WO2025158668A1 - Recipe creation device and recipe creation tool integration device - Google Patents

Abstract

The present invention relates to a recipe creation device 130 for creating a recipe for causing a system that performs inspection or measurement to automatically perform inspection or measurement. The recipe creation device 130 includes: an image classification execution unit 403 that obtains an image classification code of first image data by using an image classification model 140; a cryptographic conversion unit 405 that converts the image classification code of the first image data into a first cryptographic code by using an irreversible cryptographic method; and a recipe search unit 407 that extracts a recommended recipe by collating the first cryptographic code with a recipe database 150. The image classification model has been trained with training image data, and the first image data and the training image data have been acquired under the same default conditions. Furthermore, the recipe database has registered therein: cryptographic codes obtained by converting image classification codes by using the cryptographic method in the cryptographic conversion unit; and recommended recipes obtained by deleting unique information from existing recipes for training image data classified into the image classification codes.

Description

Recipe creation device and recipe creation tool integration device

This disclosure relates to a recipe creation device that creates recipes for automatic inspection or automatic measurement of semiconductor devices, etc., and a recipe creation tool integration device that integrates recipe creation tools created by the recipe creation device.

In recent years, semiconductor devices have become increasingly miniaturized and multi-layered, making their manufacturing processes extremely complex. For this reason, in order to improve the yield of semiconductor device production lines and ensure stable operation, it is important to be able to quickly and accurately inspect for defects caused by the manufacturing process.

Semiconductor device production lines are equipped with review SEMs (Scanning Electron Microscopes), which review defects based on coordinate information of defects detected by optical inspection equipment, and critical dimension SEMs (Critical Dimension-SEMs), which measure the dimensions of patterns, and are used for detailed inspection of these defects and measurement of patterns. In order to perform inspection and measurement accurately, it is necessary to acquire SEM images with the desired image quality. The desired image quality differs depending on, for example, whether it is desired to inspect the edge or flat surface of the circuit pattern image captured in the SEM image. For this reason, workers need to rely on their experience to adjust the numerous parameters that affect the image quality of the circuit pattern image each time they perform inspection or measurement.

Adjusting the parameters of an inspection system is a very time-consuming task, and for example, Patent Document 1 discloses technology that improves the efficiency of such parameter adjustment (recipe development).

Special Publication No. 2015-514311

As such, creating recipes takes time, hindering improvements in production efficiency. Furthermore, many parameters are adjusted by workers (inline engineers) relying on their experience, making recipe creation work highly dependent on individual skills, and a shortage of engineers can become a bottleneck when ramping up production. In particular, the image data contained in recipes contains the user's IP (Intellectual Property, hereafter referred to as User IP), such as circuit layout information, making it difficult for anyone other than the user to share it.

A recipe creation device according to one embodiment of the present disclosure acquires image data of a circuit pattern on a wafer using a charged particle beam device, and creates a recipe for automatically inspecting or measuring the circuit pattern from the image data. The recipe creation device includes an image classification execution unit that uses an image classification model to determine an image classification code for first image data acquired by the charged particle beam device according to a first initial recipe including default conditions, a cryptography conversion unit that converts the image classification code of the first image data into a first cryptography code using a non-reversible cryptography method, and a recipe search unit that compares the first cryptography code with a recipe database to extract a recommended recipe. The image classification model is trained using training image data, and the training image data is image data acquired by the charged particle beam device according to a second initial recipe. The second initial recipe is a recipe obtained by modifying an existing recipe created for a specified inspection or measurement to include default conditions. The recipe database registers the cryptography code obtained by converting the image classification code using the cryptography conversion unit, and a recommended recipe obtained by removing unique information from an existing recipe for the training image data classified with the image classification code.

This shortens the time required to create recipes, improves production line utilization, and reduces product costs. Other issues and novel features will become apparent from the description of this specification and the accompanying drawings.

FIG. 1 is a diagram illustrating an overview of a semiconductor inspection system. 1 is an example of a hardware configuration of an information processing device (computer). 1 is a flowchart of an inspection performed by a semiconductor inspection system. 1 shows a conventional inspection recipe creation flow. 10 is an example of a recipe creation screen. FIG. 10 is a functional block diagram of the recipe creation device (operation phase). 1 is a flowchart showing an inspection recipe creation flow according to the first embodiment; 10 is an example of a recipe creation screen in the first embodiment. FIG. 10 is a diagram illustrating an image classification process. FIG. 10 is a diagram illustrating an image classification process. 1 is an example of a data structure of a recipe database. FIG. 10 is a functional block diagram of a recipe creation device (preparation phase). 10 is a flow chart showing a process for creating an image classification database and a recipe database according to the first embodiment; 10 is an example of a learning screen according to the first embodiment. FIG. 2 is a functional block diagram of the recipe creation device. FIG. 1 illustrates a network connection of multiple semiconductor inspection systems and recipe creation tool integration devices. FIG. 2 is a functional block diagram of a recipe creation tool integration device. 10 is an example of an associative learning screen according to the second embodiment. 13 is an example of a recipe creation tool update screen according to the second embodiment. FIG. 10 is a diagram for explaining version management of a recipe creation tool.

Below, an embodiment of the present disclosure will be described with reference to the drawings.

FIG. 1A is a diagram showing an overview of a semiconductor inspection system, which is an example of a system for inspecting or measuring wafers according to the present disclosure. The semiconductor inspection system 801 acquires image data of a circuit pattern on a wafer and analyzes the acquired image data to inspect the circuit pattern. The semiconductor inspection system 801 has an imaging unit 101, a control unit 120, and a recipe creation device 130. The imaging unit 101 and control unit 120 constitute a charged particle beam device that acquires image data and performs inspection; here, an example is shown in which a scanning electron microscope (SEM) is used. Note that semiconductor measurement systems that measure circuit patterns have a similar configuration. Furthermore, the object to be inspected is not limited to semiconductor wafers.

In the imaging unit 101, a primary electron beam 105 emitted from an electron source 102 is accelerated by an acceleration electrode 103 to a desired acceleration voltage value and irradiated onto a sample 109 such as a wafer. Furthermore, when inspecting the side walls or bottom of a circuit pattern, the control unit 120 controls the tilt deflector 107 with a control signal 113, tilting the primary electron beam 105 to irradiate the sample 109. In addition to controlling the irradiation angle of the primary electron beam (tilt control), methods for photographing the side walls of a circuit pattern also include tilting the stage 110 (stage control).

Irradiation with the primary electron beam 105 causes secondary electrons (SE) and backscattered electrons (BSE) to be emitted from the sample 109 as signal electrons. The SEs are detected by the SE detector 112, and the BSEs are detected by the BSE detector 108, and are converted into a digital signal 114 by the A/D converter 111. The digital signal 114 is input to the control unit 120 and stored in the memory 122.

In the control unit 120, a CPU (Central Processing Unit) 121 and image processing hardware 123 perform image processing according to the purpose, and the circuit pattern on the wafer is inspected. Examples of image processing hardware 123 include an ASIC (Application Specific Integrated Circuit), FPGA (Field Programmable Gate Array), DSP (Digital Signal Processor), and GPU (Graphical Processing Unit).

In order for the semiconductor inspection system 801 to perform inspections properly, appropriate parameters must be set for the imaging unit 101 and control unit 120. The recipe creation device 130 is a device that creates an inspection recipe that defines these parameters in accordance with the circuit pattern to be inspected.

The recipe creation device 130 is realized by an information processing device (computer) 10, which primarily includes a processor (CPU) 11, memory 12, storage device 13, input interface (I/F) 14, output I/F 15, communication I/F 16, and bus 17, as shown in FIG. 1B. The processor 11 functions as a functional unit that provides specified functions by executing processes according to programs loaded into memory 12. The storage device 13 stores data and programs used by the functional unit. The input I/F 14 is connected to input devices such as a keyboard, pointing device, and operation panel, and the output I/F 15 is connected to a display device. The communication device I/F 16 enables communication with other information processing devices via a network. These are connected to each other via bus 17 so that they can communicate with each other. The recipe creation device 130 does not have to be realized by a single information processing device, but may be realized by multiple information processing devices. It may also be virtualized. Furthermore, some or all of the functions of the recipe creation device 130 may be realized as a cloud application.

In the following explanation, when describing processing by a program, the program or functional units may be described as the main focus, but the main focus of the hardware in these cases is a processor, or an information processing device comprising such a processor. The information processing device executes processing in accordance with a program read into memory, using resources such as memory and communication interfaces as appropriate through the processor. While Figure 1B shows an example of a CPU as the processor, a GPU or the like may also be used. Furthermore, the processing to realize a function is not limited to software program processing, but can also be implemented using dedicated circuits. Dedicated circuits such as FPGAs and ASICs can be used.

2 is a flowchart of the inspection of sample 109 performed by semiconductor inspection system 801. Sample 109 is placed on imaging unit 101, and the inspection flow begins. First, the recipe creation device 130 performs imaging condition setting S01, which sets imaging conditions such as the number of imaging frames and scan direction and acquires a template image for alignment, and inspection condition setting S02, which sets inspection conditions related to the inspection location and inspection detection function. The various parameter values set by imaging condition setting S01 and inspection condition setting S02, and the template image for alignment, are called an inspection recipe. The inspection recipe created by recipe creation device 130 is sent to control unit 120, which then performs wafer imaging S03 and inspection S04 in accordance with the inspection recipe. For example, the wafer is imaged to acquire a circuit pattern image, and defect locations are identified from the circuit pattern image. Finally, inspection result output S05 is performed. For example, the identified defect locations are displayed on the GUI of control unit 120 together with image information and brightness information. In this way, an inspection recipe is created and the semiconductor inspection system operates according to the inspection recipe, making it possible to automatically inspect wafers one after another as they flow down the production line.

3A shows a conventional inspection recipe creation flow, and FIG. 3B shows a GUI for creating an inspection recipe using the flow of FIG. 3A. This GUI is displayed on the display device of the recipe creation device 130. First, the name of the recipe to be created is specified in the recipe name specification section 201 on the recipe creation screen 200, and then creation of the inspection recipe begins. First, an imaging position setting S11 is performed to set the position to be inspected on the sample, and a template image for alignment is acquired. The template image includes an OM template image for rough alignment at low magnification using an optical microscope, and an SEM template image for precise alignment at high magnification using an electron microscope. The OM alignment adjustment section 202 sets the acquisition conditions for the OM template image, and the set acquisition conditions are sent to the control section 120. The desired OM template image is acquired by the optical microscope (not shown) of the imaging section 101. Similarly, the SEM alignment adjustment section 203 sets the acquisition conditions for the SEM template image, and the set acquisition conditions are sent to the control section 120. The desired SEM template image is acquired by the imaging section 101.

Next, the optical conditions for imaging (S12) are set. The optical conditions for imaging are optical conditions set in the imaging unit 101, such as the scan speed, scan direction, and number of frames, for acquiring a circuit pattern image for inspection. These are set in the SEM imaging parameter setting section 206 on the recipe creation screen 200. The optical conditions for imaging (imaging parameters) may be set as numerical values or as a selected mode. The set optical conditions for imaging are sent to the control unit 120, and the imaging unit 101 performs wafer imaging (S13). The captured circuit pattern image is then sent to the recipe creation device 130. The operator creating the inspection recipe visually checks the image quality of the circuit pattern image displayed on the display device (S14) and determines whether it is satisfactory for the image quality to be used in inspection. If the image quality is determined to be insufficient, the process returns to the optical conditions for imaging (S12) and repeats steps S12 to S14 until a circuit pattern image with satisfactory image quality is obtained.

If satisfactory image quality is obtained in the visual image quality confirmation S14, the process proceeds to inspection condition setting S15, where inspection parameters such as the inspection location, inspection method, and defect detection accuracy are set. The inspection location is selected by selecting from a coordinate list that lists the coordinates of the inspection location in the inspection location designation section 204 on the recipe creation screen 200. Inspection parameters are selected in the defect detection function selection section 205 on the recipe creation screen 200. The set optical conditions for imaging (imaging parameters) and inspection conditions (inspection parameters) are sent to the control unit 120, which then performs inspection S16 and obtains inspection results such as defect locations and defect features from the circuit pattern image.

In the visual confirmation of inspection results S17, the operator visually confirms the inspection results obtained in the inspection S16, and returns to the imaging optical condition setting S12, repeating steps S12 to S17, until the expected results are achieved.

If satisfactory inspection results are obtained in visual confirmation of inspection results S17, the process proceeds to image quality improvement condition setting S18. In the semiconductor inspection system 801, an operator visually classifies defects from an improved image obtained by improving the image quality of a circuit pattern image. For this reason, it is desirable to improve the image quality of the circuit pattern image by adjusting the image quality so that the characteristics of the defects are clearly expressed. The image quality improvement conditions for the circuit pattern image are set in the image quality adjustment parameter adjustment unit 207 on the recipe creation screen 200. The set image quality improvement conditions are sent to the control unit 120, which then executes image quality improvement processing S19 for the circuit pattern image. Note that in this disclosure, the image quality improvement conditions are also treated as part of the inspection conditions (inspection parameters).

In visual confirmation of the improved image S20, the worker visually checks the improved image obtained in the image quality improvement process S19, and returns to image quality improvement condition setting S18 and repeats steps S18 to S20 until the worker is satisfied with the image quality. A slide bar is used to adjust the parameters in the image quality adjustment parameter adjustment unit 207 so that the worker can improve the image quality intuitively.

Finally, all parameter values set from setting the optical conditions for imaging (S12) to visually checking the improved image (S20) along with the template image are saved as an inspection recipe (S21).

As such, in conventional inspection recipe creation, it took a lot of time to determine appropriate parameters for setting the optical conditions for imaging (S12), setting the inspection conditions (S15), and setting the image quality improvement conditions (S18).

In contrast, the recipe creation device 130 of this embodiment is capable of extracting existing usable inspection recipe data based on image data acquired under predetermined default conditions and keywords attached to the image data, significantly reducing the time required for recipe creation work. Furthermore, the image data required for recipe creation is confidential information that contains circuit patterns, and it is necessary to prevent such information related to user IP from leaking to the outside. For this reason, the recipe creation device 130 of this embodiment conceals image classification data that indicates the tendency of the image data through encryption, and also deletes user-specific information from existing inspection recipes registered in the recipe database. Below, the functions of the recipe creation device 130 will be explained separately for the operation phase in which inspection recipes are created, and the preparation phase in which the image classification database and recipe database required for the operation phase are created.

(Operation phase)
Fig. 4 shows a functional block diagram of the recipe creation device 130 (operation phase) of this embodiment, Fig. 5 shows the inspection recipe creation flow of this embodiment, and Fig. 6 shows a GUI for creating an inspection recipe according to the flow of Fig. 5. Note that Fig. 5 shows an extracted flow characteristic of this embodiment in the inspection recipe creation flow, and in order to explain the overall picture of the inspection recipe creation of this embodiment, the description will be made with appropriate reference to the inspection recipe creation flow of Fig. 3A.

The GUI shown in Figure 6 is displayed on the display device of the recipe creation device 130. First, the name of the recipe to be created is specified in the recipe name specification section 201 on the recipe creation screen 200a, and when the AI proposal acquisition button 211 is pressed at this time, creation of an inspection recipe according to this embodiment begins. First, imaging position setting S11 is performed in the same manner as in the conventional method (see Figure 3A).

The keyword specification unit 401 then receives a keyword specification S31 from the AI proposal specification unit 212 on the recipe creation screen 200a. Keywords are not limited to specific content, but it is desirable that they be set so that words indicating the direction of the image quality improvement process can be selected. Even for the same circuit pattern image, the image quality that prominently expresses the features of a defect differs depending on whether the inspection position is on an edge portion or a flat portion. To enable the direction of such image improvement process to be identified, the AI proposal specification unit 212 displays a keyword group 213, allowing the operator to select keywords that indicate features that interest them. Keywords may include general words that indicate image features, as well as words that indicate user-specific image features.

The recipe adjustment unit 409 specifies the inspection location in response to the operator's selection from the coordinate list in the inspection location designation unit 204 on the recipe creation screen 200a, and creates an initial recipe in which other imaging parameters and inspection parameters are set to predetermined default conditions. Specifically, the "other imaging parameters and inspection parameters" are parameters set in the defect detection function selection unit 205, SEM imaging parameter setting unit 206, and image quality adjustment parameter adjustment unit 207 on the recipe creation screen 200a. The recipe adjustment unit 409 sends the initial recipe created in this manner to the control unit 120. The control unit 120 acquires a circuit pattern image using the imaging unit 101 in accordance with the initial recipe and sends it to the recipe creation device 130 (S32). Note that the circuit pattern image acquired at this time does not need to be for all inspection locations in the coordinate list, and it is sufficient to acquire one or a small number of circuit pattern images.

Next, the image classification execution unit 403 performs image classification S33 on the circuit pattern image based on the specified keywords and the acquired initial recipe using the image classification model 140. The image classification S33 process will be explained using Figures 7A and 7B.

Figure 7A shows the image classification space formed by the image classification model 140 to classify images, and Figure 7B shows in tabular form the classification results of training image data (circuit pattern images) using the image classification space. The image classification space 300 is an n-dimensional space in which circuit pattern images are positioned, although a three-dimensional space is shown here for simplicity's sake. Circles indicate the coordinates of images positioned in the image classification space 300. The coordinate axes corresponding to each dimension are features extracted from the circuit pattern image or keywords. Features extracted from circuit pattern images include the intensity of high-frequency fluctuations in brightness that reflect the state of the edge, and the intensity of low-frequency fluctuations in brightness that reflect the state of the background. Features extracted from keywords include combination patterns of selected keywords. In the preparation phase described below, the number of dimensions n of image classification space 300 and the n-dimensional coordinate axes are defined, and training is performed using, as learning image data, circuit pattern images captured using an initial recipe created by semiconductor inspection system 801 by applying the above-mentioned default conditions to an existing recipe previously created in image classification space 300, to obtain image classification model 140 that classifies multiple circuit pattern images that are similar to each other into image groups. Here, "similar" means that they are located nearby in image classification space 300. Image groups 301 to 303 shown in Figure 7A correspond to image classification codes 1 to 3 shown in Figure 7B, respectively.

The image classification execution unit 403 inputs the circuit pattern image according to the specified keyword and the acquired initial recipe into the image classification model 140 and infers which image group it will be classified into. The inference process corresponds to the following process, for example: The coordinates of the acquired circuit pattern image in the image classification space 300 are calculated, and it is assumed that these coordinates are coordinates 311 shown in Figure 7A. In this example, the image group closest to coordinates 311 is image group 303 with image classification code "3," so the image classification code "3" is output as the image classification result. Note that while it has been explained here that only the image classification code of the image group closest in the image classification space 300 is output, if there are multiple image groups within a specified distance from coordinates 311, the image classification codes of multiple image groups may be output. In that case, it is advisable to add a priority order or classification certainty according to the distance from coordinates 311 before outputting.

The image classification code ("3" in this example) is transferred from the image classification execution unit 403 to the encryption conversion unit 405, which then performs encryption code calculation S34 based on the image classification code. Here, the encryption conversion unit 405 uses a non-reversible encryption method. As an example, a cryptographic hash function is adopted, and the encryption code is also called a hash value. However, the algorithm used should be one that was published after SHA-224, such as SHA-224, SHA-256, SHA-384, SHA-512, SHA3-224, SHA3-256, SHA3-512, Tiger(2)-192/140128, Whirlpool, MINMAX, RIPEMD-128/256, or RIPEMD-140320. A cryptographic hash function is a hash function with cryptographic mathematical properties suitable for information security applications such as encryption, converting an input of any length into an output of a fixed length. The original image classification code cannot be determined from the hash value. For this reason, even if the user's IP address or a related name is actually used as the image classification code to make it easier for the worker creating the recipe to understand, this information will not be leaked to the outside.

The encryption code calculated by the encryption conversion unit 405 is transferred to the recipe search unit 407, which compares it with the recipe database 150 to extract a recommended recipe (S35). Figure 8 shows the data structure of the recipe database 150. The hash value of the image classification code is registered in the encryption code 151, the recipe name 152 is registered with the recipe name of the inspection recipe created for inspecting the circuit pattern image included in that image classification code, and the pure recipe 153 is registered with the contents of the inspection recipe corresponding to that recipe name. Here, the contents of the inspection recipe registered in the pure recipe 153 are the inspection recipe minus the unique information related to the user IP (hereinafter referred to as unique information), and are referred to as the pure recipe hereinafter. For example, the parameters set in the defect detection function selection unit 205, SEM imaging parameter setting unit 206, and image quality adjustment parameter adjustment unit 207 on the recipe creation screen 200 correspond to the contents registered as a pure recipe.

The recipe adjustment unit 409 creates a trial recipe by replacing the default conditions set as the initial recipe with the optical conditions for imaging and the inspection conditions specified in the pure recipe of the extracted recommended recipe. The recipe adjustment unit 409 sends the trial recipe created in this way to the control unit 120. The control unit 120 performs an inspection trial (S36) according to the trial recipe. The trial results are confirmed (S37), and if the expected results are obtained, the trial recipe is selected as the inspection recipe. On the other hand, if the expected results are not obtained, parameters are adjusted from the optical conditions for imaging (S12) according to the flow of Figure 3A until the expected results are achieved. If the expected results are obtained, the trial recipe with the adjusted parameters is selected as the inspection recipe. The new recipe storage unit 411 stores the selected inspection recipe (S21). Note that if multiple recommended recipes are available, multiple trial recipes can be created based on each recommended recipe, and the operator can select the trial recipe that produces the best trial results and adjust its parameters as necessary.

Note that irreversible encryption methods are not limited to cryptographic hash functions; lossy compression functions can also be used. When a lossy compression function is used, the encryption code is a compressed code, and the hash value is converted to the compressed code.

In the inspection recipe creation method according to this embodiment, not only can the trial recipe be determined as the inspection recipe as is, but even in cases where parameter adjustment is required, adjustments can be started from parameters that already provide relatively good image quality, making it possible to create an inspection recipe more quickly than conventional recipe creation methods that require adjustments from scratch.

(Preparation phase)
FIG. 9 shows a functional block diagram of the recipe creation device 130 (preparation phase) of this embodiment, FIG. 10 shows the flow of creating an image classification database and a recipe database, and FIG. 11 shows a GUI for creating an image classification database in the flow of FIG.

The GUI shown in Figure 11 is displayed on the display device of the recipe creation device 130. First, in the learning target designation section 501 of the learning screen 500, settings are made to acquire circuit pattern images to be registered in the image classification model 140 (see Figure 7B). First, in the recipe designation section 502, an inspection recipe is designated (S41). Below, an example is described in which an existing inspection recipe is designated and the preparation phase is executed, but the preparation phase may also be executed in parallel with the creation of a new inspection recipe as shown in Figure 3A.

The keyword designation unit 401 then receives a keyword designation S42 from the feature of interest designation unit 504 on the learning screen 500. The keyword group 505 selectable in the feature of interest designation unit 504 is the same as the keyword group 213 selectable in the AI proposal designation unit 212 on the recipe creation screen 200a.

Next, a circuit pattern image (learning image data) to be used for training the image classification model 140 is acquired (S43). The circuit pattern image used for training the image classification model 140 is a circuit pattern image acquired according to an initial recipe in which the imaging parameters and inspection parameters of the inspection recipe have been replaced with predetermined default conditions. These default conditions are the same as the default conditions of the initial recipe created in the operation phase. If the operator selects the acquisition method "Execute initial recipe and capture image" in the learning image data acquisition section 503 of the learning screen 500, the recipe adjustment section 409 creates an initial recipe by replacing part of the specified inspection recipe with the default conditions. The recipe adjustment section 409 sends the initial recipe created in this manner to the control unit 120. The control unit 120 acquires a circuit pattern image using the imaging unit 101 in accordance with the initial recipe and sends it to the recipe creation device 130 (S43). Note that if a circuit pattern image has already been captured according to the initial recipe, the operator selects "Existing Image" in the learning image data acquisition section 503 of the learning screen 500 and inputs the data path in which the image is saved. The image classification learning unit 413 can acquire the circuit pattern image by accessing the input data path. If learning image data is to be added (Yes in S44), the process repeats from step S41. If learning image data is not to be added (No in S44), the process proceeds to step S45.

The image classification learning unit 413 trains the image classification model 140 using the training image data (S45). For example, by projecting each circuit pattern image into an n-dimensional image classification space based on features extracted from the acquired circuit pattern image and specified keywords, multiple similar circuit pattern images are classified into image groups. Methods that can be applied to image classification include the K-means algorithm, mixed normal distribution, Ward's method, centroid method, shortest (longest) distance method, group average method, and the Support Vector Machine, a pattern recognition model using supervised learning. Alternatively, methods such as image classification using deep learning or deep learning networks such as ChatGPT, which apply transformer technology, may be used without obtaining features. To perform image classification S45, the worker sets the parameters required for training the image classification model 140 in the learning control parameter setting unit 511. The types of parameters that need to be set vary depending on the image classification model. Additionally, the learning progress display section 521 of the learning screen 500 may display the image classification space 300, and may display the projection status of the circuit pattern images into the image classification space 300 and the classification status of the circuit pattern images. The image classification learning section 413 assigns image classification codes (labels) that uniquely identify the classified image groups. For example, image classification codes 1 to 3 are assigned to the image groups 301 to 303 shown in FIG. 11.

The image classification code ("1" to "3" in this example) set by the image classification learning unit 413 is transferred from the image classification learning unit 413 to the encryption conversion unit 405, which then performs encryption code calculation S46 based on the image classification code.

Meanwhile, the recipe adjustment unit 409 deletes the unique information from the inspection recipe specified in step S41 and creates a pure recipe (S47).

The recipe registration unit 417 creates the recipe database 150 (see Figure 8) using the image classification model 140, the encryption code obtained in step S46, and the pure recipe obtained in step S47. Specifically, it identifies which existing inspection recipe's initial recipe the circuit pattern image included in the image group with image classification code 1 was acquired from, identifies the recipe name 152 corresponding to the encrypted code with the registered hash value of the image classification code, and registers that pure recipe, thereby constructing and registering the recipe database 150 (S48). The original image classification code cannot be determined from the hash value. For this reason, even if a name related to the user IP address is actually used as the image classification code to make it easier for the worker creating the recipe to understand, this information will not be leaked to the outside.

(Modification)
Fig. 12 corresponds to a functional block diagram that integrates the functional block diagram of the recipe creation device 130 (operation phase) shown in Fig. 4 and the functional block diagram of the recipe creation device 130 (preparation phase) shown in Fig. 9. In Fig. 12, the actions between the functional blocks in the operation phase are indicated by solid lines, and the actions between the functional blocks in the preparation phase are indicated by dashed lines.

For example, in this embodiment, assume that an inspection recipe for which parameters have been adjusted is newly saved by the new recipe saving unit 411. In this case, the image classification model 140 and recipe database 150 can be updated by carrying out the above-described preparation phase and performing additional learning on the newly created inspection recipe.

In Example 2, multiple semiconductor inspection systems 801 integrate the image classification models 140 and recipe databases 150 created by each semiconductor inspection system 801 via network 802 to create an integrated image classification model 812 and an integrated recipe database 813. Normally, information related to a user IP is kept under confidential information management by the user and is not permitted to be taken outside the factory, for example. However, in the semiconductor inspection system 801 of Example 1, the output of the image classification model 140 is converted into an encrypted code, and the image classification code is converted into an encrypted code and registered in the recipe database 150, so that user IP information can be taken outside in a form that does not leak to the outside. In Example 2, the high confidentiality of the method disclosed herein is utilized to integrate the image classification models created by multiple semiconductor inspection systems 801 to create an integrated image classification model.

Figure 13A shows three semiconductor inspection systems 801a-c and a recipe creation tool integration device 803 connected via a network 802. Here, semiconductor inspection systems 801a-b are semiconductor inspection systems located on different production lines in the same factory (FabXX), and semiconductor inspection system 801c is a semiconductor inspection system located on a production line in a different factory (FabAA). Figure 13B also shows a functional block diagram of recipe creation tool integration device 803. Recipe creation tool integration device 803 is also realized by information processing device (computer) 10 shown in Figure 1B. Some or all of the functions of recipe creation tool integration device 803 may be realized as a cloud application.

The integration unit 811 creates an integrated image classification model 812 by performing federated learning on the image classification models 140 created by each of the recipe creation devices 130 of the semiconductor inspection systems 801a-c. By performing federated learning, the recipe creation tool integration device 803 can create an integrated image classification model by acquiring the image classification models 140 and performing federated learning without sending highly confidential information, such as learning image data used to train the image classification models 140, to the recipe creation tool integration device 803, which is an external device. The integration unit 811 integrates multiple image classification spaces through federated learning and creates an integrated recipe database 813 by integrating the recipe databases 150 created by each recipe creation device 130. This allows the recipe creation tools built by each of the multiple semiconductor inspection systems to be consolidated into one, integrating the knowledge and experience of engineers.

Figure 14 shows a GUI for causing the integration unit 811 to perform associative learning. The GUI shown in Figure 14 is displayed on the display device of the recipe creation tool integration device 803. First, in the associative learning information specification section 601 of the associative learning screen 600, the semiconductor inspection system to be the target of associative learning is specified. When a data path is specified in the integrated data path specification section 602, a list 604 of semiconductor inspection systems accessible by the recipe creation tool integration device 803 is displayed in the associative learning target selection section 603. The operator specifies the semiconductor inspection system for which associative learning is to be performed and presses the start learning button 605 to start associative learning. At this time, the operator sets the parameters required for associative learning in the associative learning control parameter setting section 611. It is also recommended that the progress of associative learning be displayed in the learning progress display section 621 of the associative learning screen 600.

The integrated image classification model 812 and integrated recipe database 813 created by the recipe creation tool integration device 803 are transmitted via the network 802 to the recipe creation devices 130 of the semiconductor inspection systems 801a-c, where they are used as the image classification model 140 and recipe database 150. Figure 15A shows a GUI used by the recipe creation device 130 of each semiconductor inspection system 801 to update the recipe creation tool (image classification model and recipe database) to the recipe creation tool (integrated image classification model and integrated recipe database) created by the recipe creation tool integration device 803. The GUI shown in Figure 15A is displayed on the display device of the recipe creation device 130. Since the original image classification model and recipe database differ for each integrated image classification model and integrated recipe database, version management is performed by the integration management database 815 of the integration unit 811. For example, in the version management example shown in Figure 15B, version management is performed using the creation date as a key. Note that the key for version management is not limited to creation timing, and for example, search may be possible from multiple perspectives. The worker updates the recipe creation tool using the recipe creation tool update screen 700. When a data path is specified in the AI component path specification section 701, version management information from the integrated management database 815 is read, and a version list 703 is displayed in the federated learning target selection section 702. By specifying one of the versions in the version list 703 and pressing the update start button 704, the update of the recipe creation tool in the recipe creation device 130 begins.

The present disclosure is not limited to the above-described embodiments, but includes various modifications. For example, the above-described embodiments have been described in detail to make the present disclosure easier to understand, and are not necessarily limited to those including all of the configurations described. Furthermore, it is possible to replace part of the configuration of one embodiment or modification with the configuration of another embodiment or modification, and it is also possible to add the configuration of another embodiment or modification to the configuration of one embodiment or modification. Furthermore, it is possible to add, delete, or replace part of the configuration of each embodiment or modification with other configurations.

10: Information processing device (computer), 11: Processor (CPU), 12: Memory, 13: Storage device, 14: Input interface, 15: Output interface, 16: Communication interface, 17: Bus, 101: Imaging unit, 102: Electron source, 103: Acceleration electrode, 105: Primary electron beam, 107: Tilt deflector, 108: BSE detector, 109: Sample, 110: Stage, 111: A/D converter, 112: SE detector, 113: Control signal, 114: Digital signal, 120: Control unit, 121: CPU, 122: Memory, 123: Image processing hardware hardware, 130: recipe creation device, 140: image classification model, 150: recipe database, 151: encryption code, 152: recipe name, 153: pure recipe, 200: recipe creation screen, 201: recipe name specification section, 202: OM alignment adjustment section, 203: SEM alignment adjustment section, 204: inspection location specification section, 205: defect detection function selection section, 206: SEM imaging parameter setting section, 207: image quality adjustment parameter adjustment section, 211: AI proposal acquisition button, 212: AI proposal specification section, 213, 505: keyword group, 300: image classification space, 30 1, 302, 303: Image group, 311: Coordinates, 401: Keyword specification unit, 403: Image classification execution unit, 405: Encryption conversion unit, 407: Recipe search unit, 409: Recipe adjustment unit, 411: New recipe storage unit, 413: Image classification learning unit, 417: Recipe registration unit, 500: Learning screen, 501: Learning target specification unit, 502: Recipe specification unit, 503: Learning image data acquisition unit, 504: Noteworthy feature specification unit, 506: Learning start button, 511: Learning control parameter setting unit, 521: Learning process display unit, 600: Associative learning screen, 601: Associative learning information specification unit, 602: Integration Combined data path designation unit, 603: federated learning target selection unit, 604: list, 605: learning start button, 611: federated learning control parameter setting unit, 621: learning process display unit, 700: recipe creation tool update screen, 701: AI component path designation unit, 702: federated learning target selection unit, 703: version list, 704: update start button, 801: semiconductor inspection system, 802: network, 803: recipe creation tool integration device, 811: integration unit, 812: integrated image classification model, 813: integrated recipe database, 815: integration management database.

Claims (19)

A recipe creation device that acquires image data of a circuit pattern on a wafer using a charged particle beam device, and creates a recipe for a system that inspects or measures the circuit pattern from an image of the image data to automatically perform the inspection or measurement,
an image classification execution unit that uses an image classification model to determine an image classification code of first image data acquired by the charged particle beam device according to a first initial recipe including default conditions;
a cryptographic conversion unit that converts the image classification code of the first image data into a first cryptographic code by a non-reversible cryptographic method;
a recipe search unit that compares the first encryption code with a recipe database to extract recommended recipes,
the image classification model is trained using learning image data, the learning image data being image data acquired by the charged particle beam device in accordance with a second initial recipe, the second initial recipe being a recipe obtained by modifying an existing recipe created for a predetermined inspection or measurement so as to include the default conditions,
The recipe database is characterized in that it registers an encrypted code obtained by converting an image classification code using the encryption method in the encryption conversion unit, and a recommended recipe obtained by deleting unique information from an existing recipe for learning image data classified with the image classification code. In claim 1,
the image classification model determines an image classification code for the first image data based on the first image data and a first keyword assigned to the first image data;
the image classification model is trained using the training image data and a second keyword assigned to the training image data;
A recipe creation device characterized in that the first keyword and the second keyword are selected from a predetermined group of keywords including a keyword indicating a direction of image quality improvement processing to be performed on an image of the image data. In claim 1,
The recipe creation device is characterized in that the recipe includes at least imaging parameters that specify the optical conditions for imaging when acquiring the image data using the charged particle beam device, and inspection or measurement parameters that specify the inspection or measurement conditions when inspecting or measuring the circuit pattern from an image of the image data. In claim 3,
The recipe creating device is characterized in that the default conditions include the imaging optical conditions and the inspection or measurement conditions. In claim 4,
a recipe adjustment unit that creates a trial recipe by replacing the default conditions of the first initial recipe with the recommended recipe extracted by the recipe search unit; In claim 5,
A recipe creating device further comprising a new recipe saving unit for saving a new recipe created by adjusting an imaging parameter or an inspection or measurement parameter included in the trial recipe. A recipe creation device that acquires image data of a circuit pattern on a wafer using a charged particle beam device, and creates a recipe for a system that inspects or measures the circuit pattern from an image of the image data to automatically perform the inspection or measurement,
an image classification learning unit that uses learning image data to train an image classification model that classifies the image data;
a cryptography conversion unit that converts each image classification code classified by the image classification model into a cryptography code using a non-reversible cryptography method;
a recipe registration unit that creates a recipe database in which the encryption code converted from the image classification code and the recommended recipes are registered in association with each other;
the learning image data is image data acquired by the charged particle beam device in accordance with a second initial recipe, the second initial recipe being a recipe obtained by modifying an existing recipe created for a predetermined inspection or measurement so as to include default conditions,
The recipe creation device is characterized in that the recommended recipe is obtained by deleting unique information from the existing recipe of the learning image data classified with the associated image classification code. In claim 7,
The recipe creation device is characterized in that the recipe includes at least imaging parameters that specify the optical conditions for imaging when acquiring the image data using the charged particle beam device, and inspection or measurement parameters that specify the inspection or measurement conditions when inspecting or measuring the circuit pattern from an image of the image data. In claim 8,
The recipe creating device is characterized in that the default conditions include the imaging optical conditions and the inspection or measurement conditions. In claim 9,
the image classification model is trained using the training image data and a second keyword assigned to the training image data;
The recipe creating device is characterized in that the second keyword is selected from a predetermined group of keywords including a keyword indicating a direction of image quality improvement processing to be performed on an image of the image data. In claim 10,
an image classification execution unit that uses the image classification model to determine an image classification code of first image data acquired by the charged particle beam device according to a first initial recipe including the default conditions;
a recipe search unit that compares a first cryptographic code obtained by converting the image classification code of the first image data using the cryptographic method in the cryptographic conversion unit with the recipe database to extract recommended recipes. In claim 11,
the image classification model determines an image classification code for the first image data based on the first image data and a first keyword assigned to the first image data;
The recipe creation device is characterized in that the first keyword is selected from the keyword group. In claim 12,
a recipe adjustment unit that creates a trial recipe by replacing the default conditions of the first initial recipe with the recommended recipe extracted by the recipe search unit; In claim 13,
A recipe creating device further comprising a new recipe saving unit for saving a new recipe created by adjusting an imaging parameter or an inspection or measurement parameter included in the trial recipe. In claim 14,
When the new recipe is created, the recipe creation device performs additional learning to add the new recipe to the recipe database. A recipe creation tool integration device connected to a plurality of systems via a network, the recipe creation tool integration device comprising: a charged particle beam device that acquires image data of a circuit pattern on a wafer and inspects or measures the circuit pattern from the image of the image data; and a recipe creation device that creates a recipe for automatically performing the inspection or measurement,
the recipe creation device provided in each of the plurality of systems,
an image classification learning unit that uses learning image data to train an image classification model that classifies the image data;
a cryptography conversion unit that converts each image classification code classified by the image classification model into a cryptography code using a non-reversible cryptography method;
a recipe registration unit that creates a recipe database in which the encryption code converted from the image classification code and the recommended recipes are registered in association with each other;
The recipe creation tool integration device includes:
an integration unit that acquires the image classification model from each of the recipe creation devices of the plurality of systems, performs federated learning to create an integrated image classification model, and integrates the recipe databases from each of the recipe creation devices of the plurality of systems to create an integrated recipe database;
the learning image data is image data acquired by the charged particle beam device in accordance with a second initial recipe, the second initial recipe being a recipe obtained by modifying an existing recipe created for a predetermined inspection or measurement so as to include default conditions,
The recipe creation tool integration device is characterized in that the recommended recipe is obtained by deleting unique information from the existing recipe of the learning image data classified with the associated image classification code. In claim 16,
The recipe creation tool integration device is characterized in that the integration unit is capable of selecting a system from the plurality of systems to integrate the image classification model and the recipe database. In claim 16,
The recipe creation tool integration device is characterized in that the integration unit performs version management of the integrated image classification model and the integrated recipe database using creation timing as a key. In claim 16,
A recipe creation tool integration device characterized by transmitting the integrated image classification model and the integrated recipe database to the recipe creation device in order to update the image classification model and the recipe database of the recipe creation device provided in each of the multiple systems to the integrated image classification model and the integrated recipe database, respectively.