Surface analysis device

Abstract

The present invention provides a surface analysis device that can avoid detecting pseudo clusters when automatically clustering points on a scatter plot, thereby improving the accuracy of clustering. One mode of the surface analysis device according to the present invention includes: a measurement unit (1-2, 4-8) that respectively acquires signals reflecting the amounts of a plurality of components or elements to be analyzed at a plurality of positions on a specimen (3); a scatter plot creation unit (92) that creates a two-dimensional scatter plot based on the measurement results obtained by the measurement unit; a clustering analysis unit (94) that clusters points in the two-dimensional scatter plot using a density-based clustering analysis method; and a parameter adjustment unit (93) that adjusts a distance threshold, which is one of the parameters to be set in the density-based clustering analysis, by using value distribution information of the signal values of one of the components or elements in the two-dimensional scatter plot.

Description

Technical Field

[0001] The present invention relates to a surface analysis apparatus for investigating the distribution of components or elements present in a one-dimensional or two-dimensional measurement region on a specimen. The surface analysis apparatus includes an Electron Probe MicroAnalyzer (EPMA), a Scanning Electron Microscope (SEM), an X-ray fluorescence analysis apparatus, and the like. Background Art

[0002] In elemental mapping analysis using EPMA, it is possible to investigate the types and amounts of elements contained in a large number of minute regions within a two-dimensional region on a specimen. The following method, i.e., phase analysis (or phase analysis), is often used: When analyzing the results obtained by such elemental mapping analysis, a scatter diagram of characteristic X-ray intensities for two or three elements, or a scatter diagram of elemental concentrations calculated from the intensities (a diagram obtained by taking the relative intensities of the elements for each of the two or three axes), is created, and based on the distribution of the marked points on the diagram, the types and content ratios of the compounds contained in the specimen are confirmed (see Patent Documents 1 and 2). For example, Patent Document 2 Figure 10 shows an example of a binary scatter diagram, and this document Figure 11 shows an example of a ternary scatter diagram.

[0003] One point on the scatter diagram (hereinafter, the points marked on the scatter diagram are referred to as "data points") corresponds to one point (minute region) on the specimen. Therefore, the region where the data points on the scatter diagram are dense is estimated to correspond to the portion on the specimen that contains the elements in the same ratio. Thus, in phase analysis, generally, the analyst identifies the region where the data points on the scatter diagram are dense as one cluster, that is, a set of associated data points, and uses an indicating device such as a mouse to perform an operation of surrounding the region with an appropriate figure such as a polygon, and also performs an operation of assigning a different display color to each such region. When such an operation is completed, a phase map is displayed on the screen of the display device of the EPMA, and in this phase map, the positions on the specimen corresponding to each of the data points included in one or more cluster regions are colored with the specified color.

[0004] In recent years, with the rapid development of AI (Artificial Intelligence) technology, attempts have been made to use such technology to automatically perform the process of allocating a large number of data points on the scatter plot as described above to multiple sets. In such a process, clustering analysis, which is a representative method of unsupervised machine learning, is preferably used. For clustering analysis, various algorithms of clustering analysis are known, but as a method for dividing data points on a scatter plot into multiple clusters according to their density, for example, density-based clustering analysis (Density-Based Clustering) disclosed in Non-Patent Documents 1, 2, etc. is effective. In Figure 12 an example is shown in which density-based clustering analysis is used to automatically extract clusters for a two-dimensional scatter plot obtained through actual measurement. In this example, it can be seen that 6 clusters are extracted.

[0005] Prior Art Documents

[0006] Patent Documents

[0007] Patent Document 1: Japanese Unexamined Patent Application Publication No. 2006-125952

[0008] Patent Document 2: Japanese Unexamined Patent Application Publication No. 2011-153858

[0009] Non-Patent Documents

[0010] Non-Patent Document 1: Ester M. et al. (3 others), "A Density-Based Algorithm for Discovering Clusters in Large Spatial Databases with Noise", Proceedings of 2nd International Conference on Knowledge Discovery and Data Mining (KDD-96), pp. 226-231, 1996

[0011] Non-Patent Document 2: Ricardo J.G.B.Campello et al. (2 others), "Density-Based Clustering Based on Hierarchical Density Estimates", Springer, pp. 160-172, 2013 Summary of the Invention

[0012] Problems to be Solved by the Invention

[0013] However, in the scatter diagram produced based on the data obtained by using EPMA, sometimes due to various factors, there is an uneven and specific distribution of data points. Therefore, in the case of applying the existing clustering analysis method as described above, sometimes pseudo clusters are detected.

[0014] For example, Figure 4 is a diagram showing the result obtained by automatically clustering the data points on the binary scatter diagram produced based on the measurement results. In the diagram, the range surrounded by the polygonal line is one cluster region automatically detected. At the two positions indicated by the arrows in the diagram, a plurality of small clusters extending linearly in the longitudinal direction are detected, but it is appropriate that they are originally detected as one large cluster at each position. That is, the small clusters extending linearly in the longitudinal direction are pseudo clusters.

[0015] In addition, Figure 8 is a diagram showing the result obtained by automatically clustering the data points on the ternary scatter diagram produced based on the measurement results. In the diagram, the range surrounded by the polygonal line is one cluster region automatically detected. It can be seen from Figure 8 that in the ternary scatter diagram, sometimes a set of data points connected in a line appears radially, but a plurality of small-range clusters containing a part of the set of data points are detected. Among them, there are clusters that are not necessarily suitable as clusters, and most of them are pseudo clusters.

[0016] The present invention is completed to solve the above problems, and its main object is to provide a surface analysis device that can suppress the detection of pseudo clusters when automatically clustering data points on a scatter diagram, thereby improving the accuracy of the clustering.

[0017] Solution for Solving the Problem

[0018] The first aspect of the surface analysis device according to the present invention completed to solve the above problems includes:

[0019] A measurement unit that respectively obtains signals reflecting the amounts of a plurality of components or elements to be analyzed at a plurality of positions on a sample;

[0020] A scatter diagram creation unit that creates a binary scatter diagram based on the measurement results obtained by the measurement unit;

[0021] A clustering analysis unit that clusters the data points in the binary scatter diagram by using a density-based clustering analysis method; and

[0022] A parameter adjustment unit that adjusts a distance threshold, which is one of the parameters to be set in the density-based clustering analysis, by using the distribution information of the signal values of one of the components or elements in the binary scatter diagram.

[0023] In addition, a second aspect of the surface analysis apparatus according to the present invention, which is completed to solve the above problems, includes:

[0024] A measurement unit that obtains signals reflecting the amounts of a plurality of components or elements to be analyzed at a plurality of positions on a sample;

[0025] A scatter plot creation unit that creates a ternary scatter plot based on the measurement results obtained by the measurement unit;

[0026] A data point selection unit that uses the distribution information of the sum signal values obtained by adding the signal values of three components or elements corresponding to the data points in the ternary scatter plot to exclude data points in a specified signal value range with relatively small sum signal values from all the data points existing on the ternary scatter plot; and

[0027] A clustering analysis unit that performs clustering on the data points in the ternary scatter plot that have not been excluded by the data point selection unit using a density-based clustering analysis method.

[0028] The surface analysis apparatuses according to the first and second aspects of the present invention are, for example, analysis apparatuses such as EPMA, SEM, and X-ray fluorescence analysis apparatuses. In such an analysis apparatus, it is possible to obtain signals reflecting the abundances of a plurality of elements at each of a large number of positions in a two-dimensional region or a one-dimensional region on a sample by repeating measurements while changing the position of the irradiation of excitation rays (electron rays, X-rays, etc.) on the sample.

[0029] In addition, in the surface analysis apparatuses according to the first and second aspects of the present invention, as the clustering analysis method, it is possible to use a hierarchical density-based spatial clustering analysis (Hierarchical Density-Based Spatial Clustering of Applications with Noise) obtained by improving the general density-based spatial clustering of applications with noise (DBSCAN).

[0030] Effects of the Invention

[0031] In density-based clustering analysis, the distance threshold ε is an important parameter for clustering. In particular, in hierarchical density-based spatial clustering analysis, the threshold ε is automatically adjusted according to the density of the data points on the scatter plot. As described above, in Figure 4 the binary scatter plot shown, a plurality of pseudo clusters extending linearly in the longitudinal direction are detected. On the other hand, the set of data points in the lower region (the region with high Fe intensity) in this binary scatter plot is appropriately identified as one cluster.

[0032] When making a speculation based on the density of data points on the scatter diagram, in the lower region of the binary scatter diagram, the frequency of data points is not originally very high. Therefore, the threshold value ε is appropriately set, and thereby, the distance between data points classified into one cluster is considered to become larger. On the other hand, in the upper region of the binary scatter diagram (the region where the intensity of Fe is low), the frequency of data points is very high compared to the lower region. As in this example, when the intensity of one of the two elements (Mn in this example) is extremely low and the measurement range of the data of this element (the range of X-ray intensity as the measurement result) is narrow, data points are densely aggregated in one direction (the vertical direction in this example), and data points are discretely present in the other direction (the horizontal direction in this example). Therefore, it is speculated that: in the process of normal automatic parameter adjustment, the threshold value ε is determined reflecting the state where data points in the vertical direction are extremely dense, and sets of data points discretely appearing in the horizontal direction are respectively misrecognized as independent clusters.

[0033] In contrast, in the first mode of the surface analysis apparatus according to the present invention, the parameter adjustment unit adjusts the value of the distance threshold ε by using the distribution information of the signal value of one of the elements at the data points in the binary scatter diagram, that is, the information indicating the degree of concentration or dispersion of the signal value. That is, since the measurement range is narrow, the parameter adjustment unit adjusts the distance threshold ε according to the degree of distribution of the signal values of the data points in the axial direction where data point sets that are likely to form discrete linear shapes are formed when the density of data points on the scatter diagram is high. Thereby, it is possible to adjust the distance threshold ε such that a plurality of discrete linear data point sets close to each other on the scatter diagram are included in one cluster as a whole, and thus appropriate clustering is performed.

[0034] In addition, when creating a ternary scatter diagram as Figure 8 shown, as also described in Patent Document 1, it is necessary to standardize each data point of the ternary scatter diagram with the total value of the intensities of the three elements. Therefore, when there are a large number of data points with the intensity ratios of the three elements being of the same degree, a distribution of linear data points extending radially appears on the ternary scatter diagram. The present inventors have found that: when detecting a pseudo-cluster including a linear point distribution as Figure 8 shown, there are a large number of data points with small intensities of all three elements, and these large numbers of data points contribute significantly to the generation of the linear data point distribution extending radially.

[0035] In contrast, in the second mode of the surface analysis apparatus according to the present invention, for example, as distribution information of the added signal values obtained by adding the signal values of three components or elements corresponding to the data points in the ternary scatter diagram, the data point selection unit creates a histogram showing the relationship between the added signal value steps and the frequencies. When there are a large number of data points where the intensities of the three elements are all small as described above, a large peak appears at a position where the added signal value in the histogram is small. Therefore, the data point selection unit excludes the data points forming the peak from all the data points. The clustering analysis unit performs clustering only on the remaining data points. As a result, the distribution of the linear data points radially extending in the ternary scatter diagram disappears, and the detection of the accompanying pseudo clusters is also avoided.

[0036] As described above, according to the surface analysis apparatus of the first mode and the second mode of the present invention, it is possible to suppress the detection of pseudo clusters when automatically clustering the data points on the scatter diagram, thereby improving the accuracy of clustering of the data points. That is, the accuracy of clustering of the minute regions on the specimen based on the concentrations of multiple components or elements is improved. As a result, the user can, for example, accurately and efficiently perform the phase analysis based on the clustering results of the data points on the scatter diagram. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a structural diagram of the main part of the EPMA according to the first embodiment of the present invention.

[0038] Figure 2 It is a flowchart showing an example of the clustering analysis parameter adjustment process in the EPMA of the first embodiment.

[0039] Figure 3 It is a diagram showing an example of a histogram of the intensity values of elements with a narrow measurement range.

[0040] Figure 4 It is a diagram showing an example where a pseudo cluster is detected as a result of automatically clustering the data points on the binary scatter diagram.

[0041] Figure 5 It is shown for Figure 4 It is a diagram showing the result of clustering the data points on the binary scatter diagram shown after adjusting the clustering analysis parameters.

[0042] Figure 6 It is a structural diagram of the main part of the EPMA according to the second embodiment of the present invention.

[0043] Figure 7 It is a flowchart showing an example of the phase intensity value data selection process in the EPMA of the second embodiment.

[0044] Figure 8It is a diagram showing an example where the result of automatically clustering data points on a ternary scatter plot is detected as a spurious cluster.

[0045] Figure 9 It is a diagram showing an example of a histogram of phase intensity values.

[0046] Figure 10 It is shown for Figure 9 A diagram of a histogram of phase intensity values after excluding outliers on the high-intensity side from the phase intensity values shown.

[0047] Figure 11 It is shown for Figure 8 A diagram showing the result of clustering after performing phase intensity value data selection processing on the data points on the ternary scatter plot shown.

[0048] Figure 12 It is a diagram showing an example of the result of automatically clustering a binary scatter plot.

[0049] Explanation of reference numerals

[0050] 1: Electron beam irradiation unit; 100: Electron gun; 2: Specimen stage; 3: Specimen; 4: Spectrometer crystal; 5: X-ray detector; 7: Specimen stage drive unit; 8: Analysis control unit; 9: Data processing unit; 90: Element intensity calculation unit; 91: Data storage unit; 92: Scatter plot creation unit; 93: Cluster analysis parameter adjustment unit; 94: Cluster analysis unit; 95: Cluster region detection unit; 96: Display processing unit; 97: Ternary scatter plot creation unit; 98: Phase intensity value data selection processing unit; 10: Central control unit; 11: Operation unit; 12: Display unit. Detailed implementation mode

[0051] [First implementation mode]

[0052] The EPMA of the first implementation mode of the surface analysis device according to the present invention will be described with reference to the accompanying drawings. Figure 1 It is a structural diagram of the main part of the EPMA of the first implementation mode.

[0053] As Figure 1 shown, the electron beam irradiation unit 1 includes an electron gun 100, deflection coils (not shown), etc., and is used to irradiate a specimen 3 placed on the specimen stage 2 with a micro-diameter electron beam. The specimen 3 receives this electron beam and emits characteristic X-rays with wavelengths unique to the elements from the surface of the specimen 3. In addition, secondary electrons, etc. are also emitted from the surface of the specimen 3.

[0054] The characteristic X-rays emitted from the sample 3 are wavelength dispersed by the spectroscopic crystal 4, and the diffracted X-rays of a specific wavelength are detected by the X-ray detector 5. The electron beam irradiation position on the sample 3, the spectroscopic crystal 4 and the X-ray detector 5 are always located on the Rowland circle. The spectroscopic crystal 4 is moved linearly and tilted by a driving mechanism (not shown), and the X-ray detector 5 rotates in conjunction with the movement. Thus, the wavelength scanning of the X-rays to be analyzed is achieved in a manner that satisfies the Bragg diffraction condition, that is, while maintaining the same state in which the incident angle of the characteristic X-rays incident on the spectroscopic crystal 4 and the exit angle of the diffracted X-rays emitted from the spectroscopic crystal 4. The detection signal of the X-ray intensity obtained by the X-ray detector 5 is input to the data processing unit 9.

[0055] The sample stage 2 can be moved in two mutually orthogonal axial directions, the X-axis and the Y-axis, by the sample stage driving unit 7, and the irradiation position of the electron beam on the sample 3 can be two-dimensionally scanned by this movement. In addition, the irradiation position of the electron beam on the sample 3 can be scanned by deflecting the emission direction of the electron beam in the electron beam irradiation unit 1 without moving the sample stage 2.

[0056] The data processing unit 9 includes an element intensity calculation unit 90, a data storage unit 91, a scatter diagram production unit 92, a cluster analysis parameter adjustment unit 93, a cluster analysis unit 94, a cluster region detection unit 95, and a display processing unit 96 as functional modules. In addition to controlling the operation of the sample stage driving unit 7, the analysis control unit 8 also controls the operation of the driving mechanism that moves the spectroscopic crystal 4 and the X-ray detector 5 to perform analysis on the sample 3. The central control unit 10 is responsible for the overall control of the device and input and output processing. The central control unit 10 is connected to a display unit 12 and an operation unit 11 including a keyboard, a mouse (or other pointing devices).

[0057] Furthermore, for example, all or part of the central control unit 10 , the analysis control unit 8 , and the data processing unit 9 may be constituted by a personal computer, and each function can be realized by the computer executing dedicated control / processing software installed in the computer.

[0058] When performing element mapping analysis in the EPMA of the present embodiment, the analysis control unit 8 operates the sample stage drive unit 7 and the like in the following manner: the position of the spectroscopic crystal 4 is fixed corresponding to the characteristic X-ray wavelength of the target element, and the irradiation position (micro region) of the electron beam is changed in a predetermined order in a predetermined (usually specified by the analyst) two-dimensional region on the sample 3, while repeatedly detecting the characteristic X-rays and secondary electrons. Furthermore, when the acquisition of the intensity distribution of one element is completed, the same measurement is performed on other target elements.

[0059] The element intensity calculation unit 90 obtains the intensity (concentration) of the target element for each micro region on the sample 3. The intensity data is stored in the data storage unit 91. In addition, when an energy dispersive X-ray spectrometer is used, the element intensity calculation unit 90 prepares an X-ray spectrum for each micro region in the two-dimensional region, detects a peak of a specific wavelength corresponding to the target element on the X-ray spectrum, and calculates the intensity (concentration) of the target element by obtaining the peak intensity.

[0060] When the measurement of all the micro-regions in the two-dimensional area on the sample 3 is completed and the analyst performs a predetermined operation through the operation unit 11, the scatter diagram creation unit 92 reads the predetermined data from the data storage unit 91 and creates a binary scatter diagram showing the relationship between the intensities of the predetermined two elements. Each data point on the binary scatter diagram corresponds to a micro-region on the sample 3. Therefore, for example, if the measurement is performed on 1000 micro-regions on the sample 3, the number of data points marked on the scatter diagram is 1000.

[0061] The cluster analysis unit 94 performs clustering on all data points on the created scatter diagram according to a predetermined algorithm, and assigns a label to each data point indicating which of one or more clusters the data point belongs to or whether the data point does not belong to any cluster.

[0062] It is known that there are various methods for cluster analysis. Generally, in the clustering of data points on such a scatter diagram, the distance between data points is used for clustering. In the scatter diagram obtained by surface analysis such as EPMA, there are often parts where data points exist at extremely high density and parts where data points exist at low density. In the part where data points exist at high density, if the data points are not separated into independent clusters even if the distance between the data points is relatively short, a cluster with a very large number of data points will be formed. On the contrary, in the part where data points exist at low density, if the data points are not included in the same cluster even if the distance between the data points is relatively long, a large number of clusters with very few data points will be formed. In order to cope with such a situation, here, clustering adopts the method of density-based hierarchical cluster analysis disclosed in non-patent document 2. This method is a method obtained by improving the general density-based cluster analysis disclosed in non-patent document 1. According to the research of the inventor, the data points in the scatter diagram obtained by EPMA can be clustered very well.

[0063] In the density-based cluster analysis including the above-mentioned density-based hierarchical cluster analysis, in order to determine a set of data points on a scatter plot as a cluster, the following two parameters need to be determined in advance.

[0064] (1) Minimum cluster size: the minimum number of data points required to be determined as a cluster (the minimum number of data points that constitute one cluster).

[0065] (2) Distance threshold ε: The distance threshold used to determine that two adjacent clusters are different clusters. Automatically integrates multiple clusters whose distances are closer than this threshold.

[0066] In order to perform good cluster detection on the scatter plot, the above parameters need to be set to appropriate values. However, since it is troublesome for the user (analyst) to set these parameters separately, the values determined through experiments at the manufacturer are set as default values for each parameter, and the user can manually change this value.

[0067] When the intensity ranges of the two elements reflected on the two-dimensional scatter plot are relatively close and the measurement ranges (intensity intervals) are of the same degree, even if the default value is used as the distance threshold ε, roughly appropriate clustering can be performed. However, as described above, when there is a large difference in the abundance (concentration) of the two elements and the measurement range of the element with a small abundance is extremely narrow, it is easy to detect the resulting characteristic pseudo-clusters. If the distance threshold ε is not adjusted to an appropriate value according to the spatial distribution of the data points, multiple pseudo-clusters cannot be integrated into one. Therefore, in fact, before performing the clustering analysis, the clustering analysis parameter adjustment unit 93 adjusts the distance threshold ε as follows. Figure 2 It is a flowchart showing an example of the clustering analysis parameter adjustment process.

[0068] First, in order to grasp the distribution of the intensity values of the element with a narrow measurement range ( Figure 4 Mn in the example), the clustering analysis parameter adjustment unit 93 creates a histogram of the intensity values of this element (step S1). At this time, the number of steps T of the intensity values is appropriately determined. Figure 3 shows the histogram corresponding to the data points on the two-dimensional scatter plot shown in Figure 4 . Here, the number of steps T is set to 1000. As shown in Figure 3 , it can be seen that the histogram reflecting the data points with a narrow measurement range and only discrete existence has a very high frequency in a specific step and has a comb-like gap between multiple specific steps. Since the frequency distribution of the intensity values shows such a characteristic tendency, pseudo-clusters are detected in the two-dimensional scatter plot as described above.

[0069] In density-based clustering analysis, the part with a higher degree of aggregation of data points on the scatter plot, that is, the part with a higher frequency, is more likely to form clusters. Therefore, the clustering analysis parameter adjustment unit 93 detects the maximum value, that is, the peak, in the above histogram, finds the step corresponding to this peak (step S2), and determines the step with the largest frequency in the step representing this maximum value (step S3). In Figure 3 , the step with the largest frequency in the step representing the maximum value is shown by a downward arrow.

[0070] Next, the clustering analysis parameter adjustment unit 93 calculates the number of consecutive steps NL with a frequency of 0 between the step representing the largest maximum value and the step representing the nearest maximum value, where the latter step is located below the former step (the side with a smaller intensity value) in the histogram (step S4). Similarly, the clustering analysis parameter adjustment unit 93 calculates the number of consecutive steps Nu with a frequency of 0 between the step representing the largest maximum value and the step representing the nearest maximum value, where the latter step is located above the former step (the side with a larger intensity value) in the histogram (step S5). That is, the number of consecutive steps with a frequency of 0 is calculated on both sides of the step representing the largest maximum value.

[0071] After that, the clustering analysis parameter adjustment unit 93 compares the number of consecutive steps Nu and the number of consecutive steps NL calculated in steps S4 and S5, and determines the larger one as the number of consecutive steps N (step S6). Then, it determines whether the number of consecutive steps N is greater than the distance threshold ε at this time point (step S7). If the number of consecutive steps N is less than or equal to the distance threshold ε, there is no need to correct the threshold ε, so the value is maintained (step S9), and the process ends. On the other hand, when the number of consecutive steps N is greater than the distance threshold ε, the clustering analysis parameter adjustment unit 93 corrects the value of the threshold ε using the following formula (1) (step S8), and then ends the process.

[0072] ε = (number of consecutive steps N / total number of steps T) + correction constant K …(1)

[0073] The correction constant K can be set to a value appropriately determined through experiments. Here, it is set to 0.002.

[0074] If formula (1) is used, it reflects the length of the interval with a frequency of 0 existing before and after the step representing the highest frequency, and the distance of the region where the data points concentrated in the step representing the highest frequency in the two-dimensional scatter plot are recognized as independent clusters becomes longer. Thus, it is easy to integrate multiple clusters with small interval distances into one cluster.

[0075] After that, the clustering analysis unit 94 performs clustering on the data points on the two-dimensional scatter plot using the parameters corrected as described above. As a result, a label indicating which cluster or whether it does not belong to any cluster among one or more clusters is assigned to each data point on the two-dimensional scatter plot. Since each data point is simply assigned a label in this state, it is difficult to handle it as the region occupied by the clusters in the scatter plot. Therefore, the cluster region detection unit 95 uses a suitable method such as the convex hull method to delimit a polygonal cluster region that includes all or most of the data points belonging to each cluster. In addition, in the Figure 4 shown two-dimensional scatter plot, the linear cluster region extending vertically and the rectangular region containing a large number of data points drawn in the lower region of the figure are also obtained through the above cluster region detection process.

[0076] Figure 5 This is a binary scatter diagram showing the results obtained by performing cluster analysis and cluster region detection after performing the above-described cluster analysis parameter adjustment process on the same data points as those in the binary scatter diagram shown in Figure 4 . According to this Figure 5 , it can be seen that in Figure 4 , the parts that are divided into multiple linearly extending cluster regions are respectively integrated into each large cluster region. On the other hand, in the lower region of the scatter diagram where the density of the data points is not originally very high, the shape of the cluster region does not change. Thus, in the EPMA of the present embodiment, even when there is a large difference in the abundance of the two target elements, especially when the abundance of one element is small and its deviation is also small, resulting in a narrow measurement range, it is possible to avoid detecting pseudo clusters in the binary scatter diagram and perform accurate clustering.

[0077] In addition, in the EPMA of the above-described embodiment, in the Figure 2 parameter adjustment process shown in Figure 2 , only the front and back of the step representing the maximum frequency are considered. However, it is also possible to obtain the number of consecutive steps with a frequency of 0 for all steps of the histogram (that is, the front and back of all maximum value positions), and set the maximum value of this number of consecutive steps as the consecutive step number N. However, in most cases, even if such a process is performed, the selected consecutive step number N is the same as the result of the Figure 2 process shown in

[0078] [Second Embodiment]

[0079] Next, the EPMA of the second embodiment of the surface analysis apparatus according to the present invention will be described with reference to the accompanying drawings. Figure 6 This is a structural diagram of the main part of the EPMA of the second embodiment. In Figure 6 , the same reference numerals are given to the constituent elements that are the same as or equivalent to the constituent elements of the Figure 1 device shown in

[0080] The basic structure of the EPMA of the second embodiment is the same as that of the EPMA of the first embodiment, except that in the data processing unit 9, a ternary scatter diagram production unit 97 is provided instead of the scatter diagram production unit 92, and a phase intensity value data selection processing unit 98 is provided instead of the cluster analysis parameter adjustment unit 93.

[0081] In the EPMA of the present embodiment, as in the EPMA of the first embodiment, under the control of the analysis control unit 8, analysis is performed on a large number of minute regions within a two-dimensional region on the specimen 3. The element intensity calculation unit 90 acquires intensity data reflecting the abundance of the target element for each minute region within the two-dimensional region on the specimen 3. This intensity data is stored in the data storage unit 91.

[0082] When the analyst performs a prescribed operation through the operation unit 11, the ternary scatter diagram creation unit 97 reads prescribed data from the data storage unit 91 and creates a ternary scatter diagram showing the relationship of the intensities of the specified three elements. Each data point on the ternary scatter diagram corresponds to a minute region on the specimen 3. The display processing unit 96 displays the created ternary scatter diagram on the screen of the display unit 12. As Figure 8 shown, when a set of linear data points extending radially is observed in the ternary scatter diagram, there is a high possibility that precise clustering cannot be performed even with automatic clustering analysis. Therefore, when the analyst uses the operation unit 11 to perform a prescribed operation, the phase enhanced intensity value data selection processing unit 98 performs the following data selection processing. Figure 7 This is a flowchart of this data selection processing.

[0083] The phase enhanced intensity value data selection processing unit 98 calculates the sum value of the intensities of the three elements ( Figure 8 Fe, Mg, and K in the example) shown in the ternary scatter diagram for each minute region (hereinafter referred to as "intensity sum value"), and excludes data of a prescribed ratio as outliers in descending order of the phase enhanced intensity value (step S11). The reasons for excluding outliers and the specific method are as follows.

[0084] Figure 9 This is Figure 8 a histogram of the phase enhanced intensity values of all the data points marked in the ternary scatter diagram shown (that is, the minute regions on the specimen 3). Thus, within a very narrow intensity range near the phase enhanced intensity value = 0, a considerable proportion of the data is uneven and forms a peak. It is assumed that such a large amount of data with a small phase enhanced intensity value is the main factor for the set of linear data points extending radially in the ternary scatter diagram. In order to eliminate it, all or part of this data needs to be excluded. However, in Figure 8 the histogram shown, the data extends to a large phase enhanced intensity value with a low frequency, so the peak representing a high frequency becomes a shape compressed in the horizontal axis direction, and it is difficult to determine the threshold value for screening the data to be excluded.

[0085] In the histogram, it is assumed that the frequency of data with large phase enhancement values is low, and in addition, this data is discretely present on the horizontal axis. Therefore, if such data with large phase enhancement values is temporarily excluded and a histogram is remade, the state of the peaks in the region with small phase enhancement values, that is, the distribution state of the data in the histogram, can be grasped in more detail. Therefore, as an example here, an outlier detection method based on quartiles, which is often used in statistics, is used to exclude outliers with large phase enhancement values.

[0086] Generally, in outlier detection based on quartiles, the interquartile range (IQR) is used to find outliers. The interquartile range (IQR) is a value obtained by subtracting the first quartile (Q1) corresponding to 25% of the total quantity from the third quartile (Q3) corresponding to 75% of the total quantity when all data is arranged in ascending order. Specifically, usually, the following formula is used to find the lower boundary and the upper boundary, and the data outside them is set as outliers.

[0087] Lower boundary = Q1 - IQR × 1.5

[0088] Upper boundary = Q3 + IQR × 1.5

[0089] However, here, outliers with small values are not needed. Therefore, in order to remove data with large values, data with intensities above the upper boundary is excluded. In addition, the detection method for outliers with large values is not limited to the above method. For example, other outlier detection methods such as the Smirnov-Grubbs test can also be used.

[0090] Next, the phase enhancement value data selection processing unit 98 makes a histogram based on the phase enhancement value data after excluding outliers as described above (step S12). Then, a specified algorithm is used in this histogram to detect the positions (steps) that become maximum (peaks) and minimum (step) (step S13). Figure 10 is based on the data after excluding outliers from the phase enhancement value data that is the basis of the Figure 9 shown histogram. Figure 10 In, the detected maximum values are represented by black dots, and the minimum values are represented by white dots. In addition, here, a certain detection width is set so that the increase and decrease of the frequency in a narrow intensity range are not recognized as maximum / minimum. Therefore, for example, the peak closest to intensity = 0 is not detected as a maximum.

[0091] The phase enhancement value data selection processing unit 98 determines the nearest minimum value above the maximum value representing the maximum frequency (step S14), and excludes from all the phase enhancement value data (including the outliers excluded in step S11) the phase enhancement value data included in the intensity range from the minimum intensity, that is, from intensity = 0 to the determined minimum value (step S15). In Figure 10 In the case of the example shown, all the phase enhancement value data with a phase enhancement value of 156 or less are excluded. Thus, in Figure 10 In the histogram shown, all the data forming the peak with the maximum frequency are excluded.

[0092] In addition, among the phase enhancement value data after excluding the outliers in step S11, the maximum intensity sometimes changes significantly (decreases) compared to the original data. In Figure 9 、 Figure 10 In the example shown, the number of steps of the phase enhancement value when creating the histogram is set to 256. However, assuming that the maximum intensity of the data after excluding the outliers is less than 256, it is necessary to perform processing to adjust the number of steps of the intensity value according to this situation. This is because: if such adjustment of the number of steps is not performed, steps with a frequency of 0 are generated comb-like in the created histogram, and the position of the minimum value cannot be accurately obtained.

[0093] In the above process, the phase enhancement value data selection processing unit 98 selects the data to be the clustering target by excluding the data with small phase enhancement values and high frequencies. The clustering analysis unit 94 performs clustering on the data points on the ternary scatter plot thus selected, for example, using density-based hierarchical clustering analysis. Thereby, a label indicating which cluster or whether not belonging to any cluster among one or more clusters each data point on the ternary scatter plot belongs to is assigned. The cluster region detection unit 95 delimits a polygon-shaped cluster region that includes all or most of the data points belonging to each cluster, for example, using a suitable method such as the convex hull method.

[0094] Figure 11 is a ternary scatter plot showing the result of clustering after performing the above-described phase enhancement value data selection processing on the data points on the ternary scatter plot shown in Figure 8 to reduce the data points. In Figure 11 , Figure 8 In the ternary scatter plot shown, the distribution of the radially extending linear data points that appeared in the ternary scatter plot disappears, and no clusters expected to be spurious clusters are detected. Based on this, it is estimated that the clustering has been appropriately performed.

[0095] Thus, in the EPMA of the second embodiment, it is possible to exclude the set of noise data points that appear in the ternary scatter plot from the objects to be subjected to the clustering process, thereby suppressing the detection of incorrect cluster regions. As a result, it is possible to improve the detection accuracy of the clusters on the ternary scatter plot and improve the accuracy and efficiency of the phase analysis using this method.

[0096] In addition, in the above description, in the Figure 10 histogram shown, the data included in the intensity range from intensity = 0 to the nearest minimum value above the maximum value of the maximum frequency is excluded, but the intensity range of the data to be excluded can be appropriately changed according to the distribution status of the data. For example, it may be set such that the user can select the intensity range of the data to be excluded, or it may be possible to perform clustering on multiple data groups after excluding different intensity ranges of data and obtain multiple clustering results, and summarize and present these results to the user.

[0097] In addition, it is not necessary to always perform the above-described data selection process. Therefore, the data selection process may be performed according to the user's operation as described above, or it may be set to automatically perform the data selection process based on the clustering result or the like.

[0098] In addition, although the above-described first and second embodiments are EPMA, the present invention can be comprehensively applied to various analysis devices such as SEM and X-ray fluorescence analysis devices that can obtain signals reflecting the amounts of elements or components (compounds, etc.) in a large number of minute regions in a one-dimensional or two-dimensional region on a sample. That is, the present invention does not particularly limit the measurement method and the analysis method itself, and any analysis device capable of performing mapping analysis may be used.

[0099] In addition, the above-described embodiments are merely examples of the present invention, and of course, appropriate deformations, modifications, additions, etc. within the scope of the gist of the present invention are also included in the scope of the claims of the present application.

[0100] [Various Modes]

[0101] Those skilled in the art should understand that the above-exemplified embodiments are specific examples of the following modes.

[0102] (First Item) One mode of the surface analysis device according to the present invention includes:

[0103] A measurement unit that respectively obtains signals reflecting the amounts of a plurality of components or elements to be analyzed at a plurality of positions on a sample;

[0104] A scatter plot creation unit that creates a binary scatter plot based on the measurement results obtained by the measurement unit;

[0105] A clustering analysis unit that uses a density-based clustering analysis method to cluster the points in the binary scatter plot; and

[0106] A parameter adjustment unit that uses the value distribution information of the signal values of a certain component or element in the binary scatter plot to adjust the distance threshold, which is one of the parameters to be set in the density-based clustering analysis.

[0107] (Second item) In the surface analysis device described in the first item, it can be set that: the parameter adjustment unit uses the signal value distribution of the component or element with a narrower signal value range in the binary scatter plot to adjust the distance threshold.

[0108] For example, when there is a large difference in the amounts of two elements contained in a sample, since the measurement range of the element with a smaller amount is very narrow, sometimes a set of multiple linear data points appears extremely close on the binary scatter plot. According to the surface analysis device described in the first and second items, such a set of multiple linear data points will not be misidentified as independent clusters, but can be identified as one cluster. That is, according to the surface analysis device described in the first and second items, it is possible to suppress the detection of false clusters when automatically clustering the marked data points in the binary scatter plot, thereby improving the accuracy of clustering of these data points. In other words, it improves the accuracy of clustering of micro-regions on the sample based on the amounts or concentrations of multiple components or elements. As a result, the user can, for example, accurately perform phase analysis based on the clustering results.

[0109] (Third item) In the surface analysis device described in the second item, it can be set that: the parameter adjustment unit creates a histogram of the signal values of the component or element, and adjusts the distance threshold based on the frequency distribution before and after at least one signal value step representing the maximum value in the histogram.

[0110] (Fourth item) Additionally, in the surface analysis device described in the third item, it can be set that: the parameter adjustment unit adjusts the distance threshold based on the frequency distribution before and after the signal value step representing the maximum value with the largest frequency in the histogram.

[0111] According to the surface analysis device described in the third and fourth items, it is possible to extract the region where data points are characteristically distributed and are easily misdetected as false clusters in the binary scatter plot, and appropriately determine the clustering parameter (distance threshold) in a way that avoids detecting false clusters in this region. In addition, such processing for adjusting parameters is also simple, so the processing does not take time. For example, the clustering result can be quickly displayed.

[0112] (Fifth item) In the surface analysis device described in any one of the first to fourth items, it can be set that: the clustering analysis unit performs density-based hierarchical clustering analysis.

[0113] The surface analysis device according to the fifth item can cluster data points on a binary scatter diagram well, which is made based on data collected by, for example, EPMA. Thus, the user can accurately perform phase analysis based on the clustering results, for example.

[0114] (Sixth item) Additionally, another aspect of the surface analysis device according to the present invention includes:

[0115] A measurement unit that respectively obtains signals reflecting the amounts of a plurality of components or elements to be analyzed at a plurality of positions on a sample;

[0116] A scatter diagram creation unit that creates a ternary scatter diagram based on the measurement results obtained by the measurement unit;

[0117] A data point selection unit that uses the distribution information of the added signal values obtained by adding the signal values of three components or elements corresponding to the data points in the ternary scatter diagram to exclude data points within a specified signal value range with relatively small added signal values from all the data points existing on the ternary scatter diagram; and

[0118] A clustering analysis unit that uses a density-based clustering analysis method to cluster the data points in the ternary scatter diagram that have not been excluded by the data point selection unit.

[0119] For example, when there are a large number of data points where the ratios of the signal values of three elements contained in the sample are approximately the same, a radial linear data point distribution appears on the ternary scatter diagram, which may sometimes cause the detection of false clusters. The surface analysis device according to the sixth item can eliminate the characteristic distribution of data points that cause such false clusters. Thus, it is possible to suppress the detection of false clusters when automatically clustering the data points marked in the ternary scatter diagram, thereby improving the accuracy of clustering of these data points, that is, improving the accuracy of clustering of minute regions on the sample based on the amounts or concentrations of multiple components or elements. Thus, the user can accurately perform phase analysis based on the clustering results, for example.

[0120] (Seventh item) In the surface analysis device according to the sixth item, it can be set that: the data point selection unit creates a histogram of the added signal values and uses the steps representing the detected maximum and / or minimum values in the histogram to determine the signal value range of the data to be excluded.

[0121] The surface analysis device according to the sixth item can accurately exclude data points with relatively small added signal values and high frequencies, which are likely to be the causes of pseudo clusters. Thus, not only can data points that are likely to be the causes of pseudo clusters be excluded, but also the accidental exclusion of data points that are not such can be avoided, thereby accurately delimiting the cluster region on the ternary scatter plot.

[0122] (Eighth item) In the surface analysis device according to the sixth or seventh item, it can be set that the clustering analysis unit performs density-based hierarchical clustering analysis.

[0123] The surface analysis device according to the eighth item can well cluster data points on a ternary scatter plot made from data collected using, for example, EPMA. Thus, the user can accurately perform phase analysis based on the clustering results.

Claims

1. A surface analysis device, comprising: a measurement unit that respectively obtains signals reflecting the amounts of a plurality of components or elements to be analyzed at a plurality of positions on a sample; a scatter diagram creation unit that creates a two-dimensional scatter diagram based on the measurement results obtained by the measurement unit; a clustering analysis unit that clusters the points in the two-dimensional scatter diagram using a density-based clustering analysis method; and a parameter adjustment unit that creates a histogram of the distribution information of the signal values of the components or elements with a narrow signal value range in the two-dimensional scatter diagram, and calculates the maximum number of consecutive steps with a frequency of 0 before and after the step representing the maximum value of the frequency in the histogram or between two steps with the maximum frequency among all steps, and adjusts the distance threshold according to the number of consecutive steps. The distance threshold is one of the parameters to be set in the density-based clustering analysis. When the number of consecutive steps is greater than the distance threshold, a value obtained by adding a prescribed correction constant to the value obtained by dividing the number of consecutive steps by the total number of steps in the histogram is used as the distance threshold.

2. The surface analysis device according to claim 1, wherein the parameter adjustment unit calculates the number of consecutive steps with a frequency of 0 before and after the step representing the maximum value of the frequency in the histogram, and sets the larger of the two numbers of consecutive steps as the number of consecutive steps.

3. The surface analysis device according to claim 1 or 2, wherein the correction constant is a value determined by experiments.

4. The surface analysis device according to claim 3, wherein the correction constant is 0.

002.

5. The surface analysis device according to any one of claims 1, 2, and 4, wherein the clustering analysis unit performs density-based hierarchical clustering analysis.

6. The surface analysis device according to claim 3, wherein the clustering analysis unit performs density-based hierarchical clustering analysis.

Images