the threshold voltage of the element change depending on the position of the element, as shown in FIG. 7 as an example. Therefore, if the variation in the measured value due to the abnormality is within the range of the variation in the measured value due to the process, there is a possibility that the abnormality may be overlooked based on the measured value of a single element. On the other hand, adjacent elements have little variation in wafer oxide film thickness, etc., and have similar characteristics, such as threshold voltage. It is possible to highlight anomalies as in Therefore, it is desirable that the element group be constructed using adjacent elements.

Further, since the structure of the elements on the wafer, such as the oxide film thickness, changes depending on the position from the center of the wafer, as described above, the characteristics of the elements, such as the threshold voltage, are affected by the position of the element. In other words, the characteristics of elements located at the same distance from the center of the wafer are approximately the same. Therefore, by configuring an element group using elements that are at the same distance from the center of the wafer, slight differences in measured values due to anomalies can be extracted, similar to the element groups related to adjacent elements described above, and abnormal elements can be detected. Detected. Therefore, the element group may be constructed using elements located at the same distance from the center of the wafer.

The determination unit 32 uses the measured values for each element group acquired by the acquisition unit 31 to determine whether the element group includes an abnormal element. Note that hereinafter, an element group including an abnormal element will be referred to as an "abnormal element group."

Specifically, the determination unit 32 is a learning model using a neural network that performs machine learning to determine whether or not the element group is an abnormal element group from the measured values related to the element group. As shown in FIG. 8 as an example, the neural network is composed of nodes 50 of a plurality of layers that perform processing such as an input layer, an intermediate layer (hidden layer), and an output layer. In the neural network, nodes 50 of each layer are connected by edges 51, and data processed by each node 50 is propagated to the input layer, intermediate layer, and output layer in this order to perform processing.

As shown in FIG. 8, the number of nodes 50 in the input layer of the neural network according to this embodiment is the same as the number of elements in the element group. Further, each node 50 of the input layer receives input of a measurement value while distinguishing the arrangement of each element related to the input measurement value. For example, when each node 50 of the input layer receives input of measured values related to the element group 40, the node 50A receives the measured value of the element "1", the node 50B receives the measured value of the element "2", and the node 50C receives the measured value of the element "2". The measured value of element "5" and node 50D are accepted as the measured value of element "6". In other words, by exchanging the measured values input to the node 50 in the input layer, it is possible to perform processing as input of measured values of different element groups in which the arrangement of each element is changed.

The detection unit 33 shown in FIG. 3 detects abnormal elements from the element group determined by the determination unit 32 to be an abnormal element group. Specifically, the detection unit 33 derives the standard deviation of the measured values of the elements for each element group determined to be an abnormal element group, and detects the abnormal element using the standard deviation. For example, the detection unit 33 sets a value obtained by adding the standard deviation to the average value of the measured values of the elements related to the element group as an upper limit, and a value obtained by subtracting the standard deviation from the average value as a lower limit. Set. The detection unit 33 detects an element whose measured value does not fall within the variation range (exceeds the upper limit or falls below the lower limit) as an abnormal element.

The storage unit 34 shown in FIG. 3 stores measured values of elements related to the element group as information related to the element group (hereinafter referred to as “element group information”) and results determined by the determination unit 32 (hereinafter referred to as “determination results”). ) and are stored in association with each other. Specifically, when the determination unit 32 determines that the element group is abnormal, the storage unit 34 stores the measured values of the elements related to the element group as element group information and the “abnormal element group” as the determination result. , are stored in association with each other. Furthermore, when the determination unit 32 determines that the element group is normal, the storage unit 34 associates the measured values of the elements related to the element group as element group information with the “normal element group” that is the determination result. memorize it.

The learning unit 35 causes the learning model in the determining unit 32 to learn the element group information stored in the storage unit 34 and learning data generated by a learning data generating unit 37, which will be described later, as learning data. Specifically, the learning unit 35 performs supervised learning on the learning model using the measured values of the elements related to the element group as input data and the determination results as correct data, and calculates the measured values of the elements related to the element group. Perform machine learning to determine abnormal element groups from

The element group information acquisition unit 36 acquires element group information to be increased from the storage unit 34 described above. Specifically, the element group information acquisition unit 36 acquires element group information whose determination result is “abnormal element group” from the storage unit 34 as the learning data to be increased.

The learning data generation unit 37 increases the amount of element group information acquired by the element group information acquisition unit 36 as learning data. Specifically, the learning data generation unit 37 rotates and inverts the acquired element group information to change the arrangement of elements related to the element group. As described above, the measured values of the elements related to the element group according to this embodiment are distinguished depending on the arrangement of the elements related to the element group. For example, as shown in FIG. 9 as an example, the learning data generation unit 37 changes the arrangement of elements in the element group by rotating the element group 40 counterclockwise by 90 degrees, 180 degrees, and 270 degrees. The learning data generation unit 37 also inverts the element group 40 around the horizontal direction, around the vertical direction, around a diagonal line that goes downward to the right, and around a diagonal line that goes up to the right. The arrangement of elements related to the element group is changed by . The learning data generation unit 37 increases the amount of one learning data obtained from the element group that is the "abnormal element group" to eight pieces of learning data by rotating and inverting the elements related to the element group.

Next, with reference to FIGS. 10 and 11, the operation of the abnormality detection device 2 and learning data generation device 3 according to this embodiment will be described. FIG. 10 is a flowchart illustrating an example of the abnormality detection process according to the first embodiment. When the CPU 11 reads out and executes the abnormality detection program from the ROM 12 or the storage 14, the abnormality detection process shown in FIG. 10 is executed. The abnormality detection process shown in FIG. 10 is executed by inputting the execution instruction to the abnormality detection device 2 when the user inputs an execution instruction, for example.

In step S101, the CPU 11 acquires measured values of elements related to the element group.

In step S102, the CPU 11 executes a determination of the element group using the measured values of the elements related to the element group.

In step S103, the CPU 11 determines whether the determined element group is an abnormal element group. If the determined element group is an abnormal element group (step S103: YES), the CPU 11 moves to step S104. On the other hand, if the determined element group is not an abnormal element group (step S103: NO), the CPU 11 moves to step S106.

In step S104, the CPU 11 uses the element group data to derive the average value and standard deviation of the measured values of the elements related to the element group.

In step S105, the CPU 11 detects an abnormal element from the element group determined to be an abnormal element group.

In step S106, the CPU 11 determines whether the determination has been made for all element groups. If the determination has been made for all element groups (step S106: YES), the CPU 11 moves to step S107. On the other hand, if the determination has not been made for all the element groups (step S106: NO), the CPU 11 moves to step S109.

In step S107, the CPU 11 stores the measured values of the elements related to the element group and the determination results in association with each other as element group information.

In step S108, the CPU 11 outputs the detection result. Here, when an abnormal element is detected, information to the effect that an abnormal element has been detected and information indicating the abnormal element is output.

In step S109, the CPU 11 acquires the measured values of the elements of the next element group, and proceeds to step S102.

Next, with reference to FIG. 11, the operation of the learning process of the anomaly detection system in which the anomaly detection device 2 and the learning data generation device 3 cooperate will be described. FIG. 11 is a sequence diagram showing an example of the flow of the learning process of the abnormality detection system according to the present embodiment.

In step S201, the abnormality detection device 2 transmits an instruction to generate learning data and element group information.

In step S202, the learning data generation device 3 acquires element group information from the abnormality detection device 2.

In step S203, the learning data generation device 3 increases the amount of the acquired element group information as learning data.

In step S204, the learning data generation device 3 transmits the increased amount of learning data to the abnormality detection device 2.

In step S205, the abnormality detection device 2 acquires the increased amount of learning data from the learning data generation device 3.

In step S206, the abnormality detection device 2 performs machine learning on the learning model using the stored element group information and the increased learning data acquired from the learning data generation device 3 as learning data. .

As described above, according to this embodiment, an abnormality can be detected with high accuracy even when using generated learning data.

[Second embodiment]
In the first embodiment, a mode has been described in which an abnormal element is detected using the standard deviation of the measured values of the elements in the element group. In this embodiment, a mode will be described in which an abnormal element is detected using the number of times it is determined that the element is an abnormal element group.

The configuration of the anomaly detection system 1 according to this embodiment (see FIG. 1), the hardware configuration of the anomaly detection device 2 and the learning data generation device 3 (see FIG. 2), and the functional configuration of the anomaly detection system 1 (see FIG. 2) (see FIG. 3) is the same as the first embodiment, so the explanation will be omitted. Further, a block diagram showing a group of elements according to this embodiment (see FIGS. 4, 5, 6, and 9), a graph of measured values (see FIG. 7), a neural network (see FIG. 8), and a learning process The sequence diagram (see FIG. 11) showing this is the same as that in the first embodiment, so the explanation thereof will be omitted.

In this embodiment, a mode will be described in which the detection unit 33 shown in FIG. 3 counts the number of times each element is included in the abnormal element group and detects an abnormal element.

For example, as shown in FIG. 12 as an example, the detection unit 33 stores in a list the abnormal element numbers and element numbers related to the element group determined to be the abnormal element group. Here, the abnormal element group number is a number for identifying an element group, and the element number is a number for identifying an element.

The detection unit 33 stores the abnormal element group, counts the number of times each element is included in the list for each element included in the abnormal element group, and determines, for each element, the degree of abnormal element (hereinafter referred to as (referred to as "abnormality level"). For example, as shown in FIG. 12, since the element with element number 6 is included in the abnormal element group four times, the abnormality level is set to "4". Furthermore, since the element with element number 5 is included in the abnormal element group twice, the abnormality level is set to "2". In this way, the degree of abnormality is set for each element according to the number of times counted, and an element whose degree of abnormality exceeds a threshold is detected as an abnormal element. For example, when the abnormality degree threshold is "2", element number 6 with an abnormality degree of "3" or higher is detected as an abnormal element.

Next, with reference to FIG. 13, the operation of the abnormality detection device 2 according to this embodiment will be explained. FIG. 13 is a flowchart illustrating an example of abnormality detection processing according to the second embodiment. When the CPU 11 reads out and executes the abnormality detection program from the ROM 12 or the storage 14, the abnormality detection process shown in FIG. 13 is executed. The abnormality detection process shown in FIG. 13 is executed by inputting the execution instruction to the abnormality detection device 2 when the user inputs an execution instruction, for example. Note that the steps in FIG. 13 that are the same as the abnormality detection process shown in FIG. 10 are given the same reference numerals as in FIG. 10, and the description thereof will be omitted.

In step S110, the CPU 11 determines whether the determined element group is an abnormal element group. If the determined element group is an abnormal element group (step S110: YES), the CPU 11 moves to step S111. On the other hand, if the determined element group is not an abnormal element group (step S110: NO), the CPU 11 moves to step S106.

In step S111, the CPU 11 stores the element group number of the element group and the element numbers of the elements related to the element group in a list as the abnormal element group.

In step S112, the CPU 11 determines whether an abnormal element group is stored in the list. If the abnormal element group is stored (step S112: YES), the CPU 11 moves to step S113. On the other hand, if the abnormal element group is not stored (step S112: NO), the CPU 11 moves to step S107.

In step S113, the CPU 11 derives the degree of abnormality of the elements related to the element group stored in the list.

In step S114, the CPU 11 detects an element whose degree of abnormality exceeds the threshold value of the degree of abnormality as an abnormal element.

As described above, according to the present embodiment, an abnormal element is detected from the number of times the element is counted as belonging to an abnormal element group. As explained above, according to this embodiment, the same effects as in the first embodiment can be achieved.

Note that the element group according to this embodiment includes a plurality of elements adjacent in the vertical and horizontal directions, a plurality of elements adjacent in a convex shape, a plurality of elements adjacent in a hook shape (L-shape), and a plurality of elements adjacent in an S-shape. A configuration has been described in which the device is any one of a plurality of devices located at the same distance on the wafer. However, it is not limited to this. Multiple elements adjacent in the vertical and horizontal directions, multiple elements adjacent in a convex shape, multiple elements adjacent in a hook shape (L-shape), multiple elements adjacent in an S-shape, and located at the same distance on the wafer. Two or more of the plurality of elements may be set as an element group. For example, for an element located near the center of the wafer, elements adjacent vertically and horizontally are set as an element group, and for an element located near the edge of the wafer, multiple elements adjacent in a convex shape are set. It may also be set as an element group.

Furthermore, in the present embodiment, a mode has been described in which learning data is generated when performing learning processing. However, it is not limited to this. When the number of element group information related to abnormal element groups stored in the storage unit 34 is less than a predetermined number, generation of learning data may be executed.

(Explanation of the effect of increasing data)
For example, when increasing the amount of data related to device abnormalities due to manufacturing defects in semiconductor devices, there are 400 devices in one wafer (20 vertically x 20 horizontally), and the data for 25 wafers is used as learning data. Assume that data for a certain 10,000 elements is held.

When the elements constituting an element group are four elements arranged uniformly in the vertical and horizontal directions, the number of element group data that can be obtained from one wafer is 19 vertically x 19 horizontally = 361 pieces, which can be obtained from 25 wafers. The number of data in the element group is 361×25=9025.

Also, suppose that 300 abnormal elements are included out of 10,000 elements, and one abnormal element group contains one abnormal element, then there are 4 abnormal element groups for one abnormal element. Since there are several ways, the number of data is 300 pieces x 4 ways = 1200 pieces. Furthermore, by inverting or rotating the elements related to the abnormal element group, the number of data in the abnormal element group becomes 1200×8=9600.

On the other hand, since the abnormal element group is 1200 out of 9025 element groups, the normal element group is 9025 - 1200 = 7825.

Therefore, according to the data increase method according to the present invention, it is possible to increase the number of data of an abnormal element group to 9,600 compared to 7,825 data of a normal element group. It is possible to learn the data and the data of the abnormal element group.

(others)
In addition, the configurations of the abnormality detection device 2 and the learning data generation device 3 described in the above embodiments are merely examples, and may be changed according to the situation without departing from the spirit of the invention.

Furthermore, the process flow of the program described in the above embodiment is only an example, and unnecessary steps may be deleted, new steps may be added, or the processing order may be changed without departing from the main purpose. good.

Note that various processors other than the CPU may execute the abnormality detection process, learning data generation process, and learning process that are executed by the CPU reading the software (program) in each of the above embodiments. In this case, the processor includes a PLD (Programmable Logic Device) whose circuit configuration can be changed after manufacturing, such as an FPGA (Field Programmable Gate Array), and an ASIC (Application Specific Integrated Circuit). Dedicated to execute specific processing such as An example is a dedicated electric circuit that is a processor having a circuit configuration designed to. In addition, the process of extracting, analyzing, and generating user information may be executed by one of these various processors, or by a combination of two or more processors of the same or different types (for example, multiple FPGAs). , a combination of a CPU and an FPGA, etc.). Further, the hardware structure of these various processors is, more specifically, an electric circuit that is a combination of circuit elements such as semiconductor elements.

Further, in each of the above embodiments, a mode has been described in which the abnormality detection program is stored (installed) in the ROM 12 or the storage 14 in advance, but the present invention is not limited to this. The program is recorded on recording media such as CD-ROM (Compact Disc Read Only Memory), DVD-ROM (Digital Versatile Disk Read Only Memory), and USB (Universal Serial Bus) memory. It may also be provided in a different format. Further, the program may be downloaded from an external device via a network.

11, 21 CPU
12, 22 ROM
13, 23 RAM
14, 24 Storage 15, 25 Input section 16, 26 Monitor 17, 27 Communication I/F
18, 28 Bus 31 Acquisition unit 32 Determination unit 33 Detection unit 34 Storage unit 35 Learning unit 36 Element group information acquisition unit 37 Learning data generation unit 40, 42, 43, 44 Element group 41 Wafer 50, 50A, 50B, 50C, 50D node 51 edge

Claims (8)

an acquisition unit that acquires measured values of the elements for each element group in an element group including a plurality of elements on a wafer;
The measurement values related to the element group are input into a learning device that performs machine learning to identify whether or not the element group is an abnormal element group that includes an abnormal element. a determination unit that determines whether it is an abnormal element group;
a detection unit that detects the abnormal element from the elements included in the abnormal element group;
An anomaly detection device comprising: The detection unit detects the abnormal element from the abnormal element group using the number of times the element is included in the abnormal element group, or detects the abnormal element from the abnormal element group using the measured value of the element related to the abnormal element group. The abnormality detection device according to claim 1, wherein the abnormal element is detected from a group. The element group includes elements adjacent to each other in the vertical and horizontal directions on the wafer, elements adjacent to the wafer in a convex shape, elements adjacent to the wafer in a hook shape, or elements located at the same distance from the center of the wafer. The abnormality detection device according to claim 1 or 2, which is any one of the following. The abnormality detection device acquires measured values of elements related to the element group as learning data, and uses the acquired learning data to perform machine learning on the learning device to determine whether the element group is an abnormal element group. The abnormality detection device according to any one of claims 1 to 3, further comprising a learning section. an element group information acquisition unit that acquires element group information that is element group information for each element group in an element group including a plurality of elements on a wafer;
a learning data generation unit that generates learning data by increasing the amount by changing the arrangement of the elements according to the acquired element group information;
A learning data generation device comprising: The learning according to claim 5, wherein the learning data generation unit rotates or reverses the position of the element according to the learning data, changes the arrangement of the element according to the learning data, and increases the amount of the learning data. Data generation device. An abnormality detection device according to any one of claims 1 to 4,
A learning data generation device according to claim 5 or 6;
Anomaly detection system equipped with an acquisition step of acquiring measured values of the elements for each element group in an element group including a plurality of elements on the wafer;
The measurement values related to the element group are input into a learning device that performs machine learning to identify whether or not the element group is an abnormal element group that includes an abnormal element. a determination step of determining whether it is an abnormal element group;
a detection step of detecting the abnormal element from elements included in the abnormal element group;
An anomaly detection program that causes a computer to execute