Characteristic prediction system

Abstract

A characteristic prediction system (100) is provided with a prediction computer (10), a measurement computer (20), and a measurement instrument (30). The measurement computer (20) is provided with a processor (21) that acquires the electrical characteristics of the semiconductor chip measured by the measurement device (30). A prediction computer (10) is provided with: a memory device (12) for storing a prediction model for predicting an out-of-range characteristic, which is an electrical characteristic exceeding a range that can be measured by a measurement device (30); and a processor (11) that uses the prediction model to predict out-of-range characteristics from at least the electrical characteristics acquired by the processor (21).

Description

[0001] Cross-references of related applications

[0002] This application is based on Japanese Patent Application No. 2022-143961 filed in Japan on September 9, 2022, and the contents of the basic application are cited herein in their entirety by reference. Technical Field

[0003] The present disclosure relates to a property prediction system. Background Art

[0004] Patent document 1 describes a method for manufacturing a semiconductor device. In the method for manufacturing a semiconductor device, an aging necessity determination process is performed to determine whether an aging test is necessary for each semiconductor chip based on measurement data in a probe test process. In addition, in the method for manufacturing a semiconductor device, based on the determination result of the aging necessity determination process, a first lot (lot) including packages composed of semiconductor chips determined to require an aging test and a second lot including packages composed of semiconductor chips determined to be unnecessary are divided. In addition, in the method for manufacturing a semiconductor device, an aging test is performed only on the packages included in the first lot.

[0005] Prior art literature

[0006] Patent Literature

[0007] Patent Document 1: Japanese Patent No. 6310782 Summary of the invention

[0008] However, semiconductor devices are sometimes required to operate in high temperature environments and large currents. Therefore, semiconductor devices are required to maintain specifications related to electrical characteristics in high temperature environments and large currents. However, in the case of mass production of semiconductor devices, it may be difficult to measure electrical characteristics using a measuring instrument under conditions such as high temperature environments and large currents.

[0009] One object of the disclosure is to provide a characteristic prediction system capable of predicting electrical characteristics that are difficult to measure with a measuring instrument.

[0010] The characteristic prediction system disclosed herein is characterized in that it comprises: a measuring device for obtaining electrical characteristics of a semiconductor device measured by a measuring instrument; a storage unit for storing a prediction model, the prediction model being used to predict electrical characteristics that exceed the range that can be measured by the measuring instrument, i.e., out-of-range characteristics; and a prediction device for predicting out-of-range characteristics based on at least the electrical characteristics obtained by the measuring device using the prediction model.

[0011] According to the characteristic prediction system disclosed herein, a storage unit is provided in which a prediction model for predicting out-of-range characteristics is stored. Furthermore, the characteristic prediction system uses the prediction model to predict out-of-range characteristics based on at least the electrical characteristics obtained by the measuring device. Therefore, the characteristic prediction system can predict out-of-range characteristics that are difficult to measure by the measuring device.

[0012] The multiple technical solutions disclosed in this specification adopt different technical means to achieve their respective purposes. The marks in brackets recorded in the claims and their items are illustrative of the correspondence with the parts of the embodiments described later, and are not intended to limit the technical scope. The purposes, features and effects disclosed in this specification will be more clearly understood by referring to the subsequent detailed description and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is a block diagram showing a schematic structure of a characteristic prediction system.

[0014] Figure 2 This is a flowchart showing the processing operation of the characteristic prediction system.

[0015] Figure 3 It is a plan view showing a semiconductor wafer.

[0016] Figure 4 It is a diagram showing the prediction results of the characteristic prediction system.

[0017] Figure 5 This is a diagram showing measurement data measured by a measuring device.

[0018] Figure 6 This is a graph showing non-measured data.

[0019] Figure 7 It is a graph representing teacher data.

[0020] Figure 8 is a graph showing the predicted data.

[0021] Fig. 9 It is a diagram showing the relationship between measured data and predicted data.

[0022] Fig.10 : is a flowchart showing the processing operation of the characteristic prediction system according to the modification example.

[0023] Fig.11 It is a graph showing the classification results. DETAILED DESCRIPTION

[0024] Hereinafter, an embodiment for implementing the present disclosure will be described with reference to the accompanying drawings. The characteristic prediction system 100 is a system for predicting the electrical characteristics of a plurality of semiconductor chips C1 to Cn formed on a semiconductor wafer 200. In particular, the characteristic prediction system 100 is a system for predicting electrical characteristics that are difficult to measure by the measuring device 30 when the semiconductor chips C1 to Cn are mass-produced. In addition, in the figure, the characteristic prediction system is denoted as PDIT, the semiconductor wafer is denoted as SEM, and the measuring device is denoted as MES. In addition, in the figure, the measurement computer is denoted as 1COM, the prediction computer is denoted as 2COM, the processor is denoted as PCS, and the memory device is denoted as MEM. Each semiconductor chip C1 to Cn is equivalent to a semiconductor device.

[0025] <Semiconductor Wafer>

[0026] like Figure 3 As shown, the semiconductor wafer 200 is formed with a plurality of semiconductor chips C1 to Cn. In detail, the semiconductor wafer 200 is formed with constituent elements of the semiconductor chips C1 to Cn on a substrate mainly composed of Si, SiC, etc. by a semiconductor process. In other words, the semiconductor wafer 200 is formed with a plurality of regions that become a plurality of semiconductor chips C1 to Cn. That is, the semiconductor wafer 200 is divided into a plurality of semiconductor chips C1 to Cn by dicing. In addition, the semiconductor chips C1 to Cn are, for example, MOSFET, IGBT, etc. The constituent elements are, for example, a gate electrode, a drain electrode, a source electrode, a drift layer, a buffer layer, a trench structure, etc. n is a natural number greater than 2.

[0027] <Structure>

[0028] First, use Figure 1 , the structure of the characteristic prediction system 100 will be described. The characteristic prediction system 100 includes a measurement computer 20, a prediction computer 10, and a measurement device 30.

[0029] The measurement computer 20 includes a processor 21 such as a CPU, a memory device 22 including a volatile memory and a nonvolatile memory, etc. The measurement computer 20 is configured to be able to communicate with the measurement device 30 and the prediction computer 10 .

[0030] The processor 21 executes the program stored in the memory device 22. The processor 21 executes the program and performs arithmetic processing to perform various controls. For example, the processor 21 controls the measuring device 30. The processor 21 instructs the measuring device 30 to measure the electrical characteristics of each semiconductor chip C1 to Cn. The processor 21 obtains the measurement result measured by the measuring device 30, that is, the electrical characteristics. In addition, the processor 21 inputs the electrical characteristics into the prediction computer 10. The processor 21 is equivalent to a measuring device. The processor 21 can also be said to be a measuring processor. In addition, the memory device 22 stores programs executed by the processor 21, electrical characteristics, etc.

[0031] The measuring device 30 measures the electrical characteristics of each semiconductor chip C1 to Cn according to the instruction from the processor 21. The measuring device 30 measures the electrical characteristics within the measurable range (conditions) of the measuring device 30. The measuring device 30 inputs measurement data including the measured electrical characteristics to the prediction computer 10.

[0032] In addition, the range that can be measured is a small current and a low temperature environment. A small current is a current that can be measured by the measuring device 30. A low temperature is a temperature that can be measured by the measuring device 30. In addition, the measuring device 30 here is not a device that can measure when a large current exceeds 1000A or in a high temperature environment of 200°C. The electrical characteristics measured by the measuring device 30 can be, for example, electrical characteristics when a small current is present, electrical characteristics in a low temperature environment, or electrical characteristics when a small current is present in a low temperature environment.

[0033] The measuring device 30 is as follows Figure 5 As shown, as electrical characteristics, the on-state voltage, threshold voltage, withstand voltage, forward voltage, and switching loss (Eon, Eoff, Err) are measured for each semiconductor chip C1 to Cn. In addition, the measuring device 30 can also measure the saturation current, capacitance (Crss, Ciss, Coss), voltage change (dv / dt), current change (di / dt), recovery current, surge voltage, total gate charge (Qg), etc.

[0034] Eon is the turn-on loss that occurs when the power is turned on. Eoff is the turn-off loss that occurs when the power is turned off. Err is the recovery loss.

[0035] Crss is the feedback capacitance equivalent to the parasitic capacitance between the gate and drain. Ciss is the sum of the parasitic capacitance between the gate and source and the parasitic capacitance between the gate and drain, and is the input capacitance related to the gate charging and discharging speed. Coss is the sum of the parasitic capacitance between the drain and source and the parasitic capacitance between the gate and drain, and is the output capacitance related to the switching speed between the source and drain.

[0036] dv / dt is the amount of change in drain-source voltage per unit time that occurs during the switching transition period. di / dt is the amount of change in the recovery current flowing through the body diode of the MOSFET.

[0037] In addition, as the above electrical characteristics, the measuring device 30 may also measure short circuit withstand, voltage change withstand, recovery withstand, avalanche withstand, etc. Furthermore, as the above electrical characteristics, the measuring device 30 may perform a gate screening test, a withstand voltage screening test, a reverse bias test, a high temperature gate bias test, etc. In addition, the measuring device 30 may be used as long as it measures at least one of the above electrical characteristics.

[0038] The prediction computer 10 includes a processor 11 such as a CPU, a memory device 12 including a volatile memory and a non-volatile memory, etc. The prediction computer 10 is configured to be able to communicate with the measurement computer 20. The prediction computer 10 is configured to be able to input input data and output output data. The input data is the measurement data output from the measurement computer 20, the teacher data for generating the prediction model, etc. The output data is the prediction data predicted by the prediction computer 10.

[0039] The processor 11 executes the program stored in the memory device 12. The processor 11 executes the program and performs arithmetic processing to perform various controls. For example, the processor 11 generates a prediction model by machine learning using teacher data. Figure 4 As shown, the processor 11 uses the prediction model to predict the prediction data based on at least the measurement data. The processor 11 is equivalent to the prediction device. The processor 11 can also be said to be a prediction processor. In addition, the memory device 12 stores the program executed by the processor 11, the prediction model, the prediction data, etc.

[0040] Furthermore, the prediction computer 10 may predict the prediction data based on the measurement data and non-measurement data on each of the semiconductor chips C1 to Cn that is different from the electrical characteristics. The non-measurement data corresponds to non-electrical information.

[0041] exist Figure 6 An example of unmeasured data is represented in . The unmeasured data includes at least one of the coordinate information of each semiconductor chip C1~Cn in the semiconductor wafer 200, the composition information of each semiconductor chip C1~Cn in the semiconductor wafer 200, and the manufacturing process information of the semiconductor device. The coordinate information is information indicating the X coordinate and Y coordinate of each semiconductor chip C1~Cn. The composition information includes information on the concentration of the drift layer and the film thickness of the drift layer. The manufacturing process information includes information indicating the dimensions of the constituent elements of each semiconductor chip C1~Cn measured in the manufacturing process. The dimensions of the constituent elements are the dimensions of the withstand voltage holding layer, the trench dimensions in the trench MOS, the film thickness of the gate insulating film, etc. The dimensions of these constituent elements can be measured, for example, by a length measurement SEM. Length measurement SEM is the abbreviation of Critical Dimension-Scanning Electron Microscope.

[0042] exist Figure 8An example of predicted data is shown in . The predicted data is the electrical characteristics of each semiconductor chip C1~Cn that exceeds the range that the measuring device 30 can measure. The predicted data is, for example, the predicted value of the electrical characteristics of each semiconductor chip C1~Cn when a large current of more than 1000A flows in a high temperature environment of more than 200°C. In addition, the predicted data can also be the predicted value of the electrical characteristics of each semiconductor chip C1~Cn in a high temperature environment of more than 200°C, for example, or the predicted value of the electrical characteristics of each semiconductor chip C1~Cn when a large current of more than 1000A flows. Therefore, the predicted data can be said to be electrical characteristics that are difficult to measure by the measuring device 30. The predicted data is equivalent to out-of-range characteristics. The predicted data can be, for example, electrical characteristics at a large current, electrical characteristics in a high temperature environment, or electrical characteristics at a large current in a high temperature environment.

[0043] The prediction computer 10 predicts, for example, the value of the electrical characteristics measured by the above-mentioned measuring device 30 in a high temperature environment and a large current as prediction data. In addition, as prediction data, the prediction computer 10 can also predict the results of a gate shielding test, a withstand voltage shielding test, a reverse bias test, a high temperature gate bias test, etc. in a high temperature environment and a large current. Furthermore, as prediction data, the prediction computer 10 can also predict durability performance and market failure rate. The prediction computer 10 can adopt as long as it predicts at least one of the multiple electrical characteristics as prediction data.

[0044] The prediction model is used to predict the electrical characteristics of each semiconductor chip C1 to Cn. In particular, in the present embodiment, a prediction model is used to predict the prediction data of the electrical characteristics of each semiconductor chip C1 to Cn that exceed the range that the measuring device 30 can measure. The prediction model can also be said to be a mechanism or algorithm for predicting the prediction data based on the measurement data as input data. Therefore, it can be said that the processor 11 uses the prediction model to predict the electrical characteristics of each semiconductor chip C1 to Cn under conditions that are difficult for the measuring device 30 to measure.

[0045] The prediction model is a machine learning model generated by a well-known method. Machine learning includes deep learning and simple perceptron. In the case of deep learning, it can also be based on simple perceptron or multi-layer perceptron. In the case where the teacher data is an unbalanced data set, it can also include oversampling or undersampling processes. In addition, Fig. 9 An example of the forward voltage predicted (anticipated) by the multilayer perceptron is shown in FIG. Fig. 9 The horizontal axis is the measured data (actual values), and the vertical axis is the predicted data. The machine learning model can also be said to be a learned model.

[0046] Teacher data such as Figure 7As shown, there are measurement data of multiple measurement results at low currents, electrical characteristics in high temperature environments or at high currents, and data outside of measurements. Figure 7 In the figure, only the teacher data related to the first chip of the first batch is illustrated. However, the teacher data is not limited to this. The teacher data may also include measurement data and non-measurement data of multiple chips generated from multiple batches. In addition, the electrical characteristics at high current in the teacher data are, for example, electrical characteristics measured at high current for several semiconductor chips among the multiple semiconductor chips C1~Cn. In addition, the electrical characteristics under high temperature environment in the teacher data are, for example, electrical characteristics measured in a high temperature environment for several semiconductor chips among the multiple semiconductor chips C1~Cn.

[0047] <Processing Action>

[0048] Here, use Figure 2 , prediction processing of prediction data will be described. First, in the chip preparation step S10, a semiconductor wafer 200 on which a plurality of semiconductor chips C1 to Cn are formed is prepared. That is, in the chip preparation step S10, a semiconductor wafer 200 to be predicted is prepared.

[0049] In the electrical test step S11 , the processor 21 of the measurement computer 20 issues a measurement instruction to the measurement device 30 . Then, the measurement device 30 measures the electrical characteristics of each of the semiconductor chips C1 to Cn according to the instruction from the processor 21 .

[0050] In the measurement data input step S12, the measuring device 30 Figure 5 The measured electrical characteristics, i.e., measurement data, are input to the measurement computer 20. Thus, the measurement computer 20 acquires measurement data for predicting the prediction data. The measurement data input step S12 can also be said to be a measurement data acquisition step in which the measurement computer 20 acquires the measurement data.

[0051] In the measurement data input step S13, Figure 6 The unmeasured data as shown is input to the measurement computer 20. Thus, the measurement computer 20 obtains the unmeasured data for predicting the predicted data. The unmeasured data input step S13 can also be said to be a step of obtaining unmeasured data for the measurement computer 20 to obtain the unmeasured data.

[0052] In the teacher data input step S14, Figure 7 The teacher data shown is input to the prediction computer 10. Thus, the prediction computer 10 acquires the teacher data for generating the prediction model. The teacher data input step S14 can also be said to be a teacher data acquisition step in which the prediction computer 10 acquires the teacher data.

[0053] In the learned model generation step S15, the prediction computer 10 uses Figure 7 The prediction computer 10 generates a prediction model by machine learning based on a plurality of measurement results at low currents, i.e., measurement data and unmeasured data, and electrical characteristics in a high temperature environment or at high currents, obtained as teacher data.

[0054] In the prediction step S16, the prediction computer 10 predicts the prediction data. The prediction computer 10 uses the prediction model to predict based on the measurement data input in step S12 and the non-measurement data input in step S13. Figure 8 In this way, the prediction computer 10 can obtain the electrical characteristics of each semiconductor chip C1 to C2 which are difficult to measure with the measuring device 30.

[0055] In the prediction step S16, it is preferable that the prediction computer 10 uses two or more measurement data. Thereby, the prediction computer 10 can improve the prediction accuracy, reduce the tolerance, and suppress the testing cost.

[0056] In the chip shipping step S17, the semiconductor chips C1 to C2 for which the prediction data has been measured are shipped. In the chip shipping step S17, only the semiconductor chips C1 to C2 for which the prediction data satisfies the reference value may be shipped.

[0057] The methods and / or functions provided by the computers 10 and 20 can be provided by software recorded in the physical memory devices 12 and 22 and the computers that execute it, software alone, hardware alone, or a combination thereof. For example, when the computers 10 and 20 are provided by electronic circuits as hardware, they can be provided by digital circuits including many logic circuits, or analog circuits.

[0058] <Effect>

[0059] According to the characteristic prediction system 100 disclosed herein, the memory device 12 is provided with a prediction model for predicting prediction data. Furthermore, the characteristic prediction system 100 uses the prediction model to predict the prediction data based on at least the electrical characteristics (measurement data) obtained by the measuring device 30. Therefore, the characteristic prediction system 100 can predict the prediction data (electrical characteristics) that are difficult to measure by the measuring device 30.

[0060] In particular, for semiconductor chips C1 to Cn whose main component is SiC, a temperature exceeding 200°C and a large current exceeding 1000A are required. Even for such semiconductor chips C1 to Cn, specifications are required to be guaranteed through electrical testing. In addition, in the mass production of semiconductor chips C1 to Cn, in order to perform measurements in a high temperature environment and measurements at a large current, an expensive measuring device 30 is required, and it is conceivable that the measurement cost will increase or the measurement itself cannot be performed. In contrast, according to the characteristic prediction system 100, the electrical characteristics of each semiconductor chip C1 to Cn that exceeds the range that the measuring device 30 can measure can be predicted using a prediction model. Therefore, the characteristic prediction system 100 can achieve specification guarantees while suppressing the increase in measurement costs.

[0061] In addition, it is also possible to consider using a function to predict electrical characteristics that exceed the range that the measuring device 30 can measure. However, for semiconductor chips C1 to Cn, it is necessary to consider the decrease in prediction accuracy caused by the epitaxial layer concentration and film thickness deviation and process deviation as tolerances. Furthermore, the function also becomes nonlinear, making it difficult to achieve accuracy. In this case, the amount of tolerance needs to be reflected in the chip performance, resulting in an increase in chip cost. In contrast, according to the characteristic prediction system 100, since the prediction data is predicted using a prediction model, it is possible to make a prediction with high accuracy. As a result, tolerance reduction and measurement cost suppression can be achieved.

[0062] In addition, for the deviation of the concentration and film thickness of the drift layer, there is a strong tendency to become concentric circles, and the position information becomes useful. In addition, the concentration and film thickness of the drift layer have a strong influence on the DC (direct current) characteristics and the AC (alternating current) characteristics. Therefore, the characteristic prediction system 100 preferably uses the concentration and film thickness of the drift layer as the out-of-measurement data. In other words, the characteristic prediction system 100 can improve the prediction accuracy by adding the concentration and film thickness of the drift layer as the out-of-measurement data to the machine learning. It can also be said that the characteristic prediction system 100 can generate a prediction model that can predict with good accuracy by adding the concentration and film thickness of the drift layer as the out-of-measurement data to the machine learning. In addition, the constituent information of the out-of-measurement data may also include information indicating the defect state of the semiconductor wafer 200, and information indicating the concentration and film thickness of the buffer layer.

[0063] The dimensions of the components such as the size of the withstand voltage holding layer, the trench size of the trench MOS, and the film thickness of the gate insulating film have a strong influence on the DC characteristics and AC characteristics. Therefore, the characteristic prediction system 100 can improve the prediction accuracy by adding the dimensions of the components to the machine learning as the measurement out-of-data. In addition, it can also be said that the characteristic prediction system 100 can generate a prediction model that can predict with good accuracy by adding the dimensions of the components to the machine learning.

[0064] The measured data of forward voltage, saturation current, capacitance, switching loss, voltage change, current change, recovery current, surge voltage, total gate charge, etc. are nonlinear with respect to the current value and voltage value. Furthermore, the coefficient also depends on the concentration of the drift layer, the film thickness of the drift layer, and the process deviation of the process, so even if the function is uniquely determined, high prediction accuracy cannot be obtained. Therefore, the characteristic prediction system 100 can improve the prediction accuracy by adding non-measured data to machine learning in addition to the measured data. In addition, it can also be said that the characteristic prediction system 100 can generate a prediction model that can make predictions with good accuracy by adding non-measured data to machine learning in addition to the measured data.

[0065] Regarding short-circuit tolerance, voltage change tolerance, recovery tolerance, avalanche tolerance and other tolerances, it is difficult to implement a load equivalent to the actual operation through the measurement (electrical test) of the semiconductor chips C1~Cn due to the oscillation (ringing) caused by the parasitic inductance of the measuring device 30 and the measurement limit of the measuring device 30. Therefore, it can be considered to implement a sampling test, or to implement a test after the module is built, or to add a test with a lower load than the actual operation to the electrical test. Furthermore, the number of parameters that determine the tolerance is large and complex. Therefore, it is difficult to predict the tolerance under the condition of applying a load equivalent to the actual operation based on the electrical test of the semiconductor chips C1~Cn. In contrast, the characteristic prediction system 100 can predict the tolerance under the condition of applying a desired load because it uses machine learning that can process many parameters and can build complex prediction models.

[0066] The preferred embodiments of the present disclosure have been described above. However, the present disclosure is not limited to the above embodiments, and various modifications can be made without departing from the scope of the present disclosure. The present disclosure is not limited to the combinations shown in the embodiments, and can be implemented in various combinations.

[0067] (Variation Example)

[0068] use Fig.10 , Fig.11 A characteristic prediction system 100 according to a modified example will be described. The characteristic prediction system 100 may also perform a classification process in addition to the process of the above-mentioned embodiment. Fig.10 In the embodiment, the same step numbers are assigned to the same processes as those in the above embodiment. Steps S14 and S15 are Figure 2 Steps S14 and S15 are the same, but the names are changed to distinguish them from steps S20 and S21.

[0069] The characteristic prediction system 100 performs the second teacher data input step S20, the second learned model generation step S21, and the classification step S22 between the prediction step S16 and the chip shipping step S17. However, steps S20 and S21 may be performed before step S16 is completed.

[0070] In the second teacher data input process S20, the second teacher data is input into the prediction computer 10. The second teacher data is teacher data used to generate the second learned model. That is, the second teacher data is different from the teacher data used to generate the prediction model. The prediction computer 10 obtains the second teacher data used to generate the classification model as the second learned model. In addition, the second teacher data input process S20 can also be said to be a second teacher data acquisition process in which the prediction computer 10 obtains the second teacher data.

[0071] In the second learned model generation step S21, the prediction computer 10 generates a classification model. The second learned model is different from the prediction model and is a machine learning model for classification. The prediction computer 10 generates the classification model through machine learning using the second teacher data.

[0072] The classification model may include, for example, logistic regression and linear support vector machines as linear classifiers. In addition, it may also include k-nearest neighbor method, decision tree, random forest, nonlinear support vector machine, and deep learning as nonlinear classifiers. In the case of an unbalanced data set, oversampling and undersampling may also be implemented.

[0073] In the classification step S22, the prediction computer 10 classifies the semiconductor chips C1 to Cn. The prediction computer 10 classifies (identifies, recognizes) the semiconductor chips C1 to Cn using the classification model and at least one prediction data acquired in step S16. That is, the prediction computer 10 determines the boundary line and classifies the semiconductor chips C1 to Cn based on the prediction data.

[0074] exist Fig.11 As an example, the data including one or more on-resistance, turn-on loss, turn-off loss, and recovery loss predicted by the prediction model are shown, and an example of classifying semiconductor chips C1 to Cn based on power loss that combines the turn-on loss and the switching loss is shown. The characteristic prediction system 100 classifies the semiconductor chips into the obliquely shaded category A and the dotted shaded category B based on the boundary line. In addition, as Fig.11 (a) and Fig.11 As shown in (b) of FIG. 1 , the classification in the classification step S22 can be changed according to the purpose. In other words, the boundary line can be changed according to the purpose and the like.

[0075] Thus, the prediction computer 10 obtains Fig.11The classification data for classifying the semiconductor chips C1 to Cn as shown in FIG. That is, the characteristic prediction system 100 predicts the out-of-range characteristics of the plurality of semiconductor chips C1 to Cn, and classifies the plurality of semiconductor chips C1 to Cn based on the prediction data of each semiconductor chip C1 to Cn. By implementing the classification step S22, the characteristic prediction system 100 can reduce the tolerance of each item, and can foresee the improvement of the inverter performance and the improvement of the chip yield.

[0076] In addition, for example, in a converter circuit, a structure in which semiconductor chips are driven in parallel can be considered. In such a converter circuit, a structure having a plurality of semiconductor modules in which a plurality of semiconductor chips are connected in parallel can be adopted. In this case, the characteristic prediction system 100 can also classify the semiconductor chips C1 to Cn using the threshold voltage at high temperature predicted by the prediction model. Thus, a plurality of semiconductor chips classified into the same category can be used in one module and one converter circuit. Therefore, the characteristic prediction system 100 can suppress unbalanced operation of semiconductor chips within or between modules.

[0077] The present disclosure is described based on the embodiments, but it should be understood that the present disclosure is not limited to the embodiments or structures. The present disclosure also includes various modifications or variations within the scope of equality. In addition, various combinations or forms are shown in the present disclosure, but other combinations or forms that only include one element, above or below them also fall within the scope and thought range of the present disclosure.

[0078] (Disclosure of technical ideas)

[0079] This specification discloses multiple technical ideas recorded in the following multiple items. Some items are recorded in a multiple dependent form by selectively citing the previous item in the subsequent item. Furthermore, some items are recorded in a multiple dependent form referring to another multiple dependent form. The items recorded in these multiple dependent forms define multiple technical ideas.

[0080] (Technical Thought 1)

[0081] A characteristic prediction system comprises: a measuring device (21) for obtaining electrical characteristics of a semiconductor device measured by a measuring device (30); a storage unit (12) for storing a prediction model, wherein the prediction model is used to predict the electrical characteristics that exceed the range that can be measured by the measuring device, i.e., out-of-range characteristics; and a prediction device (11) for predicting the out-of-range characteristics based on at least the electrical characteristics obtained by the measuring device using the prediction model.

[0082] (Technical Thought 2)

[0083] In the characteristic prediction system according to technical idea 1, the prediction device generates the prediction model by machine learning using teacher data.

[0084] (Technical Thought 3)

[0085] In the characteristic prediction system according to technical idea 1 or 2, the prediction device predicts the out-of-range characteristic based on the electrical characteristic and non-electrical information about the semiconductor device different from the electrical characteristic using the prediction model.

[0086] (Technical Thought 4)

[0087] In the characteristic prediction system as described in Technical Idea 3, the semiconductor device is one of multiple chips formed on a semiconductor chip; the non-electrical information includes at least one of coordinate information of each chip in the semiconductor chip, composition information of each chip in the semiconductor chip, and manufacturing process information of the semiconductor device.

[0088] (Technical Thought 5)

[0089] In the characteristic prediction system according to technical idea 4, the composition information includes information on the concentration of the drift layer and the film thickness of the drift layer.

[0090] (Technical Thought 6)

[0091] In the characteristic prediction system according to technical idea 4, the manufacturing process information includes information indicating dimensions of components of the semiconductor device measured in the manufacturing process.

[0092] (Technical Thought 7)

[0093] In the characteristic prediction system described in any one of technical ideas 1 to 6, the electrical characteristics obtained by the above-mentioned measuring device include at least one of on-state voltage, threshold voltage, withstand voltage, forward voltage, saturation current, capacitance, switching loss, voltage change, current change, recovery current, surge voltage, and total gate charge.

[0094] (Technical Thought 8)

[0095] In the characteristic prediction system according to any one of technical concepts 1 to 6, the electrical characteristic acquired by the measuring device includes at least one of short-circuit resistance, voltage variation resistance, recovery resistance, and avalanche resistance.

[0096] (Technical Thought 9)

[0097] In the characteristic prediction system described in any one of technical ideas 1 to 8, the semiconductor device is one of multiple chips formed on a semiconductor wafer; the prediction device predicts the out-of-range characteristics of the multiple chips, and classifies the multiple chips based on the out-of-range characteristics of each chip.

Claims

1. A characteristic prediction system, characterized in that: have: A measuring device (21) for obtaining electrical characteristics of a semiconductor device measured by a measuring device (30); A storage unit (12) storing a prediction model for predicting the electrical characteristics exceeding the range measurable by the measuring device, i.e., out-of-range characteristics; and The prediction means (11) predicts the out-of-range characteristics based on at least the electrical characteristics obtained by the measurement means using the prediction model.

2. The characteristic prediction system according to claim 1, wherein: The prediction device generates the prediction model by machine learning using teacher data.

3. The characteristic prediction system according to claim 1 or 2, characterized in that: The prediction device predicts the out-of-range characteristics based on the electrical characteristics and non-electrical information on the semiconductor device that is different from the electrical characteristics, using the prediction model.

4. The characteristic prediction system according to claim 3, wherein: The semiconductor device is one of a plurality of chips formed on a semiconductor wafer; The non-electrical information includes at least one of coordinate information of each chip in the semiconductor wafer, configuration information of each chip in the semiconductor wafer, and manufacturing process information of the semiconductor device.

5. The characteristic prediction system according to claim 4, characterized in that The composition information includes information on the concentration of the drift layer and the film thickness of the drift layer.

6. The characteristic prediction system according to claim 4, characterized in that The manufacturing process information includes information indicating dimensions of components of the semiconductor device measured in the manufacturing process.

7. The characteristic prediction system according to claim 2, wherein: The electrical characteristics obtained by the above-mentioned measuring device include at least one of on-state voltage, threshold voltage, withstand voltage, forward voltage, saturation current, capacitance, switching loss, voltage change, current change, recovery current, surge voltage, and total gate charge.

8. The characteristic prediction system according to claim 2, wherein: The electrical characteristics obtained by the measuring device include at least one of short-circuit resistance, voltage variation resistance, recovery resistance, and avalanche resistance.

9. The characteristic prediction system according to claim 2, wherein: The semiconductor device is one of a plurality of chips formed on a semiconductor wafer; The prediction device predicts the out-of-range characteristics of the plurality of chips, and classifies the plurality of chips based on the out-of-range characteristics of each chip.

Images