Method for determining characteristic curve from which background is removed

The segment regression piecewise linear function method with BIC/AIC and slope asymmetry indices automates the removal of background components in characteristic functions, addressing inefficiencies and uncertainties in existing manual methods, achieving precise and stable results.

JP2024166064APending Publication Date: 2024-11-28NAT INST FOR MATERIALS SCI
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
JP2024008113
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Priority Date
2023-05-16
Filing Date
2024-01-23
Publication Date
2024-11-28

AI Technical Summary

Technical Problem

Existing methods for calculating characteristic functions that follow power laws or exponential laws are inefficient and prone to human intervention, leading to uncertainty and ambiguity due to the manual removal of superimposed background components.

Method used

The application of a segment regression piecewise linear function method to determine the number of segments, slopes, and boundary points in characteristic functions, using criteria like Bayesian Information Criterion (BIC) and Akaike Information Criterion (AIC) to automatically identify and remove linear background components, employing indicators such as average slope and asymmetry index to stabilize the calculation.

Benefits of technology

This method allows for stable, automated, and accurate removal of background components from characteristic curves, ensuring high precision without human intervention.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024166064000001_ABST
    Figure 2024166064000001_ABST
Patent Text Reader

Abstract

To provide a method for easily determining a power low, exponential law characteristic curve with high background removal accuracy independent of artificiality.SOLUTION: A method for determining a characteristic curve includes the steps of: applying a segment recurrence piecewise linear function to a characteristic function Y=F(x) to determine segments and the number of segments according to an information standard; determining the inclination a[i] of a curve of the characteristic function for every segment; determining the average inclination aveS of the characteristic function; determining an asymmetry index ASY from the average value and the median value of values of data throughout all the sections; determining a determination standard DS by using the average inclination aveS and the asymmetry index ASY; when a linear component of the inclination a[i] is a[i]<DS, determining it as a background component, and when the linear component is a[i]≥DS, determining it as a non-background component; and subtracting F(0) and the background component from the characteristic function.SELECTED DRAWING: Figure 1
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to a method for determining a characteristic curve with background removed. [Background technology]

[0002] Many physical phenomena follow power laws or exponential laws. Characteristic indices derived from characteristic functions that follow these laws are very useful for evaluating physical properties and their industrial applications. Here, examples of indexes of characteristics conforming to the power law include the critical current density Jc of superconducting materials, the photoemission (photoelectric effect) threshold, and the band gap, while examples of indexes of characteristics conforming to the exponential law include the Schottky threshold of semiconductors. Examples of applications of the power law include prediction of degradation of transistor characteristics as disclosed in Patent Document 1, and application to viscosity measurement as disclosed in Patent Document 2. For this reason, characteristic functions that follow power laws or exponential laws are important, but there are two challenges to overcome in calculating them: removing superimposed background components and establishing an automated calculation method that is not dependent on human intervention.

[0003] Measured data contains various background components that are on the original characteristic curve. Most of the background is linear, and it is important to remove it. Conventionally, this calculation work was done manually. However, doing it manually is not only inefficient, but also generates artificial uncertainty and ambiguity. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2001-352059 A [Patent Document 2] Patent Publication No. 2022-185568 [Non-patent literature]

[0005]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0006] The problem of the present invention is to solve the above-mentioned conventional problems when calculating the characteristic function, to stably calculate without being influenced by humans, and to provide a method for obtaining a characteristic curve with high background removal accuracy while being a simple method.

Means for Solving the Problems

[0007] The configuration of the present invention for solving the problems is shown below. (Configuration 1) Applying the segment regression piecewise linear function method to the characteristic function Y = F(x) to be subjected to background removal, and performing segmentation based on the information criterion to obtain the number of segments m and the segmentation boundary points (BP[j]) (where j is an integer from 0 to m, BP[0] is the minimum value of x, and BP[m] is the maximum value of x); For each segment i (i is an integer from 1 to m) obtained based on the information criterion, obtaining the slope a[i] and F(0) of the curve of the characteristic function Y = F(x); When there is an a[i] that satisfies a[i] ≥ a[i + 1], setting the minimum value of i as n, and when i satisfies a[i] < a[i + 1] from 1 to m, setting n = m and obtaining the calculation target interval as the interval up to segment n; As the first index, in the interval from 1 to n of the segments, using the maximum value Y max , minimum value Y min , maximum value x max and minimum value x min of the x, (Y max - Y min ) / (x max - x minThe step of obtaining the average slope (aveS) of the characteristic function Y = F(x) defined by As a second index, the average value (D ave ) and the median (D cnt ) of the values of the data (Y) through the sections from 1 to n of the section, and an asymmetry index (ASY = D cnt / D ave ) is obtained. Using a predetermined reference value L from among the average slope (aveS), the asymmetry index (ASY), and values of 0.80 or more and 0.95 or less as the determination criterion (DS), When 0 ≦ ASY < L, DS = aveS × ASY, When ASY ≧ L, DS = aveS The step of obtaining as When the linear component of the slope a[i] (i is an integer from 1 to n) is a[i] < DS, it is a background component When a[i] ≧ DS, it is selected and determined as a non-background component. A method for obtaining a characteristic curve, including the steps of subtracting F(0) and the background component from the characteristic function Y = F(x). (Configuration 2) Applying the segment regression piecewise linear function method to the characteristic function Y = F(x) to be subjected to background removal, and obtaining the segmentation, the number of segments m, and the segmentation boundary points (BP[j]) (where j is an integer from 0 to m, BP[0] is the minimum value of x, and BP[m] is the maximum value of x) based on the information criterion. The step of obtaining the slope (a[i]) of the curve of the characteristic function Y = F(x) for each segment i (i is an integer from 1 to m) obtained based on the information criterion as the slope between the start point and the end point of each segment. When there is an a[i] that satisfies a[i] ≧ a[i + 1], the minimum value of i is set to n, and when i satisfies a[i] < a[i + 1] from 1 to m, n = m, and the calculation target interval is obtained as the interval up to segment n. As a first index, in the interval from 1 to n of the segment, the maximum value Y of the Y max , the minimum value Ymin and the maximum value x max and the minimum value x min to obtain the average slope (aveS) of the characteristic function Y = F(x) defined by (Y max - Y min ) / (x max - x min ); and As a second index, from the ratio of the average value (D ave ) and the median value (D cnt ) of the data (Y) through the sections from 1 to n of the said sections, calculate the asymmetry index (ASY) given by the following (Equation 2); Using the average slope (aveS), the asymmetry index (ASY), and a reference value L determined in advance from values of 0.80 or more and 0.95 or less as the determination criterion (DS), when 0 ≦ ASY < L, calculate according to the following (Equation 3), and when ASY ≧ L, calculate according to the following (Equation 4); Compare the said a[i] (where i is an integer from 1 to n) with the determination criterion (DS). When, within the range where i is from 1 to n, there is an i such that a[i] ≧ DS, set the smallest i that satisfies a[i] ≧ DS as t, and when a[i] < DS, set t = n, and obtain the maximum section division t indicating the linear background division section range; A method for obtaining a characteristic curve, comprising the step of obtaining a characteristic curve Y' from which the linear background has been removed according to the following (Equation 5) to (Equation 7). aveS = (Y max - Y min ) / (x max - x min ) ···(Equation 1) ASY = D cnt / D ave ···(Equation 2) DS = aveS × ASY ···(Equation 3) DS = aveS ···(Equation 4) Y' = F(x) - F background ···(Equation 5)

Number

[0008] (Configuration 6) Applying the segment regression piecewise linear function method to the characteristic function Y = F(x) to be background-removed, and performing classification according to the information amount criterion, the number of segments m, and the segmentation boundary points (BP[j]) (where j is an integer from 0 to m, BP[0] is the minimum value x min of x, and BP[m] is the maximum value x max ) of x, and For each segment i (i is an integer from 1 to m) obtained according to the information criterion, obtaining the slope (a[i]) of the curve of the characteristic function Y = F(x) as the slope between the start point and the end point of each segment; If there is an a[i] that satisfies a[i] ≥ a[i + 1], setting the minimum value of i as n, and if i satisfies a[i] < a[i + 1] from 1 to m, setting n = m, and obtaining the calculation target interval as the interval up to segment n; As the first a index, in the interval from 1 to n of the segment, using the maximum value Y max of Y, the minimum value Y min , the maximum value x max of x, and the minimum value x min of x, obtaining the average slope (aveS) of the characteristic function Y = F(x) defined by the following (Equation 1a); As the second a index, setting x min to 0, x max to 1, F(x min ) to 0, F(x maxNormalize (x, Y) with ( ) taken as 1, and obtain (1 - GiNi) calculated from the Gini coefficient (GiNi) obtained from the Lorenz curve. Determine the criterion (DSa) using the average slope (aveS) and the (1 - GiNi) which is the second a index according to the following (Equation 2a). Compare the a[i] (where i is an integer from 1 to n) with the criterion (DSa). When there is an i such that a[i] ≥ DSa within the range of i from 1 to n, let t be the smallest i that satisfies a[i] ≥ DSa. a When a[i] < DSa, then t a = n, and obtain the maximum interval division t indicating the linear background division interval range. a The step of obtaining. A method for obtaining a characteristic curve, which comprises the step of obtaining a characteristic curve Ya' from which the linear background is removed according to the following (Equation 3a). aveS = (Y max - Y min ) / (x max - x min ) ··· (Equation 1a) DSa = aveS × (1 - GiNi) ··· (Equation 2a) Ya' = F(x) - F background,a ··· (Equation 3a)

Number

Number

number

[0009] According to the present invention, there is provided a method for determining a characteristic curve that is not influenced by human intervention and is stable, and that is simple yet has high background removal accuracy. [Brief description of the drawings]

[0010] [Figure 1] FIG. 2 is a flow chart showing a first processing flow of the present invention. [Diagram 2] FIG. 2 is a configuration diagram showing the configuration of a processing device used in the present invention. [Diagram 3] FIG. 11 is a flow chart showing a second processing flow of the present invention. [Figure 4]FIG. 13 is a characteristic diagram showing an example of segmentation performed by applying a segment regression piecewise linear function method to a critical current density function. [Diagram 5] 13 is an example in which the method of the present invention is applied to critical current density measurement data to remove background components. [Figure 6] FIG. 11 is a flow chart showing a third processing flow of the present invention. [Figure 7] FIG. 13 is a flow chart showing a fourth processing flow of the present invention. [Figure 8] 13 is an example in which the method of the present invention is applied to the critical current density measurement data according to Example 3 to remove background components. [Figure 9] (a) shows an example of removing the background component from the critical current density measurement data up to the BP[1] division boundary point, and (b) shows an example of removing the background component up to the BP[2] division boundary point. (c) shows a comparison of the two. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] (Embodiment 1) <Summary> The method of removing the background component consisting of a linear function (linear background component) from the characteristic curve of the present invention is as follows: (1) Dividing the characteristic curve Y=F(x) that is to be subjected to background removal into multiple intervals, approximating each interval with a linear function, and calculating the slope for each interval. In other words, calculating the interval division and the slope of the characteristic curve for each interval, (2) It is characterized by the fact that a selection judgment is made based on an evaluation index and judgment criteria that determine whether or not something is background, and the original characteristic function is derived with the background removed.

[0012] <<Section division>> In the present invention, the characteristic curve is divided and a linear background component appropriate for each division is removed. In carrying out this, a segmented regression piecewise linear function method used for approximating curved surfaces is used (segmented linear regression method). In general, the more the number of divisions, the smaller the deviation between the approximate value and the measured data, i.e., the more the number of divisions, the better the data fitting can be obtained. However, in that case, the problem of heavy processing can occur. By using the information processing standard, it is possible to obtain the optimal number of divisions that provides sufficient fitting.

[0013] Useful information processing criteria include the Bayesian Information Criterion (BIC) or the Akaike Information Criterion (AIC).

[0014] BIC is a statistic that evaluates the complexity and goodness of fit of a model by taking into account the number of parameters estimated by the maximum likelihood method and the number of observed data, and is expressed by the following formula. BIC = -2 × log(L) + k × log(n) Here, L is the likelihood function estimated by the maximum likelihood method, k is the number of parameters in the model, and n is the number of observed data.

[0015] Like BIC, AIC is a statistic that evaluates the complexity and goodness of fit of a model by taking into account the number of parameters estimated by the maximum likelihood method and the number of observed data, and is expressed by the following formula. AIC = -2 × log(L) + 2 × k Here, L is the likelihood function estimated by the maximum likelihood method, and k is the number of parameters in the model.

[0016] Both BIC and AIC are indices that evaluate the fitness of a model by taking into account the complexity and fitness of the model, but BIC tends to favor simpler models because it penalizes the complexity of the model. On the other hand, AIC has a smaller penalty and tends to favor more complex models. Therefore, it is preferable to choose between BIC and AIC depending on the nature of the characteristic curve and the nature of the data. For example, BIC is suitable for characteristic curves with relatively little noise, and AIC is suitable for characteristic curves with a lot of noise.

[0017] <<Evaluation Interval>> The present invention provides a method for removing a linear background from a characteristic curve portion that is accelerating upward to the right. When the characteristic curve Y = F(x) has a maximum value in an intermediate interval, the linear background is removed up to the interval having the maximum value. To perform the process, the slope (a[i]) of the curve of the characteristic function Y = F(x) for each section i (i is an integer from 1 to m) obtained based on the information criterion is determined as the slope between the start point and the end point of each section. If there is an a[i] that satisfies a[i] ≥ a[i + 1], the minimum i that satisfies this relationship is set as n. When i satisfies a[i] < a[i + 1] from 1 to m, n = m, and the calculation target interval is determined as the interval up to section n.

[0018] <<Evaluation Index>> The inventor focused on the fact that curves following power or exponential functions tend to have a larger slope in the curve portion compared to the background component, and derived an evaluation index for background determination. The evaluation index consists of two parts: the average slope (aveS) and the asymmetry index (ASY).

[0019] The average slope (aveS) is the slope defined by (Y max , the minimum value Y min , the maximum value x of x max and the minimum value x min are used, and (Y max -Y min ) / (x max -x min ).

[0020] The asymmetry index (ASY) is defined as the ratio (D ave ) of the average value (D cnt ) and the median value (D cnt ) of the values of the data (Y) through all section divisions (D ave ). If it is a linear function, the mean value and the median value coincide. As it deviates from this straight line, the values of the mean and median deviate. In a characteristic curve with an accelerating upward slope on the right shoulder, the mean value is always larger than the median value. Therefore, when taking the ratio of the mean value to the median value, in the case of 1, it becomes the same as the average slope, and as the power or exponential term increases, it approaches 0. Using this property, it is used as one of the indicators for determining whether the curve represents the original characteristic or the background.

[0021] <<Judgment Criteria>> The judgment criterion (DS) is determined in advance from the average slope (aveS), the asymmetry index (ASY), and a reference value L within the range of 0.80 or more and 0.95 or less. When 0 ≦ ASY < L, DS = aveS × ASY, When ASY ≧ L, DS = aveS and is obtained as follows.

[0022] <<Selection Judgment>> Then, for each section i (where i is an integer from 1 to n) obtained based on the information criterion, the slope (a[i]) of the curve of the characteristic function Y = F(x) is obtained, and the slope a[i] is compared with the judgment criterion DS for each section i. If a[i] < DS, it is a background component. If a[i] ≧ DS, it is selected and judged as a non-background component.

[0023] <<Derivation of Characteristic Curve Excluding Background Component>> Here, Y(0) and the background component are subtracted from the characteristic function Y = F(x). Specifically, the characteristic curve Y´ excluding the background component is derived using the above-mentioned (Equations 5) to (7). Here, CMax(x - BP[i]) is a function that is 0 when x - BP[i] ≦ 0 and x - BP[i] when x - BP[i] > 0. The characteristic curve Y´ derived by this method is calculated stably without being artificially influenced. Although it is a simple method, it has a high background removal accuracy.

[0024] <Procedure, algorithm> The procedure (algorithm) of the present invention will be described below with reference to FIG. 1, which is a flow chart, and FIG. 2, which is a configuration diagram of the processing device 101 used.

[0025] The processing device 101 comprises an input facility 11, a calculation facility 14, a memory facility 15 and an output facility 17, and each facility from the input facility 11 to the output facility 17 is arranged so that information can be transmitted via an information path 16. Measurement data (input data) 12 is input to the input facility 11, and the information is processed by the calculation facility 14 and the memory facility 15 via the information path 16. The result of the information processing is sent to the output facility 17, and output data (background removal characteristic curve) 18 is output.

[0026] Here, the computing equipment 14 is a computing device mainly composed of semiconductor computing devices such as a CPU, an MPU, and a GPU, and the memory equipment 15 is composed of memory devices such as temporary memory devices such as DRAM and SRAM, and long-term memory devices (storage) such as FLASH memory, FRAM (registered trademark), EPROM, hard disks, floppy disks, and magnetic tapes. It is not necessary to have a long-term memory device, but it is useful as an option when performing advanced analysis of threshold data, such as a data reference function. It is also possible to configure the memory equipment 15 only with a long-term memory device without a temporary memory device, but in that case, it takes time to transfer data. Therefore, it is preferable to have both a temporary memory device and a long-term memory device. The information path 16 is a facility that connects the input facility 11, the calculation facility 14, the memory facility 15, and the output facility 17 to transmit data and information, and is generally made up of wiring (optical interconnects are also acceptable).

[0027] The series of steps described below are processed fully automatically by a computer-based processing device 101, making it possible to derive a characteristic curve from which linear background components have been removed in a stable manner without being affected by human intervention.

[0028] Next, each procedure (step) will be explained. First, input data into the input facility 11 (step S10 in FIG. 1). Here, the data may be input all at once by batching, or the measuring device and the threshold value calculation and analysis device 101 may be connected online to input sequentially for each measurement. Next, apply the segment regression piecewise linear function method to the characteristic function Y = F(x) to be background-removed, and obtain the segmentation, the number of segments m, and the segmentation boundary points BP[i] based on the information criterion (step S11). Here, as the information criterion, AIC or BIC can preferably be used. Thereafter, obtain the slope (a[i]) of the curve of the characteristic function Y = F(x) for each segment i (i is an integer from 1 to m) obtained based on the information criterion (step S12). After that, if there is an a[i] that satisfies a[i] ≥ a[i + 1], set the minimum i as n, and if the i from 1 to m satisfies a[i] < a[i + 1], set n = m, and obtain the calculation target interval as the interval up to segment n (step S13). Next, as the first index, in the interval from 1 to n of the segment, the maximum value Y max of the characteristic function Y, the minimum value Y min , the maximum value x max of x, and the minimum value x min are used to obtain the average slope (aveS) of the characteristic function Y = F(x) defined by (Y max -Y min ) / (x max -x min ) (step S14). Also, as the second index, obtain the asymmetry index (ASY = D ave / D cnt ) from the ratio of the average value (D cnt ) and the median value (D ave ) of the values of the data (Y) through the segments from 1 to n (step S15).

[0029] Thereafter, use the average slope (aveS), the asymmetry index (ASY), and a reference value L determined in advance from values of 0.80 or more and 0.95 or less as the determination criterion (DS) When 0 ≤ ASY < L, DS = aveS × ASY (the above-mentioned (Equation 3)), When ASY ≧ L, DS = aveS (the aforementioned (Equation 4)) is obtained as (Step S16).

[0030] Thereafter, it is determined whether the linear component of the slope a[i] (where i is an integer from 1 to n) is a background component or not, when a[i] < DS, it is a background component when a[i] ≧ DS, it is discriminated and determined as a non-background component (Step S17). Thereafter, F(0) and the said background component are subtracted from the characteristic function Y = F(x) (Step S18), and finally, the result is output from the output facility 17 and the process ends (Step S20). By this method, it becomes possible to calculate the characteristic curve Y' from which the linear background component is removed stably and simply without being affected artificially.

[0031] Note that after Step S14 and before Step S15, it is also effective to obtain the division and slope again within the calculation target range obtained in Step S14 and perform the calculation again, that is, after Step S14, limiting the calculation target interval from x = 0 to x = BP[t], returning to Step S11 and performing the calculation again up to Step S20.

[0032] <Procedure 2> The second procedure (algorithm) of the present invention is shown below with reference to FIG. 3 which is a flowchart diagram. First, data is input to the input facility 11 (Step S50 in FIG. 3). Here, the data may be input all at once by being accumulated in a batch form, or may be input sequentially for each measurement by connecting the measuring device and the threshold value calculation and analysis device 101 online. Next, the segment regression division linear function method is applied to the characteristic function Y = F(x) to be subjected to background removal, and the division, the number of divisions m, and the division boundary points BP[i] are obtained based on the information criterion (Step S51). Here, as the information criterion, AIC or BIC can be preferably used. After that, the slope (a[i]) of the curve of the characteristic function Y = F(x) for each section i (where i is an integer from 1 to m) obtained based on the information criterion is determined (step S52). After that, if there is an a[i] that satisfies a[i] ≥ a[i + 1], the smallest i that satisfies this condition is set as n. When the i from 1 to m satisfies a[i] < a[i + 1], n = m is set, and the calculation target interval is determined as the interval up to section n (step S53). Next, as the first index, in the interval from 1 to n of the said section, the maximum value Y max of the characteristic function Y, the minimum value Y min , the maximum value x max of x, and the minimum value x min are used to obtain the average slope (aveS) of the characteristic function Y = F(x) defined by (Y max -Y min ) / (x max -x min ) (step S54). Also, as the second index, the ratio of the average value (D ave ) and the median value (D cnt ) of the data (Y) values through the sections from 1 to n is used to obtain the asymmetry index (ASY = D cnt / D ave ) (step S55).

[0033] After that, the determination criterion (DS) is determined using the average slope (aveS), the asymmetry index (ASY), and a reference value L determined in advance from values between 0.80 and 0.95 as follows: When 0 ≤ ASY < L, DS = aveS × ASY, When ASY ≥ L, DS = aveS (step S56).

[0034] After that, a[i] (where i is an integer from 1 to n) is compared with the determination criterion (DS). When there is an i within the range from 1 to n such that a[i] ≥ DS, the smallest i that satisfies a[i] ≥ DS is set as t. When a[i] < DS, t = n is set, and the maximum section division t indicating the linear background interval range is obtained (step S57).

[0035] Thereafter, the characteristic curve Y' from which the background has been removed is obtained using (Equation 5) to (Equation 7) (step S58), and finally, the result is output from the output device 17 to terminate the process (step S60). Y´=F(x)-F background ...(Formula 5)

[0036]

number

[0037] This method makes it possible to calculate the characteristic curve Y' from which the linear background component has been removed in a stable and simple manner without being influenced by human intervention.

[0038] In the present invention, as described above, the segment regression piecewise linear function method is used to grasp the maximum position of the characteristic curve Y=F(x) and exclude the section after the maximum from the analysis target. Since this step is performed, it is also possible to estimate the initial value of the threshold value using the method of the present invention. In other words, the segment point where the slope is determined to be the background represents the initial value of the threshold value.

[0039] (Embodiment 2) The method of obtaining a characteristic curve with the background removed in the second embodiment uses the asymmetry index (ASY=D cnt / D ave ) is replaced by (1-GiNi), and the other steps are the same as those of the method of the first embodiment. Here, GiNi is the Gini coefficient.

[0040] The detailed procedure is given below with reference to the flow chart diagram of FIG. First, data is input to input equipment 11 (step S150 in FIG. 6). Here, data may be accumulated in a batch format and all data may be input, or the measurement device and threshold calculation analysis device 101 may be connected online and data may be input one by one for each measurement, as in the first embodiment. Next, apply the segment regression piecewise linear function method to the characteristic function Y = F(x) to be background-removed, and perform segmentation based on the information criterion to obtain the number of segments m and the segmentation boundary points BP[i] (step S151). Here, AIC or BIC can preferably be used as the information criterion. After that, obtain the slope (a[i]) of the curve of the characteristic function Y = F(x) for each segment i (where i is an integer from 1 to m) obtained based on the information criterion (step S152). After that, if there is an a[i] that satisfies a[i] ≥ a[i + 1], set the smallest i that satisfies this condition as n. If i from 1 to m satisfies a[i] < a[i + 1], set n = m, and obtain the calculation target interval as the interval up to segment n (step S153). Next, as the first index, in the interval from 1 to n of the said segmentation, the maximum value Y max of the characteristic function Y, the minimum value Y min , the maximum value x max of x, and the minimum value x min are used to obtain the average slope (aveS) of the characteristic function Y = F(x) defined by (Y max - Y min ) / (x max - x min ) (step S154). Also, as the second index, normalize (x, Y) with the minimum value x min of x being 0, the maximum value x max of x being 1, F(x min ) being 0, and F(x max ) being 1, and obtain (1 - GiNi) calculated from the Gini coefficient (GiNi) obtained from the Lorenz curve (step S155). After that, obtain the determination criterion (DSa) by the following (Equation 2a) using the average slope (aveS) and the (1 - GiNi) which is the second index (step S156). After that, compare a[i] (where i is an integer from 1 to n) with the determination criterion (DSa). When there is an i that satisfies a[i] ≥ DSa within the range of i from 1 to n, set the smallest i that satisfies a[i] ≥ DSa as t a , and when a[i] < DSa, set t a= n, the maximum interval t indicating the linear background interval range a is calculated (step S157). Finally, the characteristic curve Ya' from which the linear background has been removed is calculated using the following formula 3a (step S157), and the result is output and the process ends (step S160). aveS=(Y max -Y min ) / (x max -x min )...(Equation 1a) DSa=aveS×(1-GiNi) (Equation 2a) Ya´=F(x)-F background,a ...(Equation 3a)

number

[0041] The method of the second embodiment using the GiNi coefficient has a feature of expressing the asymmetry index as an area ratio, and it becomes possible to calculate the characteristic curve Ya' from which the linear background component has been removed in a stable and simple manner without being influenced by human intervention.

[0042] (Embodiment 3) The method of obtaining a characteristic curve with the background removed in the third embodiment focuses on the fact that in the case of a characteristic curve in which Y increases rapidly with an increase in x, such as a power or exponential function, the distance between adjacent BPs becomes smaller from the point at which the curve starts to change rapidly, and is a suitable method for characteristic curves in which the curve changes rapidly.

[0043] The detailed procedure is given below with reference to the flow chart diagram of FIG. First, data is input to input equipment 11 (step S250 in FIG. 7). Here, data may be accumulated in a batch format and all data may be input, or the measurement device and threshold calculation analysis device 101 may be connected online and data may be input one by one for each measurement, as in the first embodiment. Next, apply the segment regression piecewise linear function method to the characteristic function Y = F(x) for which background removal is to be performed, and determine the segmentation, the number of segments m, and the segmentation boundary points BP[i] based on the information criterion (step S251). Here, AIC or BIC can preferably be used as the information criterion. After that, determine the slope (a[i]) of the curve of the characteristic function Y = F(x) for each segment i (i is an integer from 1 to m) obtained based on the information criterion (step S252). After that, if there is an a[i] that satisfies a[i] ≥ a[i + 1], set the smallest i that satisfies that condition as n. If the i from 1 to m satisfies a[i] < a[i + 1], set n = m, and determine the calculation target interval as the interval up to segment n (step S253).

[0044] Next, the distance ΔBP[j] between adjacent segmentation boundary points (BP[j]) b = BP[j + 1] - BP[j] (j b is an integer from 1 to n, and j is from 0 to n - 1) is determined (step S254), and the K-means method is applied to ΔBP[j] b to classify ΔBP[j] b into two groups: a group G with a short distance and a group G with a long distance a (step S255). b Here, the K-means method is a method of classifying a set of data into several groups based on the similarity of the data. Since it is simple and has a relatively small computational amount, it is widely used as a practical clustering method.

[0045] In the K-means method, it is important to appropriately set the number of clusters k. If the number of clusters is too large, the similarity of the data decreases, and if the number of clusters is too small, the similarity of the data tends to be too high. In the K-means method, clustering is performed in the following procedure. (1) Set any k points from the given data as the centers (centroids) of the clusters. (2) Assign each data point to the cluster of the closest centroid. ​ (3) Recalculate the center of each cluster as the average value of the data belonging to that cluster. (4) Reassign each data point to the cluster that is closest to each newly calculated cluster center. (5) Repeat steps (3) and (4) until the position of the cluster center does not change. In this application, k=2 and classification is performed into two groups.

[0046] Then, the width ΔBP[j b ] belongs to group Ga, then 0 is assigned, and 1 to group Gb. b ] (step S256), and then virtual ΔBP[n+1] and ΔBP[n+2] are created by assigning the same value as ΔBP[n] to the values ​​of ΔBP[n+1] and ΔBP[n+2] (step S257). Furthermore, ΔBP[j b ] is the value of ΔBP[j b +1] and ΔBP[j b +2] b Calculate the value j b、Gamax (Step S258) is set to the maximum interval section t b A b =j b、Gamax (step S259).

[0047] Finally, the characteristic curve Yb' from which the linear background has been removed is calculated using the following formula 1b (step S260), and the result is output and the process ends (step S270). Y b ´=F(x)-F background,b ...(Equation 1b)

number

[0048] (Embodiment 4) The method for obtaining a characteristic curve with the background removed in the fourth embodiment is to use the maximum interval t c The maximum interval divisions t and t in the first to third embodiments a and t b This is a method that requires a lot of work, but is highly versatile. c By majority vote, t, t a and t b If all three values ​​of are different, then t, t a and t b The second largest value in t c It is defined as follows. The detailed procedure of the method of the fourth embodiment conforms to that of the eighth embodiment described above. The method of the fourth embodiment is a method in which the reliability of the estimation can be evaluated. a and t b If all the values ​​match, the reliability is high; if two match, the reliability is medium; and if they are all different, the reliability is low. This method makes it possible to calculate the characteristic curve Yc' from which the linear background component has been removed in a stable and simple manner without being influenced by human intervention. EXAMPLES

[0049] Example 1 In Example 1, the method of the present invention shown in the first embodiment is applied to measurement data of the critical current density Jc of a superconducting material to calculate a current-voltage characteristic curve from which a linear background component has been removed. However, it should be noted that the present invention is not limited to such a specific form, and the technical scope of the present invention is defined by the claims.

[0050] Measurement data of the critical current density Jc of the superconducting material was obtained as follows. The superconducting material used was MgB2, and measurement data showing the relationship between current J and voltage E was obtained at a temperature of 4.2 K using a measuring device developed in-house.

[0051] The measurement data was input to the input device 11 shown in Fig. 2, and the segment regression piecewise linear function method was applied to determine the divisions and the number of divisions m based on the information criterion. Here, BIC was used as the information criterion. Specifically, the divisions and the number of divisions m were determined in accordance with the method of Non-Patent Document 1. Figure 4 shows the measured data and the divisions. (a), (b), (c), (d), (e), and (f) in the figure show the results when the breakpoints are 1, 2, 3, 4, 5, and 6, respectively. The fitting with the measured data yielded BICs of 1482, 1013, 751, 538, 318, and 223 for the breakpoint numbers of 1, 2, 3, 4, 5, and 6, respectively, and the BIC was smallest when the number of divisions m was 7, indicating the highest compatibility.

[0052] Next, the slope (a[i]) of the measurement data curve for each division i (i is an integer from 1 to m) determined by the information criterion was calculated. The results are shown in Table 1.

[0053] [Table 1]

[0054] In addition, the first indicator is the average slope, aveS=(Y max -Y min ) / (x max -x min ) was calculated. The result was that aveS was 84.24. In addition, as a second indicator, the asymmetry indicator ASY=D cnt / D ave As a result, the ASY was 0.38. The criterion DS is: DS=aveS×ASY=84.24×0.38=31.864 It was. As a result, the linear components derived in the a[1] region with a slope smaller than 31.864 were regarded as linear background components and removed from the measurement data. The results after removal are compared with the measurement data in Figure 5. The corrected characteristic curve with the linear background removed had a clear rising point and was suitable for deriving the critical current density Jc.

[0055] Example 2 In the second embodiment, the method shown in the second embodiment was applied to the measurement data of the critical current density Jc of the superconducting material disclosed in the first embodiment. The GINI value calculated by the Gini coefficient method was 0.409, and therefore 1-GINI was 0.591. As shown in Example 1, aveS was 84.24, and the judgment criterion DSa was calculated as 49.797 from Equation 2a. As a result, the linear component derived in the segmented region up to a[2] where the slope is smaller than 49.797 was regarded as the linear background component, and a characteristic curve from which the background component was removed was obtained by Equation 4a and Equation 5a. The result is shown in FIG. 8.

[0056] Example 3 In the third embodiment, the method shown in the third embodiment was applied to the measurement data of the critical current density Jc of the superconducting material disclosed in the first embodiment. The values ​​of BP are shown in Table 2, and the values ​​of the intervals between BPs, ΔBP, are shown in Table 3.

[0057] [Table 2] [Table 3]

[0058] In accordance with the method described in the third embodiment, the K-means method was applied based on the data on the interval between BPs to divide BPs into groups Ga with narrow intervals between BPs and Gb with wide intervals between BPs, and BD[Jb] of 0 or 1 was assigned. As a result, the area between GP[0] and GP[1] belonged to group Ga, and the others belonged to group Gb. Based on this result, the linear component derived in the division area of ​​a[1] was regarded as a linear background component and removed from the measurement data. Therefore, the characteristic curve Yb' after removing the linear background component derived in the third embodiment became the same as the characteristic function after removing the linear background component derived in the first embodiment.

[0059] Example 4 In the fourth embodiment, the method shown in the fourth embodiment was applied to the measurement data of the critical current density Jc of the superconducting material disclosed in the first embodiment. First, the maximum interval divisions t, t in Examples 1 to 3 a and t b The maximum interval t indicating the linear background interval range is calculated. c The result was decided by majority vote. As a result, t is 1,t a is 2, and t b was 1. As a result, the characteristic curve Yb' after the linear background component was removed, which was derived in Example 4, became the same as the characteristic function after the linear background component was removed, which was derived in Examples 1 and 3. For reference, the background components of the critical current density measurement data were removed up to the BP[1] division boundary point and the BP[2] division boundary point, and a comparison of the two is shown in Figure 9. The results of Examples 1 and 3 are shown in Figure 9(a), the results of Example 2 are shown in Figure 9(b), and a comparison of the two is shown in Figure 9(c). [Industrial Applicability]

[0060] The method of the present invention can be applied to physical phenomena that follow a power law or exponential law, such as the critical current density Jc of superconducting materials. Physical phenomena that follow a power law include the derivation of the photoemission (photoelectric effect) threshold due to ultraviolet irradiation, the derivation of the band gap from the rise of optical absorption, and the derivation of the Schottky threshold of semiconductors in the case of exponential laws, and can be used to remove the background of these phenomena. By applying the method of the present invention, important indicators for analyzing these physical states and using those states to control the state and quality of materials and devices can be obtained stably and easily without being influenced by human intervention. Therefore, we believe that this method will contribute to the development of science and technology as well as to the improvement of product quality and performance, that is, to the development of industry. [Explanation of symbols]

[0061] 11: Input equipment 12: Measurement data (input data) 14: Computing equipment 15: Storage equipment 16: Information path 17: Output equipment 18: Output data (background removal characteristic curve) 101: Processing device

Claims

1. A step of applying a segment regression piecewise linear function method to a characteristic function Y = F (x) to be subjected to background removal, and determining a segment, a number of segments m, and a segment boundary point (BP [j]) (where j is an integer from 0 to m, BP [0] is the minimum value of x, and BP [m] is the maximum value of x) based on an information criterion; determining a slope a[i] and F(0) of the curve of the characteristic function Y=F(x) for each section i (i is an integer from 1 to m) determined by the information criterion; If there is an a[i] that satisfies a[i]≧a[i+1], the minimum value of said i is set to n, and if said i is from 1 to m and satisfies a[i]<a[i+1], n=m is set and the calculation target interval is determined as the interval up to division n; As a first index, the maximum value Y of Y in the section from 1 to n of the section is max , minimum value Y min , the maximum value x of the x max and the minimum value x min Using the above, (Y max -Y min ) / (x max -x min Calculating an average slope (aveS) of the characteristic function Y=F(x) defined by As a second index, the average value (D ave ) and median (D cnt ) to obtain the asymmetry index (ASY=D cnt / D ave ) The judgment criterion (DS) is determined by using the average slope (aveS), the asymmetry index (ASY), and a predetermined reference value L from a value of 0.80 to 0.

95. When 0≦ASY<L, DS=aveS×ASY, When ASY≧L, DS=aveS and The linear component of the slope a[i] (i is an integer from 1 to n) is If a[i]<DS, the background component If a[i]≧DS, the component is classified as a non-background component; subtracting F(0) and said background component from said characteristic function Y=F(x).

2. A step of applying a segment regression piecewise linear function method to a characteristic function Y = F (x) to be subjected to background removal, and determining a segment, a number of segments m, and a segment boundary point (BP [j]) (where j is an integer from 0 to m, BP [0] is the minimum value of x, and BP [m] is the maximum value of x) based on an information criterion; A step of calculating a slope (a[i]) of the curve of the characteristic function Y=F(x) for each section i (i is an integer from 1 to m) calculated by the information criterion as a slope between a start point and an end point of each section; a step of determining the interval to be calculated as up to division n when there is a a[i] that satisfies a[i]≧a[i+1], setting the minimum value of said i to n, and when said i is from 1 to m and satisfies a[i]<a[i+1], setting n=m; As a first index, the maximum value Y of Y in the section from 1 to n of the section is max , minimum value Y min , the maximum value x of the x max and the minimum value x min Using the above, (Y max -Y min ) / (x max -x min Calculating an average slope (aveS) of the characteristic function Y=F(x) defined by As a second index, the average value (D ave ) and median (D cnt Calculating an asymmetry index (ASY) given by the following (Equation 2) from the ratio of determining a judgment criterion (DS) using the average slope (aveS), the asymmetry index (ASY), and a predetermined reference value L from a value of 0.80 to 0.95, in which DS is calculated by the following formula (3) when 0≦ASY<L, and by the following formula (4) when ASY≧L; A step of comparing the a[i] (i is an integer from 1 to n) with the judgment criterion (DS), and determining a maximum interval section t indicating a linear background interval range by setting t to the smallest i that satisfies a[i]≧DS when i is in the range from 1 to n, and setting t to n when a[i]<DS; A method for determining a characteristic curve, comprising the step of determining a characteristic curve Y' from which a linear background has been removed, using the following (Equation 5) to (Equation 7). aveS = (Y max - Y min ) / (x max - x min )... (Equation 1) ASY=D cnt / D ave …………(Formula 2) DS=aveS×ASY (Formula 3) DS=aveS...(Formula 4) Y´=F(x)-F background …………(Formula 5) [0010]

3. 3. The method for determining a characteristic curve according to claim 1, wherein the information criterion is Bayesian Information Criterion (BIC) or Akaike Information Criterion (AIC).

4. 4. The method for determining a characteristic curve according to claim 1, wherein the reference value L is 0.

9.

5. 5. The method for determining a characteristic curve according to claim 1, wherein the characteristic function is a critical current density function of a superconducting material.

6. The segment regression piecewise linear function method is applied to the characteristic function Y = F (x) to be removed from the background, and the division, the number of divisions m, and the division boundary point (BP [j]) (where j is an integer from 0 to m, BP [0] is the minimum value x of x) are determined based on the information criterion. min , BP[m] is the maximum value of x max ) A step of calculating a slope (a[i]) of the curve of the characteristic function Y=F(x) for each section i (i is an integer from 1 to m) calculated by the information criterion as a slope between a start point and an end point of each section; a step of determining the interval to be calculated as up to division n when there is a a[i] that satisfies a[i]≧a[i+1], setting the minimum value of said i to n, and when said i is from 1 to m and satisfies a[i]<a[i+1], setting n=m; As the index 1a, the maximum value Y of Y in the section from 1 to n of the section max , minimum value Y min , the maximum value x of the x max and the minimum value x min calculating an average slope (aveS) of the characteristic function Y=F(x) defined by the following (Equation 1a) using As the index 2a, the x min 0, the x max 1, F(x min ) to 0, F(x max ) is set to 1, (x, Y) is normalized, and (1-GiNi) is calculated from the Gini coefficient (GiNi) obtained from the Lorenz curve; A step of determining a judgment criterion (DSa) using the average slope (aveS) and the index 2a (1-GiNi) according to the following (Formula 2a); The a[i] (i is an integer from 1 to n) is compared with the judgment criterion (DSa), and if there is an i in the range of 1 to n such that a[i] ≥ DSa, the smallest i that satisfies a[i] ≥ DSa is selected as t a , if a[i]<DSa, then t a = n, and the maximum interval section t indicating the linear background interval range a and A method for determining a characteristic curve comprising the step of determining a characteristic curve Ya' from which a linear background has been removed, using the following formula (3a): aveS = (Y max - Y min ) / (x max - x min )... (Equation 1a) DSa=aveS×(1-GiNi) (Formula 2a) Ya´=F(x)-F background,a …………(Formula 3a) [0025]

7. The segment regression piecewise linear function method is applied to the characteristic function Y = F (x) to be removed from the background, and the division, the number of divisions m, and the division boundary point (BP [j]) (where j is an integer from 0 to m, BP [0] is the minimum value x of x) are determined based on the information criterion. min , BP[m] is the maximum value of x max ) A step of calculating a slope (a[i]) of the curve of the characteristic function Y=F(x) for each section i (i is an integer from 1 to m) calculated by the information criterion as a slope between a start point and an end point of each section; a step of determining the interval to be calculated as up to division n when there is a a[i] that satisfies a[i]≧a[i+1], setting the minimum value of said i to n, and when said i is from 1 to m and satisfies a[i]<a[i+1], setting n=m; The distance ΔBP[j] between adjacent partition boundary points (BP[j]) b ]=BP[j+1]-BP[j](j b where j is an integer from 1 to n, and j is an integer from 0 to n-1; Said ΔBP[j b ] to obtain the ΔBP[j b ] to the group G with the shortest distance a And a long group G b A step of classifying the images into two groups; Width ΔBP[j b ] belongs to the group Ga, 0 is set to the division BD[j b ] as the value of A step of creating virtual ΔBP[n+1] and ΔBP[n+2] by assigning the same value as ΔBP[n] as the values ​​of ΔBP[n+1] and ΔBP[n+2]; ΔBP[j b ] is ΔBP[j b +1] and ΔBP[j b +2] is the smallest j b and the value j b、Gamax and Maximum interval segment t indicating linear background interval range b A b = j b、Gamax and calculating A method for determining a characteristic curve, comprising the step of determining a characteristic curve Yb' from which a linear background has been removed, using the following (Equation 1b): Y b ´=F(x)-F background,b …………(Formula 1b) [0030]

8. A segment regression piecewise linear function method is applied to the characteristic function Y = F (x) to be removed from the background, and the division, the number of divisions m, and the division boundary point (BP [j]) (where j is an integer from 0 to m, and the minimum value of x is x min , BP[m] is the maximum value of x max ) A step of calculating a slope (a[i]) of the curve of the characteristic function Y=F(x) for each section i (i is an integer from 1 to m) calculated by the information criterion as a slope between a start point and an end point of each section; a step of determining the interval to be calculated as up to division n when there is a a[i] that satisfies a[i]≧a[i+1], setting the minimum value of said i to n, and when said i is from 1 to m and satisfies a[i]<a[i+1], setting n=m; As a first index, the maximum value Y of Y in the section from 1 to n of the section is max , minimum value Y min , the maximum value x of the x max and the minimum value x min calculating an average slope (aveS) of the characteristic function Y=F(x) defined by the following (Equation 11) using As a second index, the average value (D ave ) and median (D cnt Calculating an asymmetry index (ASY) given by the following (Equation 12) from the ratio of determining a first judgment criterion (DS) by using the average slope (aveS), the asymmetry index (ASY), and a predetermined reference value L from a value of 0.80 to 0.95, in which 0≦ASY<L is satisfied, and by the following (Equation 13), and in which ASY≧L is satisfied, in accordance with the following (Equation 14); A step of comparing the a[i] (i is an integer from 1 to n) with the first judgment criterion (DS), and determining a maximum interval section t indicating a linear background interval range by setting t to the smallest i that satisfies a[i]≧DS when i is in the range from 1 to n, and setting t to n when a[i]<DS; As a third index, the x min 0, the x max 1, F(x min ) to 0, F(x max ) is set to 1, (x, Y) is normalized, and (1-GiNi) is calculated from the Gini coefficient (GiNi) from the Lorenz curve; A step of determining a second judgment criterion (DSa) using the average slope (aveS) and the third index (1-GiNi) according to the following (Equation 15); The a[i] (i is an integer from 1 to n) is compared with the second judgment criterion (DSa), and if there is an i in the range of 1 to n such that a[i] ≥ DSa, the smallest i that satisfies a[i] ≥ DSa is selected as t a , if a[i]<DSa, then t a = n, and the maximum interval section t indicating the linear background interval range a and The distance ΔBP[j] between adjacent partition boundary points (BP[j]) b ]=BP[j+1]-BP[j](j b where j is an integer from 1 to n, and j is an integer from 0 to n-1; Said ΔBP[j b ] to obtain the ΔBP[j b ] to the group G with the shortest distance a And a long group G b A step of classifying the images into two groups; A step of creating virtual ΔBP[n+1] and ΔBP[n+2] by assigning the same value as ΔBP[n] as the values ​​of ΔBP[n+1] and ΔBP[n+2]; ΔBP[j b ] is ΔBP[j b +1] and ΔBP[j b +2] is the smallest j b and the value j b、Gamax and Maximum interval segment t indicating linear background interval range b A b = j b、Gamax and calculating The maximum section division t, t a and t b The most frequent value of t c , or t, t a and t b If all three values ​​of are different, then t, t a and t b The second largest value in c and A method for determining a characteristic curve comprising the step of determining a characteristic curve Yc' from which a linear background has been removed, using the following (Equation 16). aveS = (Y max - Y min ) / (x max - x min )... (Equation 11) ASY=D cnt / D ave …………(Formula 12) DS=aveS×ASY (Formula 13) DS = aveS ... (Formula 14) DSa = aveS × (1 - GiNi) ... (Formula 15) Y c ´=F(x)-F background,c …………(Formula 16) [0045]

Citation Information

Patent Citations

  • Simulation method of characteristics degradation for pmos transistor

    JP2001352059A

  • Viscosity measurement method and device

    JP2022185568A