A device for generating trained models, a device for testing, a method for generating trained models, a program for trained models, a testing method, and a program for testing.

Abstract

To improve the reliability of electronic component testing. [Solution] The trained model generation device 3 generates a trained model 35 which is represented by the sum of a first model f(Rdc4,Rdc2) whose explanatory variables are the first measured value Rdc4 of the DC resistance of the DUT using the four-terminal method and the second measured value Rdc2 of the DC resistance of the DUT using the two-terminal method, and the target variable is the value of the resistance component due to the measurement system using the two-terminal method, and a second model (g(Ls)) whose explanatory variable is the fourth measured value Ls of the inductance of the DUT using the two-terminal method, and the target variable is the value of the resistance component due to the DUT. The trained model generation device 3 generates a second model by machine learning the fourth measured value Ls of the DUT where the difference between the first measured value Rdc4 and the second measured value Rdc2 is less than the threshold Rdth, and generates a first model by machine learning the first measured value Rdc4 and the second measured value Rdc2 of the DUT where the difference is greater than the threshold Rdth.

Description

Technical Field

[0004] , , , , , , , , , , , ,

[0005] , , , , , , ,

[0003]

[0001] The present invention relates to a learned model generation device, an inspection device, a learned model generation method, a program for a learned model, an inspection method, and an inspection program. For example, the present invention relates to a learned model generation device that generates a learned model for inspecting an inductor element.

Background Art

[0002] Conventionally, an inspection device that measures the electrical characteristics of electronic components such as chip inductors and determines the quality of the electronic components based on the measurement results is known. For example, Patent Document 1 discloses an inspection device that measures the AC resistance and inductance of an inductor element to be inspected, calculates a Q value using the measured values, and determines the quality of the inductor element based on the calculated Q value.

[0003] Patent Document 1 describes that, as a method for measuring the AC resistance of an inductor element, the AC resistance is calculated by subtracting an estimated value of the contact resistance of a measurement probe used when measuring by the two-terminal method from the measured value of the AC resistance measured by the two-terminal method. Further, Patent Document 1 describes that an estimated value of the contact resistance is calculated by subtracting the measured value of the DC resistance measured by the four-terminal method from the measured value of the DC resistance measured by the two-terminal method, and the measured value of the AC resistance is corrected using a value obtained by multiplying the estimated value of the contact resistance by a coefficient of 0 or more and 1 or less.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] The inspection device disclosed in Patent Document 1 corrects the measured value of AC resistance based on the assumption that the relationship between the series resistance and AC resistance in an inductor element is linear. However, the actual relationship between the series resistance and AC resistance of an inductor element is unknown, and if, for example, the relationship is nonlinear, the correction of the measured value of AC resistance may not be performed appropriately. Furthermore, in order to avoid overcorrection of AC resistance, the inspection device disclosed in Patent Document 1 corrects the measured value of AC resistance using a value obtained by multiplying the estimated contact resistance by a coefficient, but if that coefficient is not appropriate, the correction of AC resistance may not be performed appropriately.

[0006] This invention has been made in view of the above-mentioned problems, and aims to improve the reliability of inspection of electronic components. [Means for solving the problem]

[0007] A trained model generation device according to a typical embodiment of the present invention includes: a training measurement data acquisition unit that acquires training measurement data by associating a first measured value of the DC resistance of an object to be measured measured by a four-terminal method, a second measured value of the DC resistance of the object to be measured measured by a two-terminal method, a fourth measured value of the inductance of the object to be measured measured by a two-terminal method, with a third measured value of the AC resistance of the object to be measured measured by a two-terminal method; and a trained model that generates a trained model by machine learning the training measurement data to cause a computer to function to calculate the third measured value based on input data including the first measured value, the second measured value, and the fourth measured value. The system comprises a trained model generation unit, wherein the trained model is represented by the sum of a first model, which uses the first measurement value and the second measurement value as explanatory variables and the value of the resistance component due to the two-terminal measurement system as the objective variable, and a second model, which uses the fourth measurement value as an explanatory variable and the value of the resistance component due to the object being measured as the objective variable, and the trained model generation unit generates the second model by machine learning the training measurement data of the object being measured where the difference between the first measurement value and the second measurement value is less than a threshold, and generates the first model by machine learning the training measurement data of the object being measured where the difference is greater than the threshold. [Effects of the Invention]

[0008] According to the inspection apparatus of the present invention, it is possible to improve the reliability of inspection of electronic components. [Brief explanation of the drawing]

[0009] [Figure 1] This figure shows the configuration of a measurement system equipped with a trained model generation device and an inspection device according to Embodiment 1. [Figure 2] This figure shows an example of the configuration of the trained model generation device in the inspection system according to Embodiment 1. [Figure 3] This is a flowchart showing the flow of generating a trained model using the trained model generation device according to Embodiment 1. [Figure 4] This is a flowchart showing the flow of machine learning (step S3) according to Embodiment 1. [Figure 5] This figure shows an example of the configuration of a data processing control device in an inspection device according to Embodiment 1. [Figure 6] This flowchart shows the inspection process using the inspection device according to Embodiment 1. [Figure 7] This figure shows an example of the configuration of the trained model generation device in the inspection system according to Embodiment 2. [Figure 8] This is a flowchart showing the flow of machine learning (step S3) according to Embodiment 2. [Figure 9] This figure shows an example of the configuration of a data processing control device in an inspection device according to Embodiment 2. [Modes for carrying out the invention]

[0010] 1. Overview of the Embodiment First, a general overview of a typical embodiment of the invention disclosed in this application will be provided. In the following description, as an example, reference numerals on the drawings corresponding to the components of the invention are indicated in parentheses.

[0011] [1] A trained model generation device (3) according to a typical embodiment of the present invention includes: a training measurement data acquisition unit (31) that acquires training measurement data (34,34_1~34_n) by associating a first measured value (Rdc4) of the DC resistance of the object to be measured measured by the four-terminal method, a second measured value (Rdc2) of the DC resistance of the object to be measured measured by the two-terminal method, a fourth measured value (Ls) of the inductance of the object to be measured measured by the two-terminal method, with a third measured value (Rs) of the AC resistance of the object to be measured measured by the two-terminal method; and a trained model that causes a computer to function to calculate the third measured value based on input data including the first measured value, the second measured value, and the fourth measured value by machine learning the training measurement data. The system comprises a trained model generation unit (32) that generates (35), wherein the trained model is represented by the sum of a first model (f(Rdc4,Rdc2)) which uses the first measurement value and the second measurement value as explanatory variables and the value of the resistance component due to the two-terminal measurement system as the objective variable, and a second model (g(Ls)) which uses the fourth measurement value as an explanatory variable and the value of the resistance component due to the object being measured as the objective variable, and the trained model generation unit generates the second model by machine learning the training measurement data of the object being measured in which the difference between the first measurement value and the second measurement value is less than a threshold (Rdth), and generates the first model by machine learning the training measurement data of the object being measured in which the difference is greater than the threshold.

[0012] [2] Another typical embodiment of the present invention: A trained model generation device (3A) that acquires training measurement data (34A, 34A_1~34A_n) by associating a first measured value (Rdc4) of the DC resistance of an object to be measured measured by the four-terminal method, a second measured value (Rdc2) of the DC resistance of the object to be measured measured by the two-terminal method, with a third measured value (Rs) of the AC resistance of the object to be measured measured by the two-terminal method; and a trained model generation device (32) that generates a trained model (35A) for causing a computer to function to calculate the third measured value based on input data including the first and second measured values ​​by machine learning the training measurement data. A) comprises the above, wherein the trained model is represented by the sum of a first model (f(Rdc4,Rdc2)) which uses the first measurement value and the second measurement value as explanatory variables and the value of the resistance component due to the two-terminal measurement system as the objective variable, and a second model (g(Rdc4)) which uses the first measurement value as an explanatory variable and the value of the resistance component due to the object being measured as the objective variable, and the trained model generation unit generates the second model by machine learning the training measurement data of the object being measured in which the difference between the first measurement value and the second measurement value is less than a threshold (Rdth), and generates the first model by machine learning the training measurement data of the object being measured in which the difference is greater than the threshold.

[0013] [3] In the trained model generation device (3,3A) described in [1] or [2] above, the number of training measurement data used for machine learning when generating the second model may be less than the number of training measurement data used for machine learning when generating the first model.

[0014] [4] In the trained model generation device described in any one of [1] to [3] above, when the trained model generation unit generates the first model, it may adjust the parameters of the first model while fixing the parameters of the second model in the model (f(Rdc4,Rdc2)+g(Rdc4)) which is represented by the sum of the first model and the second model.

[0015] [5] An inspection apparatus (2) according to a typical embodiment of the present invention includes: a data acquisition unit (21) that acquires a first measured value (Rdc4) of the DC resistance of an object to be measured measured by the four-terminal method, a second measured value (Rdc2) of the DC resistance of the object to be measured measured by the two-terminal method, a third measured value (Rs) of the AC resistance of the object to be measured measured by the two-terminal method, and a fourth measured value (Ls) of the inductance of the object to be measured measured by the two-terminal method; a storage unit (22) that stores a learned model (35) for causing a computer to function to calculate the third measured value based on the input first measured value, second measured value, and fourth measured value; and an estimated value of the third measured value corresponding to the first measured value, second measured value, and fourth measured value acquired by the data acquisition unit, based on the learned model stored in the storage unit. The system comprises an estimation unit (23) that calculates the value of the resistance component due to the two-terminal measurement system, and the trained model is represented by the sum of a first model (f(Rdc4,Rdc2)) which uses the first measurement value and the second measurement value as explanatory variables and the value of the resistance component due to the two-terminal measurement system as the objective variable, and a second model (g(Ls)) which uses the fourth measurement value as an explanatory variable and the value of the resistance component due to the object being measured as the objective variable, wherein the second model is generated by machine learning the training measurement data from which the difference between the first measurement value and the second measurement value is smaller than a threshold, from among a plurality of training measurement data obtained by associating the third measurement value with the first measurement value, the second measurement value and the fourth measurement value that have been measured in advance, and the first model is generated by machine learning the training measurement data from which the difference between the first measurement value and the second measurement value is larger than the threshold, from among the plurality of training measurement data.

[0016] 〔6〕An inspection apparatus (2A) according to another representative embodiment of the present invention includes a data acquisition unit (21) that acquires a first measurement value (Rdc4) of the DC resistance of a measurement object measured by the four-terminal method, a second measurement value (Rdc2) of the DC resistance of the measurement object measured by the two-terminal method, and a third measurement value (Rs) of the AC resistance of the measurement object measured by the two-terminal method, a storage unit (22) that stores a learned model (35A) for causing a computer to function so as to calculate the third measurement value based on the input first measurement value and the second measurement value, and an estimation unit (23) that calculates an estimated value of the third measurement value corresponding to the first measurement value and the second measurement value acquired by the data acquisition unit based on the learned model stored in the storage unit. The learned model is represented by the sum of a first model (f(Rdc4, Rdc2)) having the first measurement value and the second measurement value as explanatory variables and a value of a resistance component caused by a measurement system using the two-terminal method as an objective variable, and a second model (g(Rdc4)) having the first measurement value as an explanatory variable and a value of a resistance component caused by the measurement object as an objective variable. The second model is generated by machine learning of the learning measurement data in which the difference between the first measurement value and the second measurement value is smaller than a threshold value among a plurality of learning measurement data in which the first measurement value, the second measurement value, and the third measurement value are associated in advance. The first model is generated by machine learning of the learning measurement data in which the difference is larger than the threshold value among the plurality of learning measurement data.

[0017] 〔7〕In the inspection apparatus according to the above [5] or [6], the number of the learning measurement data used for machine learning when generating the second model may be less than the number of the learning measurement data used for machine learning when generating the first model.

[0018] 〔8〕In the inspection apparatus according to at least one of the above [5] to [7], the parameters of the first model may be adjusted in a state where the parameters of the second model are fixed in a model represented by the sum of the first model and the second model.

[0019] 〔9〕In the inspection apparatus according to any one of the above〔5〕to〔8〕, according to the estimated value of the third measurement value, the third measurement value is corrected based on the first measurement value and the second measurement value acquired by the data acquisition unit, and a correction unit that performs a correction process of outputting the corrected third measurement value as the value of the AC resistance of the measurement object may be further provided.

[0020] 〔10〕In the inspection apparatus according to the above〔9〕, the correction unit calculates a resistance component due to the measurement system by the two-terminal method according to the first model based on the first measurement value and the second measurement value acquired by the data acquisition unit, and as the correction process, the third measurement value may be corrected based on the resistance component due to the measurement system by the two-terminal method.

[0021] 〔11〕In the inspection apparatus according to the above〔10〕, the correction unit calculates an error between the estimated value of the third measurement value calculated by the estimation unit and the third measurement value acquired by the data acquisition unit, and when the error is smaller than a predetermined threshold value, the correction process may be performed.

[0022]

[12] A method for generating a trained model according to a typical embodiment of the present invention includes: a first step (S1, S2) of acquiring training measurement data by associating a first measurement of the DC resistance of an object to be measured measured by a four-terminal method, a second measurement of the DC resistance of the object to be measured measured by a two-terminal method, a fourth measurement of the inductance of the object to be measured measured by a two-terminal method, with a third measurement of the AC resistance of the object to be measured measured by a two-terminal method; and a second step of generating a trained model by machine learning the training measurement data acquired in the first step, causing the computer to function to calculate the third measurement based on input data including the first measurement, the second measurement, and the fourth measurement. The second step includes steps (S3) and (S34), wherein the learned model is represented by the sum of a first model in which the first measurement value and the second measurement value are explanatory variables and the value of the resistance component due to the two-terminal measurement system is the objective variable, and a second model in which the fourth measurement value is an explanatory variable and the value of the resistance component due to the object being measured is the objective variable, and the second step includes steps (S31, S32) of generating the second model by machine learning the learning measurement data of the object being measured in which the difference between the first measurement value and the second measurement value is less than a threshold, and steps (S33, S34) of generating the first model by machine learning the learning measurement data of the object being measured in which the difference is greater than the threshold.

[0023]

[13] A method for generating a trained model according to another representative embodiment of the present invention includes: a first step (S1, S2) of acquiring training measurement data by associating a first measurement of the DC resistance of an object to be measured measured by a four-terminal method, a second measurement of the DC resistance of the object to be measured measured by a two-terminal method, with a third measurement of the AC resistance of the object to be measured measured by a two-terminal method; and a second step (S3) of generating a trained model by machine learning the training measurement data acquired in the first step, thereby causing the computer to function to calculate the third measurement based on input data including the first and second measurement values. The model is represented by the sum of a first model in which the first measurement value and the second measurement value are explanatory variables and the value of the resistance component due to the two-terminal measurement system is the objective variable, and a second model in which the first measurement value is an explanatory variable and the value of the resistance component due to the object being measured is the objective variable, and the second step is characterized by including the steps of generating the second model by machine learning the learning measurement data of the object being measured in which the difference between the first measurement value and the second measurement value is less than a threshold (S31, S32A), and generating the first model by machine learning the learning measurement data of the object being measured in which the difference is greater than the threshold (S33, S34).

[0024]

[14] A program for generating a trained model according to a typical embodiment of the present invention is characterized in that it causes a computer to execute each step of the trained model generation method described in

[12] or

[13] above.

[0025]

[15] A typical embodiment of the present invention provides an inspection method comprising: a first step (S11-S13) of obtaining a first measured value of the DC resistance of an object to be measured measured by a four-terminal method, a second measured value of the DC resistance of the object to be measured measured by a two-terminal method, a third measured value of the AC resistance of the object to be measured measured by a two-terminal method, and a fourth measured value of the inductance of the object to be measured measured by a two-terminal method; a second step (S14) of calculating an estimated value of the third measured value corresponding to the first measured value, second measured value, and fourth measured value obtained in the first step, based on a trained model for causing a computer to function to estimate the third measured value based on the input first measured value, second measured value, and fourth measured value; and, according to the estimated value of the third measured value, calculating the third measured value based on the first measured value and second measured value obtained in the first step The method includes a third step (S15-S17) of correcting a constant value and outputting the corrected third measurement value as the value of the AC resistance of the object to be measured. The learned model is represented by the sum of a first model, which uses the first measurement value and the second measurement value as explanatory variables and the value of the resistance component due to the two-terminal measurement system as the objective variable, and a second model, which uses the fourth measurement value as an explanatory variable and the value of the resistance component due to the object to be measured as the objective variable. The second model is generated by machine learning the learning measurement data from which the difference between the first measurement value and the second measurement value is smaller than a threshold, from among a plurality of learning measurement data obtained by associating the third measurement value with the first measurement value, the second measurement value and the fourth measurement value that have been measured in advance. The first model is generated by machine learning the learning measurement data from which the difference between the first measurement value and the second measurement value is larger than the threshold, from among the plurality of learning measurement data.

[0026]

[16] A typical inspection method according to another embodiment of the present invention includes: a first step (S11-S13) of obtaining a first measured value of the DC resistance of an object to be measured measured by a four-terminal method, a second measured value of the DC resistance of the object to be measured measured by a two-terminal method, and a third measured value of the AC resistance of the object to be measured measured by a two-terminal method; a second step (S14) of calculating an estimated value of the third measured value corresponding to the first and second measured values ​​obtained in the first step, based on a trained model for causing a computer to function to estimate the third measured value based on the input first and second measured values; and correcting the third measured value based on the first and second measured values ​​obtained in the first step according to the estimated value of the third measured value, and measuring the corrected third measured value The method includes a third step (S15-S17) of outputting the value of the AC resistance of the object, wherein the learned model is represented by the sum of a first model in which the first measurement value and the second measurement value are explanatory variables and the value of the resistance component due to the two-terminal measurement system is the objective variable, and a second model in which the first measurement value is an explanatory variable and the value of the resistance component due to the object being measured is the objective variable, wherein the second model is generated by machine learning the learning measurement data in which the difference between the first measurement value and the second measurement value is smaller than a threshold, from among a plurality of learning measurement data in which the third measurement value is associated with the first measurement value and the second measurement value that have been measured in advance, and the first model is generated by machine learning the learning measurement data in which the difference between the plurality of learning measurement data is larger than the threshold.

[0027]

[17] An inspection program according to a typical embodiment of the present invention is characterized in that it causes a computer to execute each step of the inspection method described in

[15] or

[16] above.

[0028] 2. Specific Examples of Embodiments Hereinafter, specific examples of embodiments of the present invention will be described with reference to the figures. In the following description, common components in each embodiment will be denoted by the same reference numerals, and repeated descriptions will be omitted.

[0029] <Embodiment 1> Figure 1 shows the configuration of an inspection system 1 equipped with a trained model generation device 3 and an inspection device 2 according to Embodiment 1.

[0030] The inspection system 1 shown in Figure 1 is a system for inspecting the quality of an object to be inspected (hereinafter also referred to as "DUT"). As shown in Figure 1, the inspection system 1 comprises a trained model generation device 3 that generates a trained model by learning training measurement data based on multiple measurement results of the DUT using machine learning, and an inspection device 2 that performs inspection of the DUT using the generated trained model.

[0031] Inspection device 2 is a device that measures the electrical characteristics of the DUT and inspects whether the DUT is good or bad based on the measurement results. For example, inspection device 2 is a device (a so-called chip taping machine) that inspects the quality of small electronic components (chip components) and packages the chip components that are determined to be good into a state ready for shipment.

[0032] In the following explanation, we will assume that the DUT is an inductor element (for example, a chip inductor element) as an example, but it is not limited to this.

[0033] The inspection device 2 measures the electrical characteristics of the inductor element as a DUT using a trained model, which will be described later. Specifically, as shown in Figure 1, the inspection device 2 includes a data processing control device 10, a first measurement unit 11, a second measurement unit 12, an operation unit 13, an output unit 14, and a transport mechanism 15.

[0034] The first measurement unit 11 is a device that measures the electrical characteristics of the inductor element as a DUT using the four-terminal method. Examples of the first measurement unit 11 include impedance measuring instruments such as resistance meters and LCR meters that are capable of measuring impedance using the four-terminal method.

[0035] The second measurement unit 12 is a device that measures the electrical characteristics of the inductor element as a DUT using the two-terminal method. An example of the second measurement unit 12 is an impedance measuring instrument such as an LCR meter capable of measuring impedance using the two-terminal method.

[0036] The first measurement unit 11 and the second measurement unit 12 can be any device capable of measuring the electrical characteristics of the DUT, such as impedance, and are not limited to the examples described above.

[0037] The first measurement unit 11 measures the DC resistance of the inductor element acting as a DUT using the four-terminal method in response to instructions from the data processing control device 10. For example, the first measurement unit 11 includes a moving mechanism (not shown) for moving probes 61a to 61d, a current output unit and a voltage detection unit (not shown), and a measurement value calculation unit (not shown) for calculating the measurement value based on the detection result.

[0038] For example, when the first measurement unit 11 receives a measurement execution instruction from the data processing control device 10, the first measurement unit 11's moving mechanism brings probes 61a and 61c into contact with one terminal of the inductor element that has been transported to a predetermined measurement position, and probes 61b and 61d into contact with the other terminal of the inductor element. Next, the current output unit of the first measurement unit 11 supplies a DC current to the inductor element via probes 61a and 61b. The voltage detection unit of the first measurement unit 11 detects the voltage value between the inductor terminals when the DC current is supplied to the inductor element via probes 61c and 61d. The measurement value calculation unit of the first measurement unit 11 calculates the measured value Rdc4 of the DC resistance of the inductor element based on the detected voltage value and the current value of the DC current supplied to the inductor element.

[0039] The second measurement unit 12 measures the DC resistance, AC resistance, and inductance of the inductor element acting as a DUT using the two-terminal method in response to instructions from the data processing control device 10. For example, the second measurement unit 12 includes a moving mechanism (not shown) for moving probes 62a and 62b, a current output unit and a voltage detection unit (not shown), and a measurement value calculation unit (not shown) for calculating the measurement value based on the detection result.

[0040] For example, when the second measurement unit 12 receives a measurement execution instruction from the data processing control device 10, the second measurement unit 12's moving mechanism brings probe 62a into contact with one terminal of the inductor element that has been transported to a predetermined measurement position, and probe 62b into contact with the other terminal of the inductor element. Next, the current output unit of the second measurement unit 12 supplies a DC current to the inductor element via probes 62a and 62b, and the voltage detection unit of the second measurement unit 12 detects the voltage value between both terminals of the inductor element via probes 62a and 62b. The measurement value calculation unit of the second measurement unit 12 calculates the measured value Rdc2 of the DC resistance of the inductor element based on the detected voltage value and the current value of the DC current supplied to the inductor element.

[0041] Furthermore, with the probes 62a and 62b in contact with both terminals of the inductor element by the moving mechanism of the second measuring unit 12, the current output unit of the second measuring unit 12 supplies AC current to the inductor element via the probes, and the voltage detection unit of the second measuring unit 12 detects the AC voltage value between both terminals of the inductor element via the probes 62a and 62b. The measurement value calculation unit of the second measuring unit 12 calculates the measured value Rs of the AC resistance and the measured value Ls of the inductor element based on the detected AC voltage value (effective voltage), the AC current value (effective current) of the AC current supplied to the inductor element, and the phase difference between the AC voltage and the AC current.

[0042] Furthermore, some functions of the first measurement unit 11 and the second measurement unit 12 may be implemented by the data processing control device 10. For example, the calculations performed by the measurement value calculation units of the first measurement unit 11 and the second measurement unit 12 may be performed by the data processing control device 10.

[0043] The operation unit 13 is an input interface for the user to operate the inspection device 2. Examples of the operation unit 13 include various buttons and touch panels. For example, by operating the operation unit 13, the user can set various inspection conditions for inspecting the inductor element as a DUT in the inspection device 2, and can also instruct the inspection device 2 to execute and stop inspections.

[0044] The output unit 14 is a functional unit for outputting various information such as inspection conditions and inspection results from the inspection device 2. The output unit 14 is a display device equipped with, for example, an LCD (Liquid Crystal Display) or an organic EL. For example, when the user instructs the execution of an inspection of the DUT by operating the operation unit 13, the output unit 14 displays information such as inspection results on the screen in accordance with the control of the data processing control device 10.

[0045] The output unit 14 may also be a display device equipped with a touch panel that performs some of the functions of the operation unit 13. Furthermore, the output unit 14 may include a communication circuit or the like that outputs data such as inspection results to an external source via wired or wireless connection.

[0046] The transport mechanism 15 is a device that transports the inductor elements to be inspected to an appropriate location within the inspection device 2 in accordance with the control of the data processing control device 10. For example, when the first measurement unit 11 performs measurement, the transport mechanism 15 transports the inductor elements to be inspected to a predetermined measurement position by the first measurement unit 11. Also, for example, when the second measurement unit 12 performs measurement, the transport mechanism 15 transports the inductor elements to be inspected to a predetermined measurement position by the second measurement unit 12. Furthermore, the transport mechanism 15 transports the inductor elements that have been determined to be good products after inspection to a location for packaging, and then transports the packaged inductor elements to a predetermined location in the next process.

[0047] The data processing control device 10 is a functional unit that comprehensively controls each functional unit within the inspection device 2 and performs various data processing for the inspection of the DUT. For example, the data processing control device 10 is a program processing unit having a processor such as a CPU, a storage device such as ROM, RAM, or flash memory, and peripheral circuits such as a timer. Examples of program processing units include MCUs and FPGAs.

[0048] The data processing control device 10 acquires the measurement results from the first measurement unit 11 and the second measurement unit 12, calculates an index indicating the performance of the inductor element based on the acquired measurement results, and determines whether the inductor element under inspection is good or bad based on the calculated index. Here, the index indicating the performance of the inductor element is, for example, the Q value.

[0049] As described above, when the AC resistance of an inductor element is measured using the two-terminal method, the measured value is affected by the resistance component caused by the measurement system using the two-terminal method. Therefore, when the data processing control device 10 of the inspection device 2 according to Embodiment 1 calculates an index (Q value) indicating the performance of the inductor element to be inspected based on the measurement results from the first measurement unit 11 and the second measurement unit 12, it corrects the measured value Rs of the AC resistance measured by the second measurement unit 12 using a pre-generated learned model as needed.

[0050] Before explaining in detail how the data processing control device 10 corrects the measured value Rs of AC resistance using a trained model, we will first describe how the trained model is generated.

[0051] Figure 2 shows an example of the configuration of the trained model generation device 3 in the inspection system 1 according to the embodiment.

[0052] The trained model generation device 3 is implemented by an information processing device (computer), such as a server or a personal computer (PC), and generates multiple training measurement data 34_1 to 34_n (where n is an integer of 2 or more) according to an installed trained model generation program, and generates a trained model 35 by machine learning the generated training measurement data based on a predetermined algorithm.

[0053] For the sake of explanation, Figure 1 shows the trained model generation device 3 and the inspection device 2 placed side by side. However, the trained model generation device 3 and the inspection device 2 do not necessarily have to be installed in the same location. For example, the inspection device 2 and the trained model generation device 3 may be installed in different locations and connected via a communication network such as a LAN or the Internet. In this case, the inspection device 2 and the trained model generation device 3 may send and receive various data, such as measurement data from the inspection device 2 and the trained model 35, via the communication network.

[0054] Furthermore, during DUT inspection, the inspection device 2 and the trained model generation device 3 do not necessarily need to be electrically connected to each other. For example, before or after inspection, various data such as measurement data and trained models 35 from the inspection device 2 may be exchanged via a storage medium such as a memory card.

[0055] In this embodiment, the trained model 35 is a model for estimating the measured value Rs of the AC resistance of the inductor element under test. The trained model 35 generated by the trained model generation device 3 may be distributed via a network, or it may be written to a computer-readable storage medium such as a memory card and distributed thereafter.

[0056] The trained model generation device 3 includes, for example, a training measurement data acquisition unit 31, a trained model generation unit 32, and a storage unit 33 as functional blocks for generating a trained model 35. Each of these functional blocks is realized through the cooperation of hardware resources such as the CPU and memory that constitute the information processing device as the trained model generation device 3, with software (various programs including a trained model generation program) installed on the information processing device.

[0057] The training measurement data acquisition unit 31 is a functional unit that acquires training measurement data 34_1 to 34_n necessary for generating the trained model 35.

[0058] Here, the learning measurement data 34_1 to 34_n are data pairs that associate the measured value Rdc4 (first measurement) of the DC resistance of the DUT measured by the 4-terminal method, the measured value Rdc2 (second measurement) of the DC resistance of the DUT measured by the 2-terminal method, the measured value Ls (fourth measurement) of the inductance of the DUT measured by the 2-terminal method, with the measured value Rs (third measurement) of the AC resistance of the DUT measured by the 2-terminal method. In the following explanation, when the learning measurement data 34_1 to 34_n are not distinguished, they will simply be referred to as "learning measurement data 34".

[0059] The learning measurement data acquisition unit 31 acquires data pairs including, for example, data 41 of the measured DC resistance Rdc4 of an inductor element measured by the four-terminal method, data 42 of the measured DC resistance Rdc2 of an inductor element measured by the two-terminal method, data 43 of the measured inductance Ls of the DUT measured by the two-terminal method, and data 43 of the measured AC resistance Rs of the DUT measured by the two-terminal method, via wireless or wired communication (not shown) or a storage medium such as a memory card. The learning measurement data acquisition unit 31 acquires data pairs for each inductor element that is tested. For example, the measurement results of inductor elements previously tested by the testing device 2 may be used as data pairs.

[0060] The learning measurement data acquisition unit 31 generates a learning measurement data 34 by, for example, associating the measured value of AC resistance Rs included in the acquired data pair with the measured values ​​of DC resistance Rdc4, Rdc2 and inductance Ls included in the data pair, using the measured value as the correct answer. The learning measurement data acquisition unit 31 generates learning measurement data 34_1 to 34_n for each measurement result of the inspected inductor element and stores them in the storage unit 33.

[0061] Furthermore, as described above, the learning measurement data acquisition unit 31 may acquire learning measurement data 34 generated by another information processing device or the like via communication or a storage medium, instead of generating the learning measurement data 34 itself.

[0062] The trained model generation unit 32 is a functional unit that generates a trained model 35 by machine learning multiple training measurement data 34_1 to 34_n acquired by the training measurement data acquisition unit 31.

[0063] The trained model 35 is a program generated by machine learning based on a predetermined algorithm. Examples of predetermined algorithms include polynomial regression and multiple regression.

[0064] In Embodiment 1, the trained model 35 is a function that causes a computer (MPU, etc.) to operate to estimate the measured AC resistance Rs of the DUT, measured by the two-terminal method, based on input data including the measured DC resistance Rdc4 (first measurement) of the DUT measured by the four-terminal method, the measured DC resistance Rdc2 (second measurement) of the DUT measured by the two-terminal method, and the measured inductance Ls of the DUT measured by the two-terminal method.

[0065] In other words, the trained model 35 is a program that causes the information processing device (computer) to function by performing calculations based on predetermined trained parameters on the input measurement data (measured values ​​of DC resistance Rdc4, Rdc2 and measured value of inductance Ls), and outputting a quantitative value (estimated value) of the AC resistance based on the said measurement data.

[0066] For example, the trained model 35 includes a first model representing the resistance component Rc caused by the two-terminal measurement system and a second model representing the resistance component caused by the object under test (DUT).

[0067] Here, the resistance component resulting from the two-terminal measurement system includes, for example, the resistance component of the transmission line consisting of cables and probes between the second measurement unit 12 and the DUT, and the resistance component resulting from the contact state between the probe and the DUT (so-called contact resistance).

[0068] The first model representing the resistance component Rc due to the two-terminal measurement system is a regression model in which the measured value Rdc4 (first measurement) of DC resistance obtained by the four-terminal method and the measured value Rdc2 (second measurement) of DC resistance obtained by the two-terminal method are explanatory variables, and the value of the resistance component Rc due to the two-terminal measurement system is the dependent variable. In other words, the first model representing the resistance component (Rc) due to the two-terminal measurement system is a function that calculates the resistance component Rc due to the two-terminal measurement system from the measured values ​​Rdc4 and Rdc2 of DC resistance.

[0069] The second model representing the resistance component due to the DUT is, for example, a regression model in which the measured inductance Ls (fourth measurement) obtained by the two-terminal method is the explanatory variable and the value of the resistance component (AC resistance) due to the DUT is the dependent variable. In other words, the second model representing the resistance component due to the DUT is a function that estimates the value of the AC resistance due to the DUT from the measured inductance obtained by the two-terminal method.

[0070] When the first model representing the resistance component (Rc) caused by the two-terminal measurement system is defined as "Rc=f(Rdc4,Rdc2)" and the second model representing the resistance component (AC resistance Rs0) caused by the DUT is defined as "Rso=g(Ls)", the trained model 35, which is a function for calculating the estimated value Rse of the measured AC resistance, can be expressed, for example, as Rse=Rc+Rs0=f(Rdc4,Rdc2)+g(Ls).

[0071] In other words, the estimated value Rse of the measured AC resistance is expressed as the sum of the resistance component Rc, which is due to the two-terminal measurement system determined by the first model (Rc=f(Rdc4,Rdc2)), and the resistance component (AC resistance) due to the DUT, which is determined by the second model (Rs0=g(Ls)). In the following explanation, the first model (Rc=f(Rdc4,Rdc2)) will be referred to as "model f(Rdc4,Rdc2)", and the second model (Rs0=g(Ls)) will also be referred to as "model g(Ls)".

[0072] Model f(Rdc4,Rdc2) and model g(Ls) include parameters. Parameters are, for example, the coefficients (weighting coefficients) of model f(Rdc4,Rdc2) and model g(Ls). In the following description, parameters that have been mechanically adjusted to calculate the estimated value Rse of the measured AC resistance using the training measurement data 34_1 to 34_n as input to the training program (a program based on the predetermined algorithm described above) are also referred to as "trained parameters".

[0073] The trained model generation unit 32 adjusts the parameters of model f(Rdc4,Rdc2) and model g(Ls) by machine learning the training measurement data 34_1 to 34_n, thereby obtaining trained parameters.

[0074] Specifically, the trained model generation unit 32 generates a trained model 35 (Rse = f(Rdc4, Rdc2) + g(Ls)) by separately performing machine learning on the model f(Rdc4, Rdc2) and the model g(Ls). That is, the trained model generation unit 32 separately executes a process of adjusting the parameters of the model f(Rdc4, Rdc2) by machine learning and a process of adjusting the parameters of the model g(Ls) by machine learning.

[0075] More specifically, the trained model generation unit 32 generates the model g(Ls) by performing machine learning on training measurement data 34 with a small variation in the measured value of the DC resistance, that is, a small variation in the contact resistance during measurement by the two-terminal method, and generates the model f(Rdc4, Rdc2) by performing machine learning on training measurement data 34 with a large variation in the measured value of the DC resistance.

[0076] For example, the trained model generation unit 32 generates a model (Rs = g(Ls)) by performing machine learning on training measurement data 34 where the difference between the measured value Rdc4 of the DC resistance and the measured value Rdc2 of the DC resistance is smaller than the threshold value Rdth (|Rdc2 - Rdc4| < Rdth).

[0077] In addition, the trained model generation unit 32 generates a model (Rc = f(Rdc4, Rdc2)) by performing machine learning on training measurement data 34 where the difference between the measured value Rdc4 of the DC resistance and the measured value Rdc2 of the DC resistance is larger than the threshold value (|Rdc2 - Rdc4| > Rdth). Here, the threshold value Rdth is preferably a value as close to "0" as possible.

[0078] The number of training measurement data 34 used for machine learning when generating the model g(Ls) may be less than the number of training measurement data 34 used for machine learning when generating the model f(Rdc4, Rdc2).

[0079] For example, approximately 10 samples each with different inductances (e.g., L=1nH, 5nH, 10nH, etc.) can be measured, and training measurement data 34 where the difference between the measured DC resistance Rdc4 and the measured DC resistance Rdc2 is smaller than the threshold Rdth can be used for machine learning of model g(Ls). On the other hand, several hundred to several thousand samples with different inductances can be measured, and training measurement data 34 where the difference between the measured DC resistance Rdc4 and the measured DC resistance Rdc2 is larger than the threshold Rdth can be used for machine learning of model f(Rdc4,Rdc2).

[0080] Furthermore, the trained model generation unit 32 adjusts the parameters of model g(Ls) and then adjusts the parameters of f(Rdc4,Rdc2). In other words, the trained model generation unit 32 first adjusts the parameters of the model g(Ls) using the method described above. For example, the trained model generation unit 32 calculates the difference (error) between the value of the resistance component due to the DUT (AC resistance Rs0), which is calculated by inputting the measured value Rs of the AC resistance included in the training measurement data 34 with small variation in DC resistance from the training measurement data 34_1 to 34_n into the regression model (g(Ls)), and the value of the resistance component due to the DUT, which is the correct value included in the training measurement data 34 with small variation in DC resistance (measured value Rs of AC resistance). Next, the trained model generation unit 32 generates trained parameters of the model g(Ls) by sequentially updating the parameters of the regression model (Rso=g(Ls)) so that the calculated error becomes smaller, for example by backpropagation, and stores them in the storage unit 33.

[0081] Next, the trained model generation unit 32 adjusts the parameters of f(Rdc4,Rdc2) by machine learning the training measurement data 34 in a model represented as the sum of f(Rdc4,Rdc2) and model g(Ls). Specifically, in the model (Rse=f(Rdc4,Rdc2)+g(Ls)), the trained model generation unit 32 adjusts the parameters of f(Rdc4,Rdc2) by machine learning the training measurement data 34 (|Rdc2―Rdc4|>Rdth) in which the difference between the measured DC resistance values ​​Rdc4 and Rdc2 is greater than a threshold, while fixing the trained parameters of model g(Ls).

[0082] For example, the trained model generation unit 32 calculates the difference (error) between the estimated value Rse of the AC resistance measurement, which is calculated by inputting the measured values ​​Rdc4, Rdc2, and Ls of the DC resistance measurement included in the training measurement data 34 with a large variation in DC resistance from the training measurement data 34_1 to 34_n into the regression model (Rse=f(Rdc4,Rdc2)+g(Ls)), and the correct value Rs of the AC resistance measurement included in the training measurement data 34 with a large variation in DC resistance. Next, the trained model generation unit 32 sequentially updates the parameters of the model f(Rdc4,Rdc2) in the regression model (Rse=f(Rdc4,Rdc2)+g(Ls)) to reduce the calculated error, for example by backpropagation, and generates trained parameters for the model f(Rdc4,Rdc), which are stored in the storage unit 33.

[0083] Here, the trained parameters of model g(Ls) used during machine learning of model f(Rdc4,Rdc) are the parameters adjusted by the method described above, and are not updated during machine learning of model f(Rdc4,Rdc).

[0084] The trained model generation unit 32 stores the regression model (Rse=f(Rdc4,Rdc2)+g(Ls)) containing the trained parameters adjusted by the method described above as the trained model 35 in the storage unit 33.

[0085] The memory unit 33 is a functional unit for storing various data, such as the training measurement data 34_1 to 34_n necessary for generating the trained model 35, and the generated trained model 35.

[0086] The memory unit 33 is configured to be accessible from the outside, for example. For example, by the inspection device 2 communicating with the trained model generation device 3, the inspection device 2 can read and acquire the trained model 35 from the memory unit 33. Also, for example, by the inspection device 2 communicating with the trained model generation device 3, the inspection device 2 can write measurement result data, etc., to the memory unit 33.

[0087] Figure 3 is a flowchart showing the flow of generating a trained model 35 by the trained model generation device 3 according to the embodiment.

[0088] As shown in Figure 3, first, in the trained model generation device 3, the training measurement data acquisition unit 31 acquires training measurement data 34_1 to 34_n (step S1). Specifically, as described above, the training measurement data acquisition unit 31 acquires data pairs including the measured values ​​of DC resistance Rdc4 and Rdc2 and the measured value of AC resistance Rs of inductor elements that have been inspected in the past. Based on the acquired data pairs, the training measurement data 34 is generated by assigning the measured value of AC resistance Rs as the correct value to the measured values ​​of DC resistance Rdc4 and Rdc2.

[0089] Next, the trained model generator 3 determines whether the required number of training measurement data 34 for generating the trained model 35 has been generated (step S2). For example, the trained model generator 3 has a preset number of training measurement data 34 required for generating the trained model 35, and each time the trained model generator 3 generates training measurement data 34, it increments the generation count by +1. The trained model generator 3 then determines whether the required number of training measurement data 34 has been generated by checking whether the counted generation count has reached the preset number of data.

[0090] If the necessary number of learning measurement data 34 has not been generated (step S2: NO), the learned model generation device 3 returns to step S1, obtains a data pair related to the measurement result of a new inductor element, and repeats generating learning measurement data 34 related to the inductor element (steps S1, S2).

[0091] On the other hand, if the necessary number of learning measurement data 34 has been generated (step S2: YES), the learned model generation device 3 performs machine learning using the plurality of learning measurement data 34 generated in step S1 (step S3).

[0092] FIG. 4 is a flowchart showing the flow of machine learning (step S3) according to Embodiment 1.

[0093] First, the learned model generation unit 32 extracts learning measurement data 34 with a small variation in DC resistance from the learning measurement data 34_1 to 34_n (step S31). Specifically, as described above, learning measurement data 34 in which the difference between the measured value Rdc4 of the DC resistance and the measured value Rdc2 of the DC resistance is smaller than the threshold value (|Rdc2 - Rdc4| < Rdth) is extracted.

[0094] Next, the learned model generation unit 32 generates a model g(Ls) using the learning measurement data 34 extracted in step S31 by the method described above (step S32).

[0095] Next, the learned model generation unit 32 extracts learning measurement data 34 with a large variation in DC resistance from the learning measurement data 34_1 to 34_n (step S33). Specifically, as described above, learning measurement data 34 in which the difference between the measured value Rdc4 of the DC resistance and the measured value Rdc2 of the DC resistance is larger than the threshold value (|Rdc2 - Rdc4| > Rdth) is extracted.

[0096] Next, the trained model generation unit 32 generates a model f(Rdc2, Rdc4) using the training measurement data 34 extracted in step S33 by the method described above (step S34).

[0097] Through the above processing flow, the parameters of model g(Ls) and model f(Rdc2,Rdc4) are individually adjusted based on different training measurement data 34, and a trained model 35 (Rse=f(Rdc2,Rdc4)+g(Ls)) is generated.

[0098] Next, as shown in Figure 3, the trained model 35 generated in step S3 is registered with the inspection device 2 (step S4). For example, in response to a user's operation of the operation unit 13 of the inspection device 2 or the input device (e.g., touch panel, keyboard, mouse, etc.) of the trained model generation device 3, the trained model generation device 3 transmits the trained model 35 stored in the storage unit 33 to the inspection device 2, and the inspection device 2 stores the received trained model 35 in the storage unit of the data processing control device 10. Note that the registration of the trained model 35 to the inspection device 2 may also be performed using a storage medium such as a memory card, as described above.

[0099] Furthermore, the program for generating a trained model, which causes the computer (information processing device) acting as the trained model generation device 3 to execute each of the above steps (S1 to S4), may be distributed via a network, or it may be written to a computer-readable storage medium such as a memory card and distributed therein.

[0100] Using the method described above, a trained model 35 for testing inductor elements is generated.

[0101] Next, we will explain in detail how to correct the measured value Rs of AC resistance using the trained model 35 from the inspection device 2.

[0102] Figure 5 shows an example of the configuration of the data processing control device 10 in the inspection device 2 according to Embodiment 1.

[0103] As shown in Figure 5, the data processing control device 10 of the inspection device 2 includes, for example, a data acquisition unit 21, a storage unit 22, an estimation unit 23, a correction unit 24, and a determination unit 25. These functional units are realized, for example, in a program processing device as the data processing control device 10, by the CPU executing various calculations according to a program stored in memory and controlling peripheral circuits such as counters.

[0104] The data acquisition unit 21 is a functional unit that acquires various data necessary to calculate the performance index (Q value) of the inductor element under inspection.

[0105] The data acquisition unit 21 acquires, for example, the measured value Rdc4 of the DC resistance of the DUT measured by the first measurement unit 11 using the four-terminal method and stores it in the storage unit 22. The data acquisition unit 21 also acquires, for example, the measured value Rdc2 of the DC resistance of the DUT measured by the second measurement unit 12 using the two-terminal method, the measured value Rs of the AC resistance of the DUT measured by the second measurement unit 12 using the two-terminal method, and the measured value Ls of the inductance of the DUT measured by the second measurement unit 12 using the two-terminal method, and stores them in the storage unit 22 as measurement data 50 of the object to be inspected. In addition, the data acquisition unit 21 acquires, for example, the trained model 35 generated by the trained model generation device 3 and stores it in the storage unit 22.

[0106] The memory unit 22 is a functional unit for storing various data necessary for calculating the performance index (Q value) of the inductor element under inspection, as well as the calculated Q value.

[0107] As described above, the memory unit 22 stores the measured values ​​of the DC resistance of the inductor element Rdc4 and Rdc2, the measured value of the AC resistance of the inductor element Rs, the measured value of the inductance of the inductor element Ls, and the learned model 35, all acquired by the data acquisition unit 21. The memory unit 22 also stores, for example, the estimated value of the measured value of the AC resistance Rse, the estimated value of the resistance component Rc due to the two-terminal measurement system, the value of the AC resistance Rsr, and the Q value, which will be described later.

[0108] The estimation unit 23 is a functional unit that estimates the measured value of the AC resistance of the inductor element under test. Based on the learned model 35 stored in the storage unit 22, the estimation unit 23 calculates the estimated value Rse of the measured AC resistance corresponding to the measured values ​​Rdc4 and Rdc2 of the DC resistance of the inductor element under test acquired by the data acquisition unit 21. Specifically, the estimation unit 23 inputs (substitutes) the measured values ​​Rdc4 and Rdc2 of the DC resistance and the measured value Ls of the inductor element under test acquired by the data acquisition unit 21 into the learned model 35 (function) and stores the value obtained as the estimated value Rse of the measured AC resistance in the storage unit 22.

[0109] The correction unit 24 is a functional unit for correcting the measured value Rs of the AC resistance. The correction unit 24 corrects the measured value Rs of AC resistance based on the measured values ​​Rdc4 and Rdc2 of DC resistance acquired by the data acquisition unit 21, according to the estimated value Rse of the measured value of AC resistance, and performs a correction process to output the corrected measured value Rs of AC resistance as the AC resistance value Rsr of the DUT.

[0110] More specifically, the correction unit 24 first calculates the resistance component Rc caused by the two-terminal measurement system according to the model f(Rdc4,Rdc2) described above, based on the measured values ​​Rdc4 and Rdc2 of the DC resistance of the inductor element under inspection acquired by the data acquisition unit 21. The calculated resistance component Rc data is stored, for example, in the storage unit 22.

[0111] Next, the correction unit 24 corrects the measured value Rs of the AC resistance of the inductor element under inspection based on the resistance component Rc caused by the two-terminal measurement system, according to the estimated value Rse of the measured AC resistance, and outputs the corrected measured value Rs of AC resistance as the value Rsr of the AC resistance of the inductor element under inspection.

[0112] For example, the correction unit 24 calculates the error |Rse-Rs| between the estimated value Rse of the AC resistance measured by the estimation unit 23 and the measured value Rs (third measured value) of the AC resistance acquired by the data acquisition unit 21, and evaluates the error |Rse-Rs|. Specifically, the correction unit 24 compares the error |Rse-Rs| with a threshold value Rth. The threshold value Rth is an arbitrary value set in advance.

[0113] Here, if the error |Rse-Rs| is smaller than the threshold Rth, it can be considered that the accuracy of the AC resistance estimation by the trained model 35 is high for the measurement results of the inductor element under test. In other words, it can be considered that the accuracy of the estimation of the resistance component Rc caused by the two-terminal measurement system is high by the model f(Rdc4,Rdc2) included in the trained model 35.

[0114] Therefore, if the error |Rse-Rs| is smaller than the threshold Rth, the correction unit 24 performs a correction process. Specifically, the correction unit 24 calculates the resistance component Rc caused by the two-terminal measurement system using model f(Rdc4,Rdc2), and corrects the measured value Rs (third measured value) of the AC resistance using the calculated resistance component Rc caused by the two-terminal measurement system.

[0115] For example, the correction unit 24 first inputs (substitutes) the measured values ​​Rdc4 and Rdc2 of the DC resistance of the inductor element under test into the model f(Rdc4, Rdc2) and uses the resulting value as the resistance component Rc caused by the two-terminal measurement system. Next, the correction unit 24 subtracts the resistance component Rc caused by the two-terminal measurement system from the measured value Rs (third measured value) of AC resistance acquired by the data acquisition unit 21 and outputs the resulting value as the AC resistance value Rsr (= Rs - Rc).

[0116] On the other hand, if the error |Rse-Rs| is greater than the threshold Rth, it is considered that the accuracy of the AC resistance estimation by the trained model 35 for the inductor element under test is low. In other words, it is considered that the accuracy of the resistance component Rc caused by the two-terminal measurement system is low, as indicated by model f(Rdc4,Rdc2) included in the trained model 35. In this case, if we were to use model f(Rdc4,Rdc2) to calculate the resistance component Rc caused by the two-terminal measurement system, and then use the calculated resistance component Rc to correct the measured value Rs (third measured value) of the AC resistance, it would result in an incorrect correction, and there is a risk that we would not be able to properly determine the value of the AC resistance Rsr.

[0117] Therefore, if the error |Rse-Rs| is greater than the threshold Rth, the correction unit 24 outputs the measured value Rs of the AC resistance acquired by the data acquisition unit 21 as the AC resistance value Rsr (=Rs) of the inductor element under inspection, without performing any correction processing.

[0118] The determination unit 25 is a functional unit for determining whether the DUT (inductor element) is good or bad. The determination unit 25 calculates the Q value (Q = ωL / Rsr), which is an index representing the performance of the inductor element under test, based on the AC resistance value Rsr of the inductor element under test output from the correction unit 24 and the measured value Ls of the inductance of the inductor element under test.

[0119] The determination unit 25 determines whether the inductor element under inspection is good or bad by, for example, comparing the calculated Q value with a predetermined reference value. The determination unit 25 controls the transport mechanism 15 to package the DUTs that it has determined to be good into a ready-to-ship state using a packaging device not shown in the diagram.

[0120] Next, we will explain the inspection process of the DUT using inspection device 2.

[0121] Figure 6 is a flowchart showing the inspection process by the inspection device 2 according to Embodiment 1.

[0122] For example, if a user operates the control unit 13 of the inspection device 2 to instruct the execution of a DUT (inductor element) inspection, the data processing control device 10 starts the inspection of the inductor element to be inspected.

[0123] First, the data processing control device 10 controls the first measurement unit 11 to measure the DC resistance of the inductor element to be inspected using the four-terminal method (step S11). For example, the data processing control device 10 controls the transport mechanism 15 in response to an instruction signal from the operation unit 13 to transport the inductor element to be inspected to a predetermined measurement position in the first measurement unit 11. Subsequently, the data processing control device 10 controls the first measurement unit 11 to measure the DC resistance of the inductor element to be inspected using the four-terminal method and obtains the measured value of the DC resistance, Rdc4.

[0124] Next, the data processing control device 10 controls the second measurement unit 12 to measure the DC resistance of the inductor element to be inspected using the two-terminal method (step S12). For example, the data processing control device 10 controls the transport mechanism 15 to transport the inductor element to be inspected to a predetermined measurement position in the second measurement unit 12. After that, the data processing control device 10 controls the second measurement unit 12 to measure the DC resistance of the inductor element to be inspected using the two-terminal method and obtains the measured value of the DC resistance Rdc2.

[0125] Next, the data processing control device 10 controls the second measurement unit 12 to measure the AC resistance Rs and inductance Ls of the inductor element to be inspected using the two-terminal method (step S13). For example, with the inductor element to be inspected placed in the same measurement position as in step S12, the data processing control device 10 controls the second measurement unit 12 to measure the AC resistance of the inductor element to be inspected and obtains the measured value Rs of AC resistance and the measured value Ls of inductance, respectively.

[0126] In steps S11 to S13, the DC resistance measurements Rdc4 and Rdc2, the AC resistance measurement Rs, and the inductance measurement Ls acquired by the data processing control device 10 are stored in the storage unit 22 as measurement data 50 of the inductor element to be inspected.

[0127] Next, the data processing control device 10 calculates an estimated value Rse of the measured AC resistance of the inductor element to be inspected based on the measurement data 50 of the inductor element to be inspected (step S14). Specifically, the estimation unit 23 inputs the measured values ​​Rdc4 and Rdc2 of the DC resistance acquired in steps S11 and S12 into the trained model 35 using the method described above, thereby obtaining an estimated value Rse of the measured AC resistance output from the trained model 35.

[0128] Alternatively, after step S13, the DC resistance of the inductor element under test may be measured using the two-terminal method, and the measured value of DC resistance Rdc2 may be obtained again. In this case, the estimation unit 23 may compare the measured value of DC resistance Rdc2 measured in step S12 with the measured value of DC resistance Rdc2 measured again after step S13, and use the smaller of the two measured values ​​of DC resistance Rdc2 to calculate the estimated value of AC resistance Rse in step S14.

[0129] Next, the correction unit 24 determines whether the difference |Rse-Rs| between the estimated value Rse of the measured AC resistance calculated in step S14 and the measured value Rs of the AC resistance obtained in step S13 is smaller than the threshold Rth (step S15).

[0130] If the difference |Rse-Rs| is smaller than the threshold Rth (step S15: YES), the correction unit 24 calculates the resistance component Rc caused by the two-terminal measurement system using model f(Rdc4, Rdc2) (step S16). Specifically, the correction unit 24 obtains the resistance component Rc caused by the two-terminal measurement system by inputting the measured values ​​of DC resistance Rdc4 and Rdc2 obtained in steps S11 and S12 into model f(Rdc4, Rdc2).

[0131] Next, the correction unit 24 corrects the measured value Rs of AC resistance obtained in step S13 using the resistance component Rc caused by the two-terminal measurement system calculated in step S16, and outputs the corrected value as the AC resistance value Rsr (step S17). Specifically, the correction unit 24 subtracts the resistance component Rc caused by the two-terminal measurement system calculated in step S16 from the measured value Rs of AC resistance obtained in step S13 and outputs the value obtained as the AC resistance value Rsr (= Rs - Rc).

[0132] On the other hand, if the difference |Rse-Rs| is greater than the threshold Rth (step S15: NO), the correction unit 24 outputs the measured value Rs of the AC resistance obtained in step S13 as the AC resistance value Rsr (=Rs) without performing any correction processing (step S18).

[0133] Next, the determination unit 25 calculates the Q value of the inductor element to be inspected based on the AC resistance value Rsr output from the correction unit 24 in step S17 or step S18 and the inductance measurement value Ls obtained in step S13 (step S19). After that, the determination unit 25 determines whether the inductor element to be inspected is good or bad based on the Q value calculated in step S19 (step S20). Inductor elements determined to be good are transported by the transport mechanism 15 and packaged.

[0134] Furthermore, the test program used to cause the computer (information processing device) acting as the data processing control device 10 to execute each of the above steps (S11 to S20) may be distributed via a network, or it may be written to a computer-readable storage medium such as a memory card and distributed therein.

[0135] In the inspection system 1 according to Embodiment 1, the trained model generation device 3 generates a trained model 35, which includes a model f(Rdc4, Rdc2) for calculating the resistance component due to the two-terminal measurement system and a model g(Ls) for calculating the resistance component due to the DUT (inductor element), by machine learning multiple training measurement data 34_1 to 34_n, which are generated by labeling the measured value Rdc4 of the DC resistance of the DUT measured by the four-terminal method, the measured value Rdc2 of the DC resistance of the DUT measured by the two-terminal method, the measured value Ls of the inductance of the DUT measured by the two-terminal method, and the measured value Rs of the AC resistance of the DUT measured by the two-terminal method.

[0136] According to this, even if the relationship between the measured DC resistance values ​​Rdc4 and Rdc2 of the inductor element and the measured AC resistance value Rs is nonlinear, a trained model 35 (function) can be obtained that appropriately represents the relationship between the measured DC resistance values ​​Rdc4 and Rdc2, the measured inductance value Ls, and the measured AC resistance value Rs.

[0137] Furthermore, in the trained model 35, model f(Rdc4,Rdc2) is a regression model in which the measured values ​​of DC resistance Rdc4 and Rdc2 are explanatory variables and the value of the resistance component due to the two-terminal measurement system is the dependent variable, and model g(Ls) is a regression model in which the measured value of inductance Ls is the explanatory variable and the value of the resistance component due to the DUT is the dependent variable.

[0138] According to this, since the learned model 35 can be represented by a simpler function, it is possible to avoid the black-boxing of the learned model 35, which is a concern in machine learning. Also, in the regression model (g(Ls)) representing the value of the resistance component caused by the DUT, by using the measured value Ls of the inductance as an explanatory variable, it becomes possible to estimate the resistance component caused by the DUT with higher accuracy.

[0139] Also, the learned model generation device 3 generates the model g(Ls) by performing machine learning on the learning measurement data 34 where the difference between the measured value Rdc4 of the DC resistance and the measured value Rdc2 of the DC resistance is smaller than the threshold (|Rdc2 - Rdc4| < Rdth), and generates the model f(Rdc2, Rdc4) by performing machine learning on the learning measurement data 34 where the difference between the measured value Rdc4 of the DC resistance and the measured value Rdc2 of the DC resistance is larger than the threshold (|Rdc2 - Rdc4| > Rdth).

[0140] The model f(Rdc2, Rdc4) representing the resistance component Rc caused by the measurement system using the two-terminal method can generate a more accurate model by using the measurement results of samples with large variations in the series resistance (variations in the contact resistance) as the learning measurement data 34 for machine learning. On the other hand, for the model g(Ls) representing the resistance component caused by the DUT, it is not necessary to use the measurement results of samples with large variations in the series resistance (variations in the contact resistance). Rather, if the measurement results of samples with large variations in the series resistance are used for the machine learning of the model g(Ls), the accuracy of the model g(Ls) may decrease.

[0141] Therefore, like the learned model generation device 3, by performing machine learning using the measurement results of samples with small variations in the series resistance (variations in the contact resistance) as the learning measurement data 34, it becomes possible to generate a more accurate model g(Ls). Therefore, according to the learned model generation device 3 according to Embodiment 1, it becomes possible to obtain a more accurate learned model 35 (Rse = f(Rdc2, Rdc4) + g(Ls)).

[0142] Furthermore, the number of training measurement data 34 used for machine learning when generating model g(Ls) is less than the number of training measurement data used for machine learning when generating model f(Rdc2,Rdc4). As mentioned above, the training measurement data 34 used for machine learning when generating model g(Ls) are measurement results of samples with small variations in series resistance, so it is possible to generate a highly accurate model g(Ls) even with a small number of measurement results used for machine learning. Therefore, the time required for machine learning of the trained model 35 (Rse=f(Rdc2,Rdc4)+g(Ls)) can be shortened.

[0143] Furthermore, the trained model generator 3 adjusts the parameters of model f(Rdc2,Rdc4) by performing machine learning with the trained parameters of model g(Ls) fixed, in a regression model represented as the sum of model f(Rdc2,Rdc4) and model g(Ls). According to this, it is possible to prevent the model g(Ls), whose parameters have already been trained and adjusted, from being retrained using other training measurement data 34. Therefore, it is possible to obtain a highly accurate model f(Rdc2,Rdc4) while maintaining the accuracy of model g(Ls).

[0144] Furthermore, in the inspection system 1 according to Embodiment 1, the inspection device 2 estimates the measured value of the AC resistance of the DUT using the trained model 35 generated by the trained model generation device 3, and calculates the resistance component Rc caused by the two-terminal measurement system using the model f(Rdc4, Rdc2) included in the trained model 35 according to the estimation result. Then, the inspection device 2 corrects the measured value of the AC resistance Rs based on the calculated resistance component Rc caused by the two-terminal measurement system, and outputs the corrected value as the AC resistance value of the DUT.

[0145] According to this method, the measured value of AC resistance Rs is corrected using the resistance component Rc, which is calculated based on a high-precision model f(Rdc4,Rdc2) generated by machine learning from past inductor element measurement results, and which is derived from the two-terminal measurement system. This allows for more accurate acquisition of AC resistance values ​​than conventional methods.

[0146] Furthermore, the inspection device 2 calculates the error |Rse-Rs| between the estimated value Rse of the measured AC resistance estimated based on the trained model 35 and the actually measured value Rs of the AC resistance. If the error |Rse-Rs| is smaller than the threshold Rth, the inspection device 2 outputs the measured value Rs of the AC resistance corrected based on the resistance component Rc caused by the two-terminal measurement system as the AC resistance value Rsr (=Rse-Rc). On the other hand, if the error |Rse-Rs| is greater than the threshold Rth, the inspection device 2 outputs the actually measured value Rs of the AC resistance as the AC resistance value Rsr (=Rs).

[0147] According to this, instead of uniformly correcting the measured AC resistance for all inductor elements under test, the correction of the measured AC resistance is applied only to inductor elements where the error between the measured AC resistance and the estimated value based on the learned model 35 is small, i.e., inductor elements for which it is considered appropriate to apply the learned model 35. This prevents overcorrection and allows for obtaining a more accurate AC resistance value.

[0148] Thus, the trained model generation device 3 and inspection device 2 according to Embodiment 1 can improve the reliability of inspection of electronic components.

[0149] <Embodiment 2> Figure 7 shows an example of the configuration of the trained model generation device 3A in the inspection system 1A according to Embodiment 2.

[0150] The inspection system 1A according to Embodiment 2 differs from the inspection system 1 according to Embodiment 1 in that, in the regression model representing the value Rs0 of the resistance component caused by the DUT as a second model, the DC resistance Rdc4 measured by the four-terminal method is used as the explanatory variable instead of the measured value Ls of inductance, but in other respects it is the same as the inspection system 1 according to Embodiment 1.

[0151] In the trained model generation device 3A, the training measurement data acquisition unit 31A, similar to the training measurement data acquisition unit 31 according to Embodiment 1, acquires a data pair for each inspected inductor element, including data 41 of the measured DC resistance Rdc4 of the inductor element measured by the four-terminal method, data 42 of the measured DC resistance Rdc2 of the inductor element measured by the two-terminal method, and data 43 of the measured AC resistance Rs of the inductor element measured by the two-terminal method, via wireless or wired communication (not shown) or a storage medium such as a memory card.

[0152] The learning measurement data acquisition unit 31A generates a learning measurement data 34 by associating the measured value of AC resistance Rs included in the acquired data pair with the measured values ​​of DC resistance Rdc4 and Rdc2 included in the same data pair, using Rs as the correct value. The learning measurement data acquisition unit 31 generates learning measurement data 34A_1 to 34A_n for each measurement result of the inspected inductor element and stores them in the storage unit 33.

[0153] The trained model generation unit 32A generates a trained model 35 (Rse=f(Rdc4,Rdc2)+g(Rdc4)) by individually training the first model (Rc=f(Rdc4,Rdc2)) and the second model (Rs0=g(Rdc4)), similar to the trained model generation unit 32 in Embodiment 1. That is, the trained model generation unit 32A separately executes the process of adjusting the parameters of the first model (Rc=f(Rdc4,Rdc2)) using machine learning and the process of adjusting the parameters of the second model (Rs0=g(Rdc4)) using machine learning. The second model (Rs0=g(Rdc4)) is also referred to as "model g(Rdc4)".

[0154] The learned model generation unit 32A generates a model g(Rdc4) by performing machine learning on learning measurement data 34 with small variations in the measured values of the DC resistance (variations in the contact resistance), and generates f(Rdc4, Rdc2) by performing machine learning on learning measurement data 34 with large variations in the measured values of the DC resistance.

[0155] Specifically, the learned model generation unit 32A generates g(Rdc4) by performing machine learning on learning measurement data 34 where the difference between the measured value Rdc4 of the DC resistance and the measured value Rdc2 of the DC resistance is smaller than the threshold value Rdth (|Rdc2 - Rdc4| < Rdth). Also, similar to the learned model generation unit 32 according to Embodiment 1, the learned model generation unit 32A generates f(Rdc4, Rdc2) by performing machine learning on learning measurement data 34 where the difference between the measured value Rdc4 of the DC resistance and the measured value Rdc2 of the DC resistance is larger than the threshold value (|Rdc2 - Rdc4| > Rdth).

[0156] Similar to Embodiment 1, the number of learning measurement data 34 used for machine learning when generating the model g(Rdc4) may be less than the number of learning measurement data 34 used for machine learning when generating the model f(Rdc4, Rdc2).

[0157] Also, the learned model generation unit 32A adjusts the parameters of f(Rdc4, Rdc2) after adjusting the parameters of the model g(Rdc4). In other words, the trained model generation unit 32A first obtains trained parameters for model g(Rdc4) using the method described above. For example, the trained model generation unit 32A calculates the difference (error) between the value of the resistance component (Rs0) caused by the DUT, which is calculated by inputting the measured value Rdc4 of the DC resistance of the training measurement data 34 with small variation in DC resistance from the training measurement data 34_1 to 34_n into the regression model (g(Rdc4)), and the correct value, which is the value of the resistance component caused by the DUT of the training measurement data 34 with small variation in DC resistance (measured value of AC resistance Rs). Next, the trained model generation unit 32A generates model g(Rdc4) by sequentially updating the parameters of the regression model (g(Rdc4)) so that the calculated error becomes smaller, for example by backpropagation, and stores it in the storage unit 33.

[0158] Next, the trained model generation unit 32A adjusts the parameters of f(Rdc4,Rdc2) by machine learning the training measurement data 34 in a model represented as the sum of f(Rdc4,Rdc2) and model g(Rdc4). Specifically, similar to the trained model generation unit 32 in Embodiment 1, the trained model generation unit 32A adjusts the parameters of f(Rdc4,Rdc2) in the model (Rse=f(Rdc4,Rdc2)+g(Rdc4)) by machine learning the training measurement data 34 (|Rdc2―Rdc4|>Rdth) in which the difference between the measured value of DC resistance Rdc4 and the measured value of DC resistance Rdc2 is greater than a threshold, while fixing the trained parameters of model g(Rdc4). At this time, the trained parameters of model g(Rdc4) are not updated.

[0159] The trained model generation unit 32A stores a regression model (Rse=f(Rdc4,Rdc2)+g(Rdc4)) containing the adjusted trained parameters as a trained model 35 in the storage unit 33.

[0160] The overall flow of the method for generating a learned model according to Embodiment 2 is the same as that of the method for generating a learned model according to Embodiment 1 (Steps S1 to S4). Some of the processes in the machine learning according to Embodiment 2 (Step S3) are different from some of the processes in the machine learning according to Embodiment 1.

[0161] FIG. 8 is a flowchart showing the flow of the machine learning (Step S3) according to Embodiment 2.

[0162] In Step S3, first, the learned model generation unit 32A extracts the learning measurement data 34A with a small variation in DC resistance from the learning measurement data 34A_1 to 34A_n (Step S31). Specifically, as described above, the learning measurement data 34A (|Rdc2 - Rdc4| < Rdth) in which the difference between the measured value Rdc4 of the DC resistance and the measured value Rdc2 of the DC resistance is smaller than the threshold value is extracted.

[0163] Next, the learned model generation unit 32A generates the model g(Rdc4) using the learning measurement data 34 extracted in Step S31 by the method described above (Step S32A).

[0164] Next, the learned model generation unit 32A extracts the learning measurement data 34A with a large variation in DC resistance from the learning measurement data 34A_1 to 34A_n (Step S33). Specifically, as described above, the learning measurement data 34A (|Rdc2 - Rdc4| > Rdth) in which the difference between the measured value Rdc4 of the DC resistance and the measured value Rdc2 of the DC resistance is larger than the threshold value is extracted.

[0165] Next, the learned model generation unit 32A generates the model f(Rdc2, Rdc4) using the learning measurement data 34A extracted in Step S33 by the method described above (Step S34).

[0166] Through the above processing flow, the parameters of the model g(Rdc4) and the parameters of the model f(Rdc2, Rdc4) are individually adjusted based on different learning measurement data 34, and a learned model 35 (Rse = f(Rdc2, Rdc4) + g(Rdc4)) is generated.

[0167] FIG. 9 is a diagram showing an example of the configuration of the data processing control device 10A in the inspection device 2A according to the second embodiment.

[0168] As shown in FIG. 9, the data processing control device 10A of the inspection device 2A stores the learned model 35A (Rse = f(Rdc2, Rdc4) + g(Rdc4)) in the storage unit 22. Similar to the data processing control device 10 according to the first embodiment, the data processing control device 10A calculates the estimated value Rse of the AC resistance of the DUT using the learned model 35A (Rse = f(Rdc2, Rdc4) + g(Rdc4)). For other processes, they are the same as those by the data processing control device 10 according to the first embodiment.

[0169] As described above, the learned model generation device 3A according to the second embodiment generates the model g(Rdc4) by machine learning the learning measurement data 34 (|Rdc2 - Rdc4| < Rdth) in which the difference between the measured value Rdc4 of the DC resistance and the measured value Rdc2 of the DC resistance is smaller than the threshold value, and generates the model f(Rdc2, Rdc4) by machine learning the learning measurement data 34 (|Rdc2 - Rdc4| > Rdth) in which the difference between the measured value Rdc4 of the DC resistance and the measured value Rdc2 of the DC resistance is larger than the threshold value.

[0170] According to this, similar to the learned model generation device 3 according to the first embodiment, it is possible to obtain more accurate models g(Rdc4) and f(Rdc2, Rdc4).

[0171] ≪Expansion of the Embodiment≫ Although the invention made by the present inventor has been specifically described above based on embodiments, it goes without saying that the present invention is not limited thereto and can be modified in various ways without departing from its essence.

[0172] For example, the model (trained model 35) used by the inspection device 2 for inspection only needs to be a function that shows the correspondence between the measured values ​​of DC resistance Rdc4 and Rdc2 and the measured value of AC resistance Rs, and may be a model generated by a method other than machine learning. For example, the inspection device 2 may use a model that includes a model f(Rdc4, Rdc2) whose coefficients have been adjusted by a method other than machine learning, and a model g(Ls) (or model g(Rdc4)), and perform inspection of the inductor element in the same manner as described above.

[0173] Furthermore, in the above embodiment, the example given was that the inspection device 2(2A) is an integrated device comprising components such as the data processing control device 10(10A), the first measurement unit 11, the second measurement unit 12, the operation unit 13, the output unit 14, and the transport mechanism 15. However, some components constituting the inspection device 2(2A) may be configured separately from other components. For example, the data processing control device 10(10A), the operation unit 13, and the output unit 14 may be implemented by a first device (for example, an information processing device such as a PC), while the first measurement unit 11, the second measurement unit 12, and the transport mechanism 15 may be implemented by a second device different from the first device. In this case, the first device and the second device may be connected via a wired or wireless network.

[0174] The flowchart described above is merely an example illustrating the operation, and is not limited to it. In other words, the steps shown in each diagram of the flowchart are specific examples and are not limited to this flow. For example, the order of some processes may be changed, other processes may be inserted between each process, or some processes may be performed in parallel. [Explanation of symbols]

[0175] 1,1A…Inspection system, 2,2A…Inspection device, 3,3A…Trained model generation device, 10,10A…Data processing control device, 11…First measurement unit, 12…Second measurement unit, 13…Operation unit, 14…Output unit, 15…Transport mechanism, 21…Data acquisition unit, 22…Storage unit, 23…Estimation unit, 24…Correction unit, 25…Determination unit, 31,31A…Training measurement data acquisition unit, 32,32A…Trained model generation unit, 33…Storage unit, 34,34_1 ~34_n,34A,34A_1~34A_n…Training measurement data, 35,35A…Trained model, 50…Measurement data, Rc…Resistance component due to the measurement system using the two-terminal method, Rdc2…Measured value of DC resistance using the two-terminal method, Rdc4…Measured value of DC resistance using the four-terminal method, Rs…Measured value of AC resistance using the two-terminal method, Rse…Estimated value of the measured value of AC resistance using the two-terminal method, Rsr…Value of AC resistance, Rth,Rdth…Threshold.

Claims

1. A learning measurement data acquisition unit acquires learning measurement data by associating a first measured value of the DC resistance of the object to be measured measured by the four-terminal method, a second measured value of the DC resistance of the object to be measured measured by the two-terminal method, a fourth measured value of the inductance of the object to be measured measured by the two-terminal method, with a third measured value of the AC resistance of the object to be measured measured by the two-terminal method. The system includes a trained model generation unit that generates a trained model for causing a computer to function to calculate the third measurement based on input data including the first measurement, second measurement, and fourth measurement, by machine learning the aforementioned training measurement data. The trained model is represented by the sum of a first model, which uses the first and second measured values ​​as explanatory variables and the value of the resistance component due to the two-terminal measurement system as the objective variable, and a second model, which uses the fourth measured value as an explanatory variable and the value of the resistance component due to the object being measured as the objective variable. The trained model generation unit generates the second model by machine learning the fourth measurement value included in the training measurement data of the object to be measured where the difference between the first measurement value and the second measurement value is less than a threshold, and generates the first model by machine learning the first and second measurement values ​​included in the training measurement data of the object to be measured where the difference is greater than the threshold. A pre-trained model generator.

2. A learning measurement data acquisition unit acquires learning measurement data by associating a first measured value of the DC resistance of the object to be measured, measured by the four-terminal method, a second measured value of the DC resistance of the object to be measured, measured by the two-terminal method, with a third measured value of the AC resistance of the object to be measured, also measured by the two-terminal method. The system includes a trained model generation unit that generates a trained model for causing a computer to function to calculate a third measurement based on input data including the first and second measurement values ​​by machine learning the aforementioned training measurement data, The trained model is represented by the sum of a first model, in which the first and second measured values ​​are explanatory variables and the value of the resistance component due to the two-terminal measurement system is the objective variable, and a second model, in which the first measured value is the explanatory variable and the value of the resistance component due to the object being measured is the objective variable. The trained model generation unit generates the second model by machine learning the first measurement value included in the training measurement data of the object to be measured where the difference between the first measurement value and the second measurement value is less than a threshold, and generates the first model by machine learning the first and second measurement values ​​included in the training measurement data of the object to be measured where the difference is greater than the threshold. A pre-trained model generator.

3. In the trained model generation device according to claim 1 or 2, The number of training measurement data used for machine learning when generating the second model is less than the number of training measurement data used for machine learning when generating the first model. A pre-trained model generator.

4. In the trained model generation device according to claim 1 or 2, When the trained model generation unit generates the first model, it adjusts the parameters of the first model while fixing the parameters of the second model in the model represented by the sum of the first and second models. A pre-trained model generator.

5. A data acquisition unit that acquires a first measured value of the DC resistance of the object to be measured measured by the four-terminal method, a second measured value of the DC resistance of the object to be measured measured by the two-terminal method, a third measured value of the AC resistance of the object to be measured measured by the two-terminal method, and a fourth measured value of the inductance of the object to be measured measured by the two-terminal method. A storage unit that stores a trained model for causing a computer to function to calculate the third measurement based on the input first measurement, second measurement, and fourth measurement; The system includes an estimation unit that calculates estimated values ​​of the third measurement corresponding to the first measurement, second measurement, and fourth measurement acquired by the data acquisition unit, based on the trained model stored in the storage unit, The trained model is represented by the sum of a first model, which uses the first and second measured values ​​as explanatory variables and the value of the resistance component due to the two-terminal measurement system as the objective variable, and a second model, which uses the fourth measured value as an explanatory variable and the value of the resistance component due to the object being measured as the objective variable. The second model is generated by machine learning the fourth measurement of a learning measurement data set, which is obtained by associating the third measurement with the first, second, and fourth measurements that have been measured in advance, and in which the difference between the first and second measurements is smaller than a threshold. The first model is generated by machine learning the first and second measurement values ​​of the training measurement data from among the plurality of training measurement data, where the difference is greater than the threshold. Inspection device.

6. A data acquisition unit that acquires a first measured value of the DC resistance of the object to be measured by the four-terminal method, a second measured value of the DC resistance of the object to be measured by the two-terminal method, and a third measured value of the AC resistance of the object to be measured by the two-terminal method. A storage unit that stores a trained model for causing the computer to function to calculate the third measurement based on the input first and second measurement values, The system includes an estimation unit that calculates estimated values ​​of the third measurement corresponding to the first and second measurement obtained by the data acquisition unit, based on the trained model stored in the storage unit, The trained model is represented by the sum of a first model, in which the first and second measured values ​​are explanatory variables and the value of the resistance component due to the two-terminal measurement system is the objective variable, and a second model, in which the first measured value is the explanatory variable and the value of the resistance component due to the object being measured is the objective variable. The second model is generated by machine learning the learning measurement data from a plurality of learning measurement data sets, which are obtained by associating the first measurement value with the second measurement value and the third measurement value, where the difference between the first measurement value and the second measurement value is smaller than a threshold. The first model is generated by performing machine learning on the training measurement data from among the plurality of training measurement data where the difference is greater than the threshold. Inspection device.

7. In the inspection apparatus according to claim 5 or 6, The number of training measurement data used for machine learning when generating the second model is less than the number of training measurement data used for machine learning when generating the first model. Inspection device.

8. In the inspection apparatus according to claim 5 or 6, The parameters of the first model are adjusted while the parameters of the second model are fixed in a model that is represented as the sum of the first model and the second model. Inspection device.

9. In the inspection apparatus according to claim 5 or 6, The system further includes a correction unit that performs a correction process to correct the third measurement value based on the first and second measurement values ​​acquired by the data acquisition unit, according to the estimated value of the third measurement value, and outputs the corrected third measurement value as the value of the AC resistance of the object being measured. Inspection device.

10. In the inspection apparatus according to claim 9, The correction unit calculates the resistance component caused by the two-terminal measurement system according to the first model based on the first and second measurement values ​​acquired by the data acquisition unit, and as a correction process, corrects the third measurement value based on the resistance component caused by the two-terminal measurement system. Inspection device.

11. In the inspection apparatus according to claim 10, The correction unit calculates the error between the estimated value of the third measurement calculated by the estimation unit and the third measurement acquired by the data acquisition unit, and performs the correction process if the error is smaller than a predetermined threshold. Inspection device.

12. The first step is to acquire learning measurement data by associating a first measured value of the DC resistance of the object to be measured measured by the four-terminal method, a second measured value of the DC resistance of the object to be measured measured by the two-terminal method, a fourth measured value of the inductance of the object to be measured measured by the two-terminal method, with a third measured value of the AC resistance of the object to be measured measured by the two-terminal method. The process includes a second step of generating a trained model by machine learning the training measurement data acquired in the first step, causing the computer to function to calculate the third measurement based on input data including the first measurement, the second measurement, and the fourth measurement, The trained model is represented by the sum of a first model, which uses the first and second measured values ​​as explanatory variables and the value of the resistance component due to the two-terminal measurement system as the objective variable, and a second model, which uses the fourth measured value as an explanatory variable and the value of the resistance component due to the object being measured as the objective variable. The second step includes generating the second model by machine learning the training measurement data of the object to be measured in which the difference between the first measurement and the second measurement is less than a threshold, and generating the first model by machine learning the training measurement data of the object to be measured in which the difference is greater than the threshold. Method for generating pre-trained models.

13. The first step is to acquire learning measurement data by associating a first measured value of the DC resistance of the object to be measured, measured by the four-terminal method, a second measured value of the DC resistance of the object to be measured, measured by the two-terminal method, with a third measured value of the AC resistance of the object to be measured, measured by the two-terminal method. The second step includes, by machine learning the training measurement data acquired in the first step, generating a trained model that causes a computer to function to calculate a third measurement based on input data including the first and second measurement values, The trained model is represented by the sum of a first model, in which the first and second measured values ​​are explanatory variables and the value of the resistance component due to the two-terminal measurement system is the objective variable, and a second model, in which the first measured value is the explanatory variable and the value of the resistance component due to the object being measured is the objective variable. The second step includes generating the second model by machine learning the training measurement data of the object to be measured in which the difference between the first measurement and the second measurement is less than a threshold, and generating the first model by machine learning the training measurement data of the object to be measured in which the difference is greater than the threshold. Method for generating pre-trained models.

14. The computer is made to perform each step in the trained model generation method described in claim 12 or 13. A program for generating pre-trained models.

15. The first step is to obtain a first measured value of the DC resistance of the object to be measured using the four-terminal method, a second measured value of the DC resistance of the object to be measured using the two-terminal method, a third measured value of the AC resistance of the object to be measured using the two-terminal method, and a fourth measured value of the inductance of the object to be measured using the two-terminal method. A second step of calculating an estimated value of the third measurement corresponding to the first measurement, second measurement, and fourth measurement obtained in the first step, based on a trained model that causes the computer to function to estimate the third measurement based on the input first measurement, second measurement, and fourth measurement; The third step includes correcting the third measurement value based on the first and second measurement values ​​obtained in the first step, according to the estimated value of the third measurement value, and outputting the corrected third measurement value as the value of the AC resistance of the object to be measured. The trained model is represented by the sum of a first model, which uses the first and second measured values ​​as explanatory variables and the value of the resistance component due to the two-terminal measurement system as the objective variable, and a second model, which uses the fourth measured value as an explanatory variable and the value of the resistance component due to the object being measured as the objective variable. The second model is generated by machine learning the learning measurement data from a plurality of learning measurement data obtained by associating the third measurement with the first measurement, second measurement, and fourth measurement, which have been measured in advance, and in which the difference between the first measurement and the second measurement is smaller than a threshold. The first model is generated by performing machine learning on the training measurement data from among the plurality of training measurement data where the difference is greater than the threshold. Testing method.

16. The first step is to obtain a first measured value of the DC resistance of the object to be measured using the four-terminal method, a second measured value of the DC resistance of the object to be measured using the two-terminal method, and a third measured value of the AC resistance of the object to be measured using the two-terminal method. A second step of calculating an estimated value of the third measurement corresponding to the first and second measurements obtained in the first step, based on a trained model that causes the computer to function to estimate the third measurement based on the input first and second measurements, The third step includes correcting the third measurement value based on the first and second measurement values ​​obtained in the first step, according to the estimated value of the third measurement value, and outputting the corrected third measurement value as the value of the AC resistance of the object to be measured. The trained model is represented by the sum of a first model, in which the first and second measured values ​​are explanatory variables and the value of the resistance component due to the two-terminal measurement system is the objective variable, and a second model, in which the first measured value is the explanatory variable and the value of the resistance component due to the object being measured is the objective variable. The second model is generated by machine learning the learning measurement data from a plurality of learning measurement data sets, in which the difference between the first measurement and the second measurement is smaller than a threshold, obtained by associating the third measurement with the first measurement and the second measurement, which have been measured in advance. The first model is generated by performing machine learning on the training measurement data from among the plurality of training measurement data where the difference is greater than the threshold. Testing method.

17. Cause the computer to perform each step of the inspection method described in claim 15 or 16. Testing program.

Images