Device management system and device management method
The equipment management system addresses the challenge of individual differences in scanning electron microscope devices by standardizing index values and generating prediction models, enhancing measurement accuracy and reducing dimensional differences between devices.
Patent Information
- Application Number
- PCT/JP2023/018661
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-18
- Publication Date
- 2024-11-21
AI Technical Summary
Existing methods for managing scanning electron microscope devices fail to accurately account for individual differences between devices, leading to prolonged reduction of measured dimensional differences due to hardware variations and image conditions like blurring and brightness, which affects measurement accuracy.
An equipment management system and method that generates training data sets to predict index values for image data, adjusting devices to standardize index values and generate a prediction model that removes the influence of individual differences, allowing for accurate machine difference factor analysis.
Enables effective management of scanning electron microscope devices by eliminating the impact of individual differences, thereby improving the accuracy of machine difference factor analysis and reducing measurement discrepancies between devices.
Smart Images

Figure JP2023018661_21112024_PF_FP_ABST
Abstract
Description
Device management system and device management method
[0001] The present invention relates to a device management system and a device management method.
[0002] As patterns formed on semiconductor wafers become finer, higher-precision dimensional measurement is required in the semiconductor manufacturing process. The demand for measurement accuracy extends beyond improving the measurement accuracy of individual measurement devices to reducing the difference in measurement dimensions between multiple devices on a production line and reducing fluctuations in measurement dimensions due to changes in the device over time.
[0003] To measure the width of a fine pattern on the order of nanometers, a critical dimension SEM, which is a scanning electron microscope for measuring pattern width and capable of magnifying and capturing an image of the pattern, has been conventionally used. In order to reduce the measurement dimensional discrepancy between multiple scanning electron microscopes installed on a production line, attempts have been made to analyze and adjust the factors behind the machine differences from the data captured during measurement.
[0004] Patent Document 1 discloses a system for managing instrumental differences, which are differences in measurement dimensions between multiple scanning electron microscopes, with the aim of reducing instrumental differences between multiple scanning electron microscopes and instrumental differences due to changes over time. Specifically, Patent Document 1 describes: "A system and method for managing instrumental differences between instruments and instrumental differences due to changes over time in a scanning electron microscope, characterized by comprising: measuring means 10, 301a, 301b; 18, 301a for measuring instrumental differences between instruments and instrumental differences due to changes over time based on secondary electron image data obtained by imaging a standard wafer and substantially simultaneously measuring index values indicating various instrument conditions; instrumental difference factor analysis units 301c, 301e for analyzing the relationship between the instrumental differences measured by the measuring means and the index values indicating the various instrument conditions to estimate the factors causing the instrumental differences; and output means 302 for displaying and outputting the factors causing the instrumental differences estimated by the instrumental difference factor analysis unit."
[0005] JP 2007-122995 A
[0006] In the invention disclosed in Patent Document 1, in a scanning electron microscope apparatus, image feature amounts are calculated directly from acquired data, and the relationship between the machine difference and the feature amounts is analyzed to estimate the cause of the machine difference.
[0007] However, the above technology does not take into account the effects of individual differences that exist between multiple devices. Here, "individual differences" refers to hardware differences between devices, as well as complex differences in the blur, brightness, clarity, and other aspects that appear in images captured by each device. Extracting features from images without considering these individual differences results in extracting features with variable distributions that differ between devices due to individual differences as a cause of machine differences. Adjusting these features does not eliminate machine differences, which may result in a longer time required to reduce measurement dimensional differences.
[0008] The present invention has been made in view of the above circumstances, and aims to manage devices without being affected by individual differences between multiple devices, thereby contributing to improving the accuracy of analysis of the causes of machine differences.
[0009] In order to achieve the above object, one representative equipment management system of the present invention is an equipment management system that manages equipment that images wafers and outputs measurement values of patterns, and includes: a training data generation unit that generates training data including a first learning data set and a second learning data set; and an index value prediction model generation unit that generates an index value prediction model that outputs a predicted index value for a predetermined index that indicates a state of input image data, using the first learning data set and the second learning data set. The training data generation unit generates the first learning data set by associating the same first index value with a plurality of first image data obtained by each of devices in a first equipment group, each of which has an image of a first wafer, where the difference in measurement values for the same wafer is equal to or less than a predetermined reference value; and, after adjusting each device in the first equipment group so that the index value changes by a predetermined value from the first index value, generates the second learning data set by associating a second index value, which is a value that changes by a predetermined value from the first index value, with a plurality of second image data obtained by each of devices after adjustment, Furthermore, one representative equipment management method of the present invention is an equipment management method for managing an equipment that images a wafer and outputs measurement values of a pattern, wherein the equipment management system includes a training data generation step of generating training data including a first learning data set and a second learning data set, and an index value prediction model generation step of generating an index value prediction model that outputs a predicted index value for a predetermined index that indicates a state of input image data, using the first learning data set and the second learning data set, wherein the training data generation step generates the first learning data set by associating the same first index value with a plurality of first image data obtained by each of devices in a first equipment group, each of which has an imaged first wafer, and whose measurement value difference for the same wafer is equal to or less than a predetermined reference value; and after adjusting each device in the first equipment group so that the index value changes from the first index value by a predetermined value, generates the second learning data set by associating a second index value, which is a value changed from the first index value by a predetermined value, with a plurality of second image data obtained by each of devices after adjustment.
[0010] According to the present invention, devices can be managed without being affected by individual differences between multiple devices.
[0011] 1 is a block diagram showing an example of the configuration of a machine-difference factor analysis device and a machine-difference factor analysis system including a device management method according to the first embodiment. FIG. 2 is a flowchart showing an example of a method for generating training data to be stored in a training data storage unit in the machine-difference factor analysis device according to the first embodiment. FIG. 3 is a diagram showing an example of an image obtained by capturing an image of a line pattern. FIG. 4 is a diagram showing an example of a histogram of noise values calculated from images captured by two different devices, including individual differences. FIG. 5 is a diagram showing an example of data in which image data is associated with corrected index values. FIG. 6 is a diagram showing an example of data in which image data is associated with adjusted index values. FIG. 7 is a flowchart showing an example of processing in a machine-difference factor analysis unit in the machine-difference factor analysis device according to the first embodiment. FIG. 8 is a diagram showing an example of a data structure stored in an analysis data storage unit in the machine-difference factor analysis device according to the first embodiment. FIG. 9 is a diagram showing an example of a data structure of predicted index value data stored in an analysis result storage unit by the machine-difference factor analysis unit. FIG. 10 is a flowchart showing an example of processing in a machine-difference determination unit in the machine-difference factor analysis device. FIG. 11 is a diagram showing an example of a data structure stored in an analysis result storage unit by the machine-difference determination unit. FIG. 12 is a diagram showing an example of an output screen output to an input / output device by execution of the machine-difference factor analysis device according to the first embodiment. 1. A flowchart showing an example of processing in an analysis result processing unit in the machine-difference factor analysis device according to the first embodiment. A block diagram showing an example of the configuration of a machine-difference factor analysis device and a machine-difference factor analysis system including an apparatus management method according to the second embodiment. A diagram showing an example of data in which image data, index values, and actual measurement length values are associated with each other. A flowchart showing an example of processing in a machine-difference factor analysis unit in the machine-difference factor analysis device according to the second embodiment. A diagram showing an example of a structure of predicted measurement length value data stored in an analysis result storage unit by the machine-difference factor analysis unit. A diagram showing an example of an output screen output to an input / output device by execution of the machine-difference factor analysis device according to the second embodiment. A block diagram showing an example of the configuration of a machine-difference factor analysis device and a machine-difference factor analysis system including an apparatus management method according to the third embodiment. A flowchart showing an example of processing in a machine-difference factor analysis unit in the machine-difference factor analysis device according to the third embodiment. A flowchart showing an example of processing in a measurement value prediction model generation unit in the machine-difference factor analysis device according to the third embodiment. A diagram showing an example of the structure of data stored in a history information storage unit. A diagram showing an example of the structure of data extracted by the measurement value prediction model generation unit.10 is a flowchart showing an example of processing in a machine-difference factor analysis unit in a machine-difference factor analysis device according to a fifth embodiment. FIG. 11 is a diagram showing an example of the structure of machine-difference factor history data stored in an analysis result storage unit. FIG. 12 is a diagram showing an example of the structure of data stored in the analysis result storage unit by the machine-difference factor analysis unit according to the fifth embodiment.
[0012] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
[0013] 1 shows a machine-difference factor analysis device 101 including a device management method according to an embodiment of the present invention, and a machine-difference factor analysis system 100. The machine-difference factor analysis system 100 of this embodiment includes the machine-difference factor analysis device 101 communicably connected via a network 102, and an input / output device 103 used by a user.
[0014] The network 102 enables the input / output device 103 and the machine-difference factor analysis device 101 to communicate with each other. The input / output device 103 is an information processing device such as a PC (Personal Computer). A user inputs predetermined information into this input / output device, and the input / output device outputs the results calculated by the machine-difference factor analysis device 101. As a specific example, when analyzing the machine-difference factors of two devices in a certain device group, the input information is the device numbers to be compared, and the output information is variables that are candidate machine-difference factors.
[0015] The machine-difference factor analysis device 101 generates an index value prediction model that predicts index values from images based on a data set that associates images captured by a scanning electron microscope with index values from which individual differences have been removed, and analyzes machine-difference factors in a group of devices based on this model.
[0016] The machine-difference factor analysis device 101 includes a storage unit 110 and a calculation unit 120. The storage unit 110 includes a training data storage unit 111 that stores image data and index value information used as training data for constructing an index value prediction model, an analysis data storage unit 112 that stores data used for analysis acquired from the analysis target device group 10, and an analysis result storage unit 113 that stores results calculated by machine-difference factor analysis.
[0017] Furthermore, the training data storage unit 111 can transmit and receive data to and from the analysis target device 10 via a data bus or the Internet 11 .
[0018] The calculation unit 120 includes a machine difference determination unit 121 that calculates the difference in measurement values from data acquired from the group of devices 10 to be analyzed and calculates the magnitude of the machine difference; an index value prediction model generation unit 122 that constructs an index value prediction model based on the training data; a machine difference factor analysis unit 123 that analyzes machine difference factors by applying the index value prediction model to the group of devices to be analyzed; an analysis result processing unit 124 that calculates the results of comparing the analyzed results between devices; and a training data generation unit 125 that generates training data.
[0019] The communication unit 130 communicates with the machine difference factor analysis device 101 and other devices via the network 102 .
[0020] In this embodiment, two devices are assumed as the analysis target device group 10, with device numbers being device A and device B, respectively. In particular, this embodiment assumes a case where the device state of device B has fluctuated, causing a measurement value variation of 0.5 nm. In other words, it is assumed that device difference factor analysis is performed on device A, which is a normal device, and device B, which is a device with a device difference. Here, a device with a device difference is a device with a device difference equal to or greater than a threshold value.
[0021] Here, the number of devices in the analysis target device group 10 may be more than two. In that case, the number of devices to which the index value prediction model described below is applied may be increased to match the number of devices in the analysis target device group.
[0022] In this embodiment, it is assumed that the training data storage unit 111 stores data acquired from the group of devices 10 to be analyzed.
[0023] Here, the training data storage unit 111 may store some of the data of devices not included in the group of devices to be analyzed 10, or only the data of devices not included in the group of devices to be analyzed 10.
[0024] 2 shows a flow for creating data to be stored in the training data storage unit 111. In the following embodiment, an example of a method for creating data to be stored in the training data storage unit 111 will be described in accordance with this flow.
[0025] In this embodiment, the group of devices stored in the training data storage unit 111 is referred to as a first group of devices. After the process starts (step S1), in step S202, the training data generation unit 125 adjusts each scanning electron microscope in the first group of devices so that the difference in the length measurement values obtained falls within a certain standard.
[0026] Here, the standard for a certain measurement value difference is arbitrary. However, in order to create a data set that eliminates the influence of individual differences based on the device state set here, it is desirable to set it within the measurement value difference that is considered to be a machine difference. In this embodiment, the standard for machine difference is set to the magnitude of the measurement value difference from the measurement value of device A in the range of -0.1 nm to +0.1 nm.
[0027] This criterion may be changed as appropriate, such as whether the difference from the maximum measurement value obtained by the first device group or the difference from the average measurement value obtained by the first device group is within a certain range.
[0028] Furthermore, it is desirable that the dimension measurement pattern of the standard wafer to be measured has a very sharp edge portion, and the measurement value changes in response to even a slight difference in the state of the apparatus.
[0029] It is also desirable to use the same standard wafer as the one used for the machine-variation factor analysis as the measurement target. If the design dimensions or pattern types are different, the accuracy of the generated index value prediction model may decrease.
[0030] Furthermore, it is desirable that the standard wafer to be measured has small variations in pattern dimensions on the same wafer. The reason for this is that, in order to prevent the influence of shape dependency of measurement values due to rare changes in the wafer shape, measurement values are generally obtained at multiple locations using the same equipment, and the average value is often used as the representative measurement value for each equipment.
[0031] In this embodiment, the dimension measurement pattern of the target standard wafer is assumed to be a linear pattern with very sharp edges, called a line pattern.
[0032] 3A shows an example of an image acquired by capturing an image of a line pattern. The white line 302 in the captured image 301 is the pattern, and the scanning electron microscope can measure the length of the pattern width 304. The surrounding gray area 303 is the wafer surface.
[0033] As a method for adjusting the device, it is assumed that the device control parameters are adjusted as appropriate, such as the focal position of the electron beam and the retarding voltage that adjusts the acceleration voltage of the electron beam near the wafer. This operation changes the state of the captured image, and therefore the acquired length measurement values change.
[0034] Next, in step S203, the training data generation unit 125 acquires image data of a pattern on a target standard wafer captured by each of the first device group, and then acquires index values that may be machine-variable factors indicating the state of the image data of each device.
[0035] In this embodiment, the index values assumed are a tilt angle index representing the angle of the electron beam, a contrast index representing the brightness contrast of the image, and a noise index quantifying the amount of noise contained in the image. The contrast index and noise index can be calculated from the captured image. The tilt angle index may be calculated from the captured image or may be obtained from the setting value at the time of capturing the image.
[0036] 4 shows a histogram 401 of noise values calculated from images captured by device A and device B. If there are individual differences among devices, even if the difference in measurement values is within the standard, differences in index values may occur, as shown in histogram 411 of device A and histogram 412 of device B. If such index values are used as they are, there is a risk that this variable will be extracted as a factor of machine difference.
[0037] Therefore, in step S204, the training data generation unit 125 first corrects each acquired index value so that the values are the same for the first device group in order to remove the influence of individual differences as shown in Figure 4. In this embodiment, the value of each index value is corrected to 0. This makes it possible to create a data set from which the influence of index value differences due to individual differences that appear between devices has been removed. Furthermore, by constructing an index value prediction model based on this data set and using it in device difference factor analysis, it becomes possible to perform device difference factor analysis from which the influence of individual differences that exist in the device group has been removed.
[0038] Then, in step S204, the image data and the corrected index values are associated with each other and stored in the training data storage unit 111. Here, the image data is not processed to correspond to the correction of the index values.
[0039] 5 shows an example of a data set in which image data and corrected index values are associated. Data set 501 in which image data and corrected index values are associated is composed of an equipment number 511, a path 512 to the image data, and the analysis target index values of tilt angle 513, contrast 514, and noise 515. As described above, a correction value of 0 is stored for tilt angle 513, contrast 514, and noise 515. In FIG. 5, each equipment obtains image data from multiple locations on one wafer.
[0040] Here, different correction index values may be stored for each image. For example, the index values for each device may be standardized so that the average value is 0 and the standard deviation is 1.
[0041] Next, in step S205, the training data generating unit 125 adjusts the equipment in each of the first equipment group so that the image index value changes by a predetermined value, and then images the same wafer to acquire image data.
[0042] The predetermined adjustment value can be determined arbitrarily. In this embodiment, four adjustment values of +100, −100, +50, and −50 are assumed for each index value.
[0043] The more types of adjustment values are added, the more training data is used to generate the index value prediction model described below, and it is therefore expected that a more accurate prediction model will be constructed.
[0044] In this embodiment, it is first assumed that the noise index value for each device is adjusted by +50. The noise index value is adjusted by adjusting the size of the aperture hole, adjusting image processing parameters, or the like.
[0045] Figure 3(b) shows an example of an image acquired by capturing an image of a line pattern after adjusting the device state. The white line 312 in the captured image 311 is the pattern, and the length of the pattern width 314 is measured using a scanning electron microscope. The surrounding gray area 313 is the wafer surface. It is expected that the image will be more distorted by noise than in Figure 3(a), and the width 314, which is the measurement value portion, will be larger, resulting in larger acquired measurement values.
[0046] Next, in step S206, the training data generation unit 125 stores the adjusted index values together with the images in the training data storage unit 111. Fig. 6 shows an example of a data set in which image data and adjusted index values are associated with each other.
[0047] A data set 601 that associates image data with adjusted index values is made up of an apparatus number 611, a path 612 to the image data, and the index values to be analyzed, that is, a tilt angle 613, contrast 614, and noise 615. Here, the noise 615 stores a value obtained by adding the adjustment value corresponding to the adjusted index value in step S204.
[0048] Finally, in step S207, steps S205 and S206 are repeated until data acquisition is completed for all adjustment values determined in step S205.
[0049] The data set generated by the above flow is stored in the training data storage unit 111. In the machine difference factor analysis device of the present invention, a model for predicting index values is generated using this data set, making it possible to perform machine difference factor analysis while removing the influence of individual differences.
[0050] Next, the function of the index value prediction model generation unit 122 will be described. The index value prediction model generation unit 122 receives image data from a dataset stored in the training data storage unit 111 as input and generates a machine learning model that predicts a corresponding index value. The machine learning model outputs output data that is likely to correspond to the input data, in accordance with the trends of the training dataset used for its construction. Therefore, a machine learning model trained with a dataset stored in the training data storage unit 111 from which individual differences have been removed will output an index value from which the influence of individual differences has been removed, even in the predicted value.
[0051] In this embodiment, a convolutional neural network is assumed as the machine learning algorithm.
[0052] As with conventional techniques, it is possible to obtain index values by performing image processing on image data or acquiring electron microscope settings. However, the machine-difference factor analysis device 101 does not obtain index values from image processing on image data or electron microscope settings. Instead, it generates a model that predicts index values by inputting image data through machine learning. The training data for machine learning includes data to which the same index value is assigned for image data whose measurement values fall within a predetermined range, index values that are changed from the data, and image data corresponding to the changed index values. In other words, a model that predicts index values is generated by preparing sets of index values and multiple image data that can correspond to the index values for two or more index values and using these sets as training data. A model that predicts index values may be generated for each type of index value. For example, a tilt angle prediction model, a contrast prediction model, and a noise prediction model may be generated separately. Furthermore, a model that outputs multiple types of index values may be generated.
[0053] 7 shows a processing flow of the machine difference factor analysis unit 123. In the following embodiment, the function of the machine difference factor analysis unit 123 will be described according to this flow.
[0054] In this embodiment, the group of devices that are the subject of the machine-difference factor analysis is called the second group of devices. Here, it is assumed that the data acquired by the second group of devices is stored in the analysis data storage unit 112 in a format to be described later. As described above, it is desirable to use the same standard wafer as the wafer used in the machine-difference factor analysis for the length measurement.
[0055] First, in step S702, the machine difference factor analysis unit 123 reads the data set of the second device group stored in the analysis data storage unit 112.
[0056] FIG. 8 shows an example of a data set acquired from the second device group and stored in the analysis data storage unit 112.
[0057] A data set 801 stored in the analysis data storage unit 112 is made up of an apparatus number 811, a data acquisition date and time 812, a path 813 to the image data, and a measurement value 814 that is the measurement result.
[0058] Next, in step S703, the machine difference factor analysis unit 123 inputs the read image data of each device into the index value prediction model and obtains the output value.
[0059] Next, in step S704, the machine difference factor analysis unit 123 calculates a representative index value for each index of each device from the acquired predicted index values, and stores the calculated representative index value in the analysis result storage unit 113.
[0060] In this embodiment, the average value of each index predicted value for all images of each device is obtained as the representative index value of each device. Here, the method for calculating the representative index value is not limited to the average value, and may be the median value.
[0061] 9 shows an example of a data set stored in the analysis result storage unit 113. A data set 901 stored in the analysis result storage unit 113 is composed of an apparatus number 911 and a predicted index value 912, and the predicted index value 912 is composed of a tilt angle 921, a contrast 922, and a noise 923, which are analysis target index values.
[0062] Here, the tilt angle 921, contrast 922, and noise 923 store representative index values for each index of each device calculated in step S704.
[0063] 10 shows a processing flow of the machine difference determination unit 121. In the following embodiment, the function of the machine difference determination unit 121 will be described according to this flow.
[0064] First, in step S1002 , the machine difference determination unit 121 reads the measurement values of each device from the data set of the second device group stored in the analysis data storage unit 112 .
[0065] Next, in step S1003, the machine difference determination unit 121 calculates a representative measurement value for each device from the measurement values of each image of each device read in step S1002. In this embodiment, the average value is used as the method for calculating the representative measurement value. Here, the method for calculating the representative measurement value is not limited to the average value, and may be the median or the like. Next, in step S1004, the machine difference determination unit 121 compares the representative measurement values of each device based on a predetermined machine difference standard and determines the machine difference device. This result is then stored in the analysis result storage unit 113.
[0066] In this embodiment, the standard is the same as the standard used when acquiring the data to be stored in the training data storage unit 111, and the range of the difference from the measurement value of device A is set to −0.1 nm to +0.1 nm.
[0067] 11 shows an example of a data set stored in the analysis result storage unit 113 by the above process. The data set 1101 is made up of an apparatus number 1111, a length measurement value 1112 of each apparatus, and an apparatus difference determination result 1113.
[0068] In this embodiment, the machine difference judgment result 1113 is set to a machine difference standard within the range of -0.1 nm to +0.1 nm of the measurement value difference from device A, so device A is not a machine difference device, and device B is determined to be a machine difference device because the measurement value difference from device A is 0.5 nm. Therefore, the machine difference judgment result 1113 stores the machine difference judgment of "applicable" only for device B.
[0069] FIG. 12 shows an example of an output screen 1201 output to the input / output device 103, obtained by executing the device difference factor analysis device.
[0070] 1211, which displays the evaluation results based on the data set stored in the analysis result storage unit 113, is composed of 1212, which outputs the devices determined to be due to a machine difference from the data set of FIG. 11, graph 1213, which visualizes the measurement values acquired from each device from the data set of FIG. 11 and the machine difference judgment reference range, and table 1214, which displays the measurement values of each device from the data set of FIG. 11 and the predicted index values from the data set of FIG. 9.
[0071] The display 1221, which is generated by the analysis result processing unit 124 and displays the calculation results of the difference in predicted index values based on the data set stored in the analysis result storage unit 113, includes a table for selecting the device to be compared, a button 1222 for executing the analysis, a graph 1223 of the difference in index values between the compared devices analyzed by the analysis result processing unit 124 based on the data stored in the analysis result storage unit 113 for the compared device selected in 1222, and a table 1224 listing the difference in index values between the compared devices analyzed by the analysis result processing unit 124 based on the data stored in the analysis result storage unit 113.
[0072] 13 shows the processing flow of the analysis result processing unit 124. In the following embodiment, the function of the analysis result processing unit 124 will be described according to this flow.
[0073] First, in step S1302, the analysis result processing unit 124 reads the analysis results of the devices included in the comparison device selected by the input / output device from the analysis result storage unit 113. In this example, device A was selected as comparison device 1 and device B as comparison device 2, so the data sets for these devices are read from the analysis result storage unit. In other words, the data set in FIG. 9 is read.
[0074] Next, in step S1303, the analysis result processing unit 124 calculates and outputs the difference between each prediction index value between comparison device 1 and comparison device 2 based on the loaded data set. In this embodiment, device A was selected as comparison device 1 and device B was selected as comparison device 2, so the difference between them is calculated.
[0075] Here, if multiple devices are selected as comparison devices, appropriate measures may be taken, such as obtaining the average value of each prediction index value within the comparison devices.
[0076] Next, a method will be described in which the present device for analyzing the cause of machine differences performs processing to predict the magnitude of the change in the measurement value contributed by each factor based on the index value predicted from the data of the second device group.
[0077] In the method of Example 1, devices from the second device group are compared, the difference in the predicted values of each index value is calculated, and the cause of the machine difference is determined. However, it is not possible to determine the extent to which the difference in index values affects the measured value. In particular, even if the difference in index values is the same, it is expected that the amount of change in the measured value will change depending on the difference in the magnitude of the index value predicted directly from the image. For example, if the noise predicted values of device A and device B are 0 and 50, respectively, and 150 and 200, respectively, the difference in the predicted values will be the same, but the amount of change in the measured value will be different.
[0078] Therefore, in the machine difference factor analysis device of the second embodiment, the magnitude of the change in the measurement value due to each factor is predicted from the predicted index value of each device in the second device group.
[0079] In this process, the magnitude of the actually occurring difference in the measurement value can be compared with the magnitude of the change in the measurement value caused by each factor, so that the factor of the machine difference can be found with higher accuracy.
[0080] 14 shows a machine-difference factor analysis device 1401 including a device management method according to an embodiment of the present invention, and a machine-difference factor analysis system 1400. The machine-difference factor analysis system 1400 of this embodiment includes the machine-difference factor analysis device 1401 communicably connected via a network 1402, and an input / output device 1403 used by a user.
[0081] The network 1402 enables the input / output device 1403 and the machine difference factor analysis device 1401 to communicate with each other.
[0082] The input / output device 1403 is an information processing device such as a PC (Personal Computer). A user inputs predetermined information into this input / output device, and the input / output device outputs the results calculated by the machine difference factor analysis device. As a specific example, if you want to analyze the machine difference factors of two devices in a certain device group, the input information would be the device numbers you want to compare, and the output information would be variables that are candidate machine difference factors.
[0083] The machine-difference factor analysis device 1401 generates an index value prediction model that predicts index values from images based on a data set that associates images captured by a scanning electron microscope with index values from which individual differences have been removed, and analyzes machine-difference factors in a group of devices based on this model.
[0084] The machine-difference factor analysis device 1401 includes a storage unit 1410 and a calculation unit 1420. The storage unit 1410 includes a training data storage unit 1411 that stores image data and index value information used as training data for constructing an index value prediction model, an analysis data storage unit 1412 that stores data used for analysis acquired from the analysis target device group 20, and an analysis result storage unit 1413 that stores results calculated by machine-difference factor analysis. The training data storage unit 1411 can send and receive data to and from the analysis target device 20 via a data bus or the Internet 21.
[0085] The calculation unit 1420 includes a machine difference determination unit 1421 that calculates the measurement value difference from data acquired from the group of devices 20 to be analyzed and calculates the magnitude of the machine difference, an index value prediction model generation unit 1422 that constructs an index value prediction model based on the training data, a measurement value prediction model generation unit 1423 that constructs a measurement value prediction model based on the training data, a machine difference factor analysis unit 1424 that predicts the measurement value of each factor in each device by inputting predicted index values obtained by applying the index value prediction model to the group of devices to be analyzed into the measurement value prediction model, an analysis result processing unit 1425 that calculates the results of comparing the analyzed results between devices, and a training data generation unit 1426 that generates training data.
[0086] The communication unit 1430 communicates with the machine difference factor analysis device 1401 and other devices via the network 1402 .
[0087] The machine difference factor analysis device 1401 and machine difference factor analysis system 1400 in Fig. 14 have the same configuration as in Fig. 1, except for the addition of a measurement value prediction model generation unit 1423. In this embodiment, various settings such as the first device group and second device group to be analyzed and the machine difference judgment criteria are assumed to be in the same format as in embodiment 1. Any differences will be described accordingly.
[0088] First, a data set is stored in the training data storage unit 1411 by the same process as in the first embodiment.
[0089] In this case, the data sets to be stored are data sets including the length measurement values 1516 and length measurement values 1536 of each image, as shown in FIGS. 15(a) and 15(b).
[0090] These measurement values are those obtained from each image by a scanning electron microscope. Next, an index value prediction model is constructed based on the index value prediction model generating unit 1422 in the same manner as in the first embodiment.
[0091] The measurement value prediction model generation unit 1423 receives index values from the data set stored in the training data storage unit 1411 as input and generates a measurement value prediction model that predicts the measurement value of each factor. In this embodiment, it is assumed that a model that predicts the measurement value from the index value of each factor is generated. Here, there is no problem in generating a single model that receives index values of all factors as input and predicts the measurement value of each factor.
[0092] 16 shows the processing flow of the machine difference factor analysis unit 1424. In the following embodiments, the function of the machine difference factor analysis unit 1424 will be described according to this flow. Here, it is assumed that data acquired by the second device group is stored in the analysis data storage unit 1412 in the same format as in the first embodiment. First, in step S1602, the data set of the second device group stored in the analysis data storage unit 1412 is read in the same manner as in the first embodiment.
[0093] Next, in step S1603, the image data read from each device is input to an index value prediction model and its output value is obtained in the same manner as in embodiment 1. Next, in step S1604, the predicted index value obtained in S1603 is input to a length measurement value prediction model and predicted length measurement values of each index value of each device are obtained.
[0094] Next, in step S1605, the representative index values for each index of each device calculated from the predicted index values obtained in S1603 and the representative measurement values for each index of each device calculated from the predicted measurement values in S1604 are stored in the analysis result storage unit 1413.
[0095] In this embodiment, for both index values and length measurement values, the average value of predicted values acquired for each index in all images of each device is acquired as a representative value. Here, the method for calculating the representative value is not limited to the average value, and may be the median value.
[0096] FIG. 17 shows an example of a data set in which representative values of predicted length measurement values are stored, which is stored in the analysis result storage unit 1413 in step S1605.
[0097] The data set 1701 stored in the analysis result storage unit 1413 is made up of an apparatus number 1711 and a predicted length measurement value 1712, and the predicted length measurement value 1712 is made up of analysis target index values, that is, a tilt angle 1721, a contrast 1722, and a noise 1723. Next, a machine difference determination is performed based on the machine difference determination unit 1421 in the same manner as in the first embodiment.
[0098] FIG. 18 shows an example of an output screen 1801 output to the input / output device 1403 obtained by executing the device difference factor analysis device.
[0099] The format is the same as Figure 12 of Example 1, except for table 1824, which is a component of 1821 that displays the calculation results of the difference between the predicted index value and the predicted measurement value based on the data set stored in the analysis result storage unit 1413 by the analysis result processing unit 1425, and which lists the index value difference and the measurement value difference between the compared devices analyzed by the analysis result processing unit 1425 based on the data stored in the analysis result storage unit 1413.
[0100] The method of calculating the index value difference in the analysis result processing unit 1425 for displaying the table 1824 is performed in the same manner as in Example 1. In addition, the method of calculating the measurement value difference in the analysis result processing unit 1425 can be performed by replacing the calculation method of the index value difference with the measurement value.
[0101] Next, a method for using historical data of the second group of devices acquired in the past to generate a measurement value prediction model in the device difference factor analysis device will be described.
[0102] In the method of Example 2, only the data stored in the training data storage unit 1411 is used to generate the measurement value prediction model, but it is expected that a combination of index values not included in this data will be input during machine difference factor analysis, in which case the measurement value prediction accuracy may decrease.
[0103] The reason for this is that when several factors change in combination, the image state changes in a compound manner, and the magnitude of the measurement value fluctuation due to a certain factor may change from the case of a single factor change. Therefore, when predicting a certain measurement value, if the combination of input index values differs significantly from the data stored in the training data storage unit 1411, the magnitude of the predicted measurement value change will differ from the actual situation, and an analyst using this machine-difference factor analysis device may erroneously determine the machine-difference factor.
[0104] Therefore, in the machine difference factor analysis device of Example 3, by using past actual values in the second device group, it is possible to generate a measurement value prediction model for each factor from data that is not stored in the training data storage unit 1411.
[0105] In this process, the combination of index values analyzed during the machine difference factor analysis is more likely to be similar to the data used as training data when generating the measurement value prediction model, thereby ensuring accuracy.
[0106] FIG. 19 shows an apparatus difference factor analysis device 1901 including an apparatus management method according to an embodiment of the present invention, and an apparatus difference factor analysis system 1900.
[0107] The machine difference factor analysis system 1900 of this embodiment includes a machine difference factor analysis device 1901 communicably connected via a network 1902, and an input / output device 1903 used by a user.
[0108] The network 1902 enables the input / output device 1903 and the machine difference factor analysis device 1901 to communicate with each other.
[0109] The input / output device 1903 is an information processing device such as a PC (Personal Computer). A user inputs predetermined information into this input / output device, and the input / output device outputs the results calculated by the machine difference factor analysis device. As a specific example, if you want to analyze the machine difference factors of two devices in a certain device group, the input information would be the device numbers you want to compare, and the output information would be variables that are candidate machine difference factors.
[0110] The machine-difference factor analysis device 1901 generates an index value prediction model that predicts index values from images based on a data set that associates images captured by a scanning electron microscope with index values from which individual differences have been removed, and analyzes machine-difference factors in a group of devices based on this model.
[0111] The machine-difference factor analysis device 1901 includes a storage unit 1910 and a calculation unit 1920. The storage unit 1910 includes a training data storage unit 1911 that stores image data and index value information used as training data for constructing an index value prediction model, an analysis data storage unit 1912 that stores data used for analysis acquired from the analysis target device group 30, an analysis result storage unit 1913 that stores results calculated by machine-difference factor analysis, and a history information storage unit 1914 that stores data that associates the acquired data, index values predicted by the index value prediction model, and actual length values.
[0112] Furthermore, the training data storage unit 1911 can transmit and receive data to and from the analysis target device 30 via a data bus or the Internet 31 .
[0113] The calculation unit 1920 includes an instrument difference determination unit 1921 that calculates measurement value differences from data acquired from the group of devices 30 to be analyzed and calculates the magnitude of the instrument difference; an index value prediction model generation unit 1922 that constructs an index value prediction model based on the training data; a measurement value prediction model generation unit 1923 that constructs a measurement value prediction model based on the training data; an instrument difference factor analysis unit 1924 that predicts measurement values of each factor in each device by inputting predicted index values obtained by applying the index value prediction model to the group of devices to be analyzed into the measurement value prediction model; an analysis result processing unit 1925 for calculating the results of comparing the analyzed results between devices; and a training data generation unit 1926 that generates training data.
[0114] A communication unit 1930 communicates with the machine difference factor analysis device 1901 and other devices via a network 1902 .
[0115] The machine difference factor analysis device 1901 and machine difference factor analysis system 1900 in FIG. 19 have the same configuration as that in FIG. 14, except for the addition of a history information storage unit 1914 .
[0116] In this embodiment, various settings such as the first device group and second device group to be analyzed and the machine difference judgment criteria are assumed to be in the same format as in embodiment 2. Any differences will be described accordingly. First, a data set is stored in the training data storage unit 1911 using the same processing as in embodiment 2, and an index value prediction model is constructed.
[0117] 20 shows the processing flow of the machine difference factor analysis unit 1924. In the following embodiment, the function of the machine difference factor analysis unit 1924 will be described according to this flow.
[0118] Here, it is assumed that the analysis data storage unit 1912 stores data acquired by the second device group in the same format as in the second embodiment.
[0119] First, in steps S2002 and S2003, in the same manner as in Example 2, the data set of the second device group stored in the analysis data storage unit 1912 is read, input into the index value prediction model, and the output value is obtained.
[0120] Next, in step S2004, a measurement value prediction model is generated by the measurement value prediction model generation unit 1923 based on the predicted index values acquired in step S2003. Here, in this embodiment, it is assumed that the index values of the factors (tilt angle, contrast, noise) predicted by the index value prediction model for device A and device B in the second device group are (100, 100, 30) and (110, 100, 100), respectively.
[0121] 21 shows the processing flow of the measurement value prediction model generating unit 1923. In the following embodiment, the function of the measurement value prediction model generating unit 1923 will be described according to this flow.
[0122] First, in step S2102, the index value prediction result calculated by the machine difference factor analysis unit 1924 is read.
[0123] Next, in step S2103, one factor is selected from the factors to be analyzed, and data having a similar combination of index values other than the selected factor is extracted from the history information storage unit 1914. In this embodiment, it is assumed that a noise index is first selected.
[0124] 22 shows an example of data stored in the history information storage unit 1914. A data set 2201 stored in the history information storage unit 1914 is composed of an apparatus number 2211, an acquisition date and time 2212, a path to an image 2213, an actual measurement value 2214 of a length measurement value, and a predicted index value 2215. The predicted index value 2215 is composed of a tilt angle 2221, a contrast 2222, and a noise 2223, which are index values to be analyzed.
[0125] This data set is a data set that includes image data of the second device group acquired in past operations, measurement values, and index value prediction results acquired by analysis using an index value prediction model.
[0126] From this data set, only data in which the combination of index values other than the selected factors is similar to that of the analysis target is extracted and used to generate a measurement value prediction model.
[0127] Regarding the definition of similar data sets, in this embodiment, the average value is calculated for each device for index values other than the factors selected in step S2103, and the data is defined as data in the range of -10 to +10 from that average value.
[0128] That is, in this embodiment, when generating a length measurement value prediction model targeting noise, the prediction index values for factors other than the selection factors (tilt angle, contrast) for device A and device B are (100, 100) and (110, 100), respectively, so the average of these is calculated as (105, 100). Therefore, data in which the tilt angle is in the range of -95 to +115 and the contrast is in the range of -90 to +110 is considered to be similar data.
[0129] 23 shows an example of the data set extracted in step S2103 in this embodiment. The data set 2301 extracted in step S2103 is composed of an acquisition date and time 2311, a path to the image 2312, an actual measurement value 2313 of the length measurement value, and a predicted index value 2314. The predicted index value 2314 is composed of a tilt angle 2321, a contrast 2322, and a noise 2323, which are the index values to be analyzed.
[0130] Looking at the tilt angle 2321 and contrast 2322, it can be seen that the index values are included in the range set for the similarity criterion described above.
[0131] Next, in step S2104, it is determined whether the number of data extracted in step S2103 is equal to or greater than a certain value. If the number of data is small, it may be impossible to generate an appropriate measurement value prediction model, so this process is performed to check. In this embodiment, the standard number of data is set to 10 or more. This standard may be changed as appropriate, for example, by confirming accuracy through experiments or the like and finding an appropriate value.
[0132] Next, in step S2105, a measurement value prediction model is constructed based on the extracted data. By generating a measurement value prediction model through this process, the combination of index values input during machine difference factor analysis is similar to the data used as training data, and the accuracy of the measurement value prediction model can be ensured.
[0133] In S2104, if the number of extracted data is small, as described in step S2106, a measurement value prediction model is constructed based on the data set stored in the training data storage unit 1911 using a method similar to that of Example 2.
[0134] Next, in step S2107, it is determined whether measurement value prediction models have been generated for all factors. If measurement value prediction models have not been generated for all factors, in step S2103 a factor for which no measurement value prediction models have been generated is selected, and the process is repeated.
[0135] The processing flow up to this point is the function of the measurement value prediction model generating unit 1923. Next, in step S2005, the acquired predicted index values are input to the measurement value prediction model, and predicted measurement values for each index value of each device are acquired.
[0136] Next, in step S2006, as in Example 2, the representative index values for each index of each device calculated from the predicted index values and the representative measurement values for each index of each device calculated from the predicted measurement values are stored in the analysis result storage unit 1913.
[0137] Thereafter, by performing the same processing as in the second embodiment, this measurement value prediction model can be used for analyzing the causes of machine differences.
[0138] In Examples 1, 2, and 3, the present invention is used to estimate the cause of differences in measurement values between different devices, but it may also be used to estimate the cause of differences in measurement values that occur in the same device over different time series, instead of between different devices.
[0139] Therefore, in the fourth embodiment, the devices compared in FIGS. 12 and 18 are the same device in time series.
[0140] Therefore, the group of devices to be analyzed 10 in the system configuration is configured with at least one scanning electron microscope device.
[0141] Next, a method will be described in which the present device for analyzing machine difference factors estimates and outputs machine difference factor candidates based on machine difference factor history data of the second device group that has been analyzed in the past.
[0142] In the method of the first embodiment, when analyzing the cause of the machine difference, if an image similar to an image that previously caused a misjudgment of the cause is acquired, there is a risk of causing a similar misjudgment.
[0143] Therefore, the machine-difference factor analysis device of Example 5 can refer to the machine-difference factors of images that were previously analyzed by the second device group and for which the factors were incorrectly determined, determine whether the image data is similar to the image data used for machine-difference factor analysis, and infer machine-difference factor candidates. This process can prevent misjudgments similar to those that occurred in the past due to machine-difference factors.
[0144] The configuration of the machine-difference factor analysis device including the device management method in this embodiment is the same as that in Fig. 1. However, unlike in the first embodiment, the analysis result storage unit 113 stores machine-difference factor history data that was created when the second device group was analyzed in the past and the analyst confirmed the factors, and the machine-difference factor analysis unit has a function of analyzing the relationship between the images acquired using the machine-difference factor history data and the images obtained from the second device group.
[0145] In this embodiment, various settings such as the first device group and second device group to be analyzed and the machine difference judgment criteria are assumed to be in the same format as in embodiment 1. If there are any differences, they will be described accordingly.
[0146] In this embodiment, the machine learning model used as the index value prediction model is assumed to be a machine learning model having a neural network inside.
[0147] A neural network, also known as a multi-perceptron, is a technology that uses multiple perceptrons stacked in layers to provide the ability to solve linearly inseparable problems. Here, the intermediate layer shown in the examples may be any of the multiple perceptron layers used to calculate the prediction results.
[0148] The convolutional neural network, which is the machine learning algorithm assumed in Example 1, has a neural network inside, and therefore this algorithm is also assumed in this Example.
[0149] 24 shows the processing flow of the machine difference factor analysis unit 123 in this embodiment. In the following embodiment, the function of the machine difference factor analysis unit 123 will be described according to this flow.
[0150] In steps S2402 and S2403, as in Example 1, the data set of the second device group stored in the analysis data storage unit 112 is read, the image data of each device is input into the index value prediction model, and the output value is obtained.
[0151] Next, in step S2404, the read image data is input into an index value analysis model, and intermediate layer values are obtained as a first feature vector. Here, the feature vector is assumed to be a vector obtained by inputting each image into an index value prediction model and averaging the intermediate layer values of each obtained image across all images. In other words, in this step, feature vectors are calculated for the number of second device groups. Here, the method for calculating the feature vector is not limited to averaging, and other statistics such as the median may also be obtained.
[0152] Next, in step S2405, the data stored in the analysis result storage unit 113 is read and input to the index value analysis model, and the value of the intermediate layer is obtained as a second feature vector.
[0153] 25 shows an example of a machine-difference factor history data set stored in the analysis result storage unit 113. A data set 2501 stored in the analysis result storage unit 113 is composed of a device number 2511, an acquisition date and time 2512, a path 2513 to the image data, and an actual machine-difference factor history 2514 determined by the analyst.
[0154] The machine difference factors are assumed to be input by an analyst using the machine difference factor analysis device. The number of data items read from the analysis result storage unit must be one or more. Therefore, the feature vectors calculated in this step are the same as the number of data items read from the analysis result storage unit.
[0155] Next, in step S2406, the similarity between the first feature amount vector and the second feature amount vector is calculated.
[0156] In this embodiment, the reciprocal of the mean square error is used as the similarity between vectors. By taking the reciprocal of the error, the larger the value, the more similar the vectors are.
[0157] Here, the method for calculating the similarity may be appropriately selected, such as cosine similarity calculated from the angle formed by vectors.
[0158] Next, in step S2407, the data with the highest similarity is selected from the data stored in the analysis result storage unit 113, and it is determined whether the similarity exceeds a set threshold value. If it does, the machine difference factor of the target data is acquired.
[0159] In this embodiment, the threshold value of the similarity is set to 0.5, but this threshold value may be changed as appropriate.
[0160] In this embodiment, it is assumed that the image of device B has the highest similarity with data in which the machine-difference factor is contrast among the machine-difference factor history data sets stored in the analysis result storage unit 113, and the similarity of the feature vectors exceeds a threshold value. It is also assumed that the image of device A has no similar image among the machine-difference factor history data sets stored in the analysis result storage unit 113.
[0161] Finally, in step S2408, a representative index value for each index of each device is calculated from the predicted index value in the same manner as in the first embodiment, and stored in the analysis result storage unit 113 together with the machine difference factors acquired in step S2407.
[0162] 26 shows an example of a data set stored in the analysis result storage unit 113 by the above processing. A data set 2601 stored in the analysis result storage unit 113 is composed of an apparatus number 2611, an apparatus difference factor 2612, and a predicted index value 2613, and the predicted index value 2613 is composed of a tilt angle 2621, a contrast 2622, and a noise 2623, which are the index values to be analyzed.
[0163] By outputting the stored results to an input / output device, when analyzing the causes of machine differences, it is possible to infer possible causes of machine differences based on the machine difference cause history of images previously analyzed by the second device group.
[0164] As described above, the disclosed system is an equipment management system for managing equipment that images wafers and outputs measurement values of patterns, and includes a training data generation unit (125, 1426, 1926) that generates training data including a first learning data set and a second learning data set, and an index value prediction model generation unit (122, 1422, 1922) that generates an index value prediction model that outputs a predicted index value for a predetermined index that indicates a state of input image data, using the first learning data set and the second learning data set. The training data generation unit generates the first learning data set by associating the same first index value with a plurality of first image data obtained by each equipment of a first equipment group, each of which has an image of a first wafer, where the difference in measurement values for the same wafer is equal to or less than a predetermined reference value, and after adjusting each equipment of the first equipment group so that the index value changes from the first index value by a predetermined value, generates the second learning data set by associating a second index value, which is a value changed from the first index value by a predetermined value, with a plurality of second image data obtained by each equipment after adjustment, In this way, by generating an index value prediction model by assigning the same index value to image data whose measurement values fall within a predetermined range, it is possible to manage devices using index values that are not affected by individual differences between multiple devices.
[0165] The index is an index that can be a factor in instrumental differences, and includes any one of a tilt angle index that indicates the angle of the electron beam of the device, which is a scanning electron microscope, a contrast index that indicates the brightness contrast of the image data, and a noise index that quantifies the amount of noise contained in the image data. In this way, the disclosed system can eliminate the influence of individual differences between scanning electron microscopes and determine the index that can be a factor in instrumental differences.
[0166] The training data generation unit calculates index values from the first image data and corrects the calculated index values for each type of index so that the values are the same for the first device group, thereby obtaining the first index values. Therefore, index values for training data can be obtained by using the index values calculated from the first image data.
[0167] The disclosed system further includes an instrument-difference factor analysis unit (123, 1424, 1924) that acquires second image data obtained by imaging a second wafer from each of the second tool groups, and analyzes instrument-difference factors based on predicted index values output by inputting the second image data into the index value prediction model, and an analysis result processing unit (124, 1425, 1925) that outputs the analysis results of the instrument-difference factors. With this configuration, the disclosed system can analyze instrument-difference factors and output the analysis results without being affected by individual differences.
[0168] The disclosed system further includes a measurement value prediction model generation unit (1423, 1923) that uses a third learning data set in which measurement values of the first image data correspond to the first index values to generate a measurement value prediction model that receives the predicted index values as input and outputs predicted measurement values, and the machine-difference factor analysis unit estimates an index that is a factor of the machine difference based on the predicted measurement values output by inputting the predicted index values based on the second image data to the measurement value prediction model. With this configuration, the disclosed system can predict measurement values and analyze machine-difference factors without being affected by individual differences.
[0169] Furthermore, the measurement value prediction model generation unit references historical data that associates predicted index values based on the second image data with measurement values, and uses data with similar conditions other than the index to be predicted as the third learning data set. This allows for highly accurate analysis of machine-variation factors even when multiple factors are involved.
[0170] Furthermore, the machine-difference factor analysis unit estimates an index that is a candidate factor for the difference in measurement values over time, based on image data of the second wafer captured by one apparatus at a different time point. Therefore, it is possible to estimate the factor for the difference in measurement values due to changes over time.
[0171] The machine-difference factor analysis unit also acquires image data when a machine-difference factor is misjudged and information on the registered machine-difference factor, and estimates an index that is a candidate for the machine-difference factor from the relationship between the image data and the second image data. This allows for past judgment errors to be reflected and improves the accuracy of the analysis of machine-difference factors.
[0172] The present invention is not limited to the above-described embodiments, but includes various modifications. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to those including all of the described configurations. Furthermore, not only can the configurations be deleted, but also replacements and additions of configurations are possible.
[0173] 10, 20, 30: Group of devices to be analyzed 11, 21, 31: Internet 100, 1400, 1900: Machine difference factor analysis system 101, 1401, 1901: Machine difference factor analysis device 102, 1402, 1902: Network 103, 1403, 1903: Input / output device 110, 1410, 1910: Memory unit 111, 1411, 1911: Training data storage unit 112, 1412, 1912: Analysis data storage unit 113, 1413, 1913: Analysis result storage unit 120, 1420, 1920: Calculation unit 121, 1421, 1921: Machine difference determination unit 122, 1422, 1922: Index value prediction model generation unit 123, 1424, 1924: Machine difference factor analysis unit 124, 1425, 1925: Analysis result processing unit 125, 1426, 1926: Training data generation unit 1423, 1923: Measurement value prediction model generation unit
Claims
1. An equipment management system for managing equipment that images a wafer and outputs measurement values of a pattern, comprising: a training data generation unit that generates training data including a first learning data set and a second learning data set; and an index value prediction model generation unit that generates an index value prediction model that outputs a predicted index value for a predetermined index that indicates a state of input image data, using the first learning data set and the second learning data set, wherein the training data generation unit generates the first learning data set by associating the same first index value with a plurality of first image data obtained by each equipment of a first equipment group, the first image data being captured by each equipment of a first equipment group having a measurement value difference for the same wafer that is equal to or less than a predetermined reference value; and, after adjusting each equipment of the first equipment group so that the index value changes by a predetermined value from the first index value, generates the second learning data set by associating a second index value that is changed by a predetermined value from the first index value with a plurality of second image data obtained by each equipment after adjustment of the first wafer.
2. An equipment management system as described in claim 1, wherein the index is an index that can be a factor of machine difference and includes any one of a tilt angle index that represents the angle of the electron beam of the equipment, which is a scanning electron microscope, a contrast index that represents the light-dark contrast of image data, and a noise index that quantifies the amount of noise contained in image data.
3. An equipment management system as described in claim 1, wherein the training data generation unit calculates an index value from the first image data, corrects the calculated index value for each type of index so that the values are the same for the first equipment group, and sets the calculated index value as the first index value.
4. An equipment management system as described in claim 1, further comprising: an equipment difference factor analysis unit that acquires second image data obtained by imaging a second wafer from each of a second equipment group, and analyzes equipment difference factors based on a predicted index value output by inputting the second image data into the index value prediction model; and an analysis result processing unit that outputs the analysis result of the equipment difference factors.
5. An equipment management system as described in claim 4, further comprising a measurement value prediction model generation unit that uses a third learning data set in which the measurement values of the first image data correspond to the first index values to generate a measurement value prediction model that inputs the predicted index value and outputs a predicted measurement value, and the machine difference factor analysis unit estimates an index that is a factor of the machine difference based on the predicted measurement value output by inputting the predicted index value based on the second image data into the measurement value prediction model.
6. An equipment management method as described in claim 5, characterized in that the measurement value prediction model generation unit refers to historical data that associates predicted index values based on the second image data with measurement values, and uses data with similar conditions other than the index to be predicted as the third learning data set.
7. An equipment management system according to claim 4, wherein the machine difference factor analysis unit estimates an index which is a candidate factor for the difference in measurement values over time based on image data of the second wafer captured by one equipment at a different time point.
8. An equipment management system as described in claim 4, wherein the machine-difference factor analysis unit acquires image data when a machine-difference factor is misjudged and information on registered machine-difference factors, and infers an index that is a candidate for the machine-difference factor from the relationship between the image data and the second image data.
9. An equipment management method for managing an equipment that images a wafer and outputs measurement values of a pattern, the equipment management system comprising: a training data generation step for generating training data including a first learning data set and a second learning data set; and an index value prediction model generation step for generating an index value prediction model that outputs a predicted index value for a predetermined index indicating a state of input image data, using the first learning data set and the second learning data set, wherein the training data generation step generates the first learning data set by associating the same first index value with a plurality of first image data obtained by each equipment of a first equipment group, the first image data being captured by each equipment having a measurement value difference for the same wafer that is equal to or less than a predetermined reference value; and, after adjusting each equipment of the first equipment group so that the index value changes by a predetermined value from the first index value, generates the second learning data set by associating a second index value that is changed by a predetermined value from the first index value with a plurality of second image data obtained by each equipment after adjustment of the first wafer.
Citation Information
Patent Citations
Apparatus difference control system in scanning electron microscope apparatus and its method
JP2007122995A
Pattern dimension measurement method using electron microscope, pattern dimension measurement system and monitoring method of secular change of electron microscope device
JP2012026989A
