JP2021100039A - Device, method, and program - Google Patents

Abstract

To solve the problem in which it is inefficient to identify a configuration by trial and error.SOLUTION: A device includes a configuration acquisition unit that acquires configuration data indicating a configuration of a semiconductor device, a characteristic acquisition unit that acquires characteristic data indicating characteristics of the semiconductor device, and a first learning processing unit that performs a learning process of a first model that outputs recommended configuration data indicating a configuration of a semiconductor device to be recommended in response to an input of target characteristic data indicating a characteristic of a target semiconductor device, using learning data including the acquired configuration data and characteristic data.SELECTED DRAWING: Figure 1

Description

The present invention relates to devices, methods and programs.

Conventionally, when a semiconductor device is manufactured using a film forming apparatus such as an organic metal vapor phase growth apparatus, a preferable configuration has been found by trial and error by a skilled operator in order to obtain a semiconductor device having desired characteristics (). For example, see Patent Document 1).
Patent Document 1 Chinese Patent Application Publication No. 106521459

However, it is inefficient to identify the configuration by trial and error.

In order to solve the above problems, an apparatus is provided in the first aspect of the present invention. The device may include a configuration acquisition unit that acquires configuration data indicating the configuration of the semiconductor device. The device may include a characteristic acquisition unit that acquires characteristic data indicating the characteristics of the semiconductor device. The device uses the acquired configuration data and learning data including the characteristic data to generate the recommended configuration data indicating the configuration of the semiconductor device recommended according to the input of the target characteristic data indicating the characteristics of the target semiconductor device. A first learning processing unit that executes the learning processing of the first model to be output may be provided.

In the second aspect of the present invention, the device is provided. The device may include a configuration acquisition unit that acquires configuration data indicating the configuration of the semiconductor device. The device may include a characteristic acquisition unit that acquires characteristic data indicating the characteristics of the semiconductor device. The apparatus may include a control condition acquisition unit that acquires control condition data indicating the control conditions of the semiconductor device manufacturing apparatus. The device uses the acquired configuration data, characteristic data, and training data including control condition data to input the target characteristic data indicating the characteristics of the target semiconductor device, and the device selects the recommended semiconductor device configuration. A third learning processing unit that executes the learning processing of the third model that outputs at least one of the recommended configuration data shown and the recommended control condition data indicating the recommended control condition may be provided.

In the third aspect of the present invention, the device is provided. The apparatus may include a target characteristic acquisition unit that acquires target characteristic data indicating the characteristics of the target semiconductor device. The apparatus supplies the target characteristic data acquired by the target characteristic acquisition unit to the first model that outputs the recommended configuration data indicating the recommended configuration of the semiconductor device in response to the input of the target characteristic data. It may have a part. The apparatus may include a recommended configuration acquisition unit that acquires recommended configuration data output by the first model in response to supplying target characteristic data to the first model.

In a fourth aspect of the invention, the device is provided. The apparatus may include a target characteristic acquisition unit that acquires target characteristic data indicating the characteristics of the target semiconductor device. The device outputs at least one of the recommended configuration data indicating the recommended semiconductor device configuration and the recommended control condition data indicating the recommended control conditions in response to the input of the target characteristic data for the third model. , The first supply unit may be provided to supply the target characteristic data acquired by the target characteristic acquisition unit. The apparatus may include an acquisition unit that acquires at least one of the recommended configuration data output by the third model in response to the supply of the target characteristic data to the third model and the recommended control condition data.

In a fifth aspect of the invention, a method is provided. The method may include a configuration acquisition step of acquiring configuration data indicating the configuration of the semiconductor device. The method may include a characteristic acquisition step of acquiring characteristic data indicating the characteristics of the semiconductor device. The method uses the acquired configuration data and learning data including the characteristic data to generate the recommended configuration data indicating the configuration of the semiconductor device recommended according to the input of the target characteristic data indicating the characteristics of the target semiconductor device. A first learning processing step for executing the learning processing of the first model to be output may be provided.

In a sixth aspect of the invention, a method is provided. The method may include a configuration acquisition step of acquiring configuration data indicating the configuration of the semiconductor device. The method may include a characteristic acquisition step of acquiring characteristic data indicating the characteristics of the semiconductor device. The method may include a control condition acquisition step of acquiring control condition data indicating the control conditions of the semiconductor device manufacturing apparatus. The method uses the acquired configuration data, characteristic data, and training data including control condition data to input the target characteristic data indicating the characteristics of the target semiconductor device, and the recommended semiconductor device configuration is configured. A third learning processing step for executing the training processing of the third model that outputs at least one of the recommended configuration data shown and the recommended control condition data indicating the recommended control condition may be provided.

In a seventh aspect of the invention, a method is provided. The method may include a target characteristic acquisition step of acquiring target characteristic data indicating the characteristics of the target semiconductor device. The method is to supply the target characteristic data acquired in the target characteristic acquisition stage to the first model that outputs the recommended configuration data indicating the recommended semiconductor device configuration in response to the input of the target characteristic data. It may have stages. The method may include a recommended configuration acquisition step of acquiring recommended configuration data output by the first model in response to supplying target characteristic data to the first model.

In the eighth aspect of the present invention, a method is provided. The method may include a target characteristic acquisition step of acquiring target characteristic data indicating the characteristics of the target semiconductor device. The method is for a third model that outputs at least one of the recommended configuration data indicating the recommended semiconductor device configuration and the recommended control condition data indicating the recommended control conditions in response to the input of the target characteristic data. , A first supply stage may be provided to supply the target characteristic data acquired by the target characteristic acquisition stage. The method may include an acquisition step of acquiring at least one of the recommended configuration data output by the third model in response to the supply of the target characteristic data to the third model and the recommended control condition data.

In a ninth aspect of the invention, a program is provided. The program may cause the computer to function as a configuration acquisition unit that acquires configuration data indicating the configuration of the semiconductor device. The program may cause the computer to function as a characteristic acquisition unit that acquires characteristic data indicating the characteristics of the semiconductor device. The program recommends the configuration of the semiconductor device that the computer recommends according to the input of the target characteristic data indicating the characteristics of the target semiconductor device using the acquired configuration data and the training data including the characteristic data. It may function as the first learning processing unit that executes the learning processing of the first model that outputs the configuration data.

In a tenth aspect of the invention, a program is provided. The program may cause the computer to function as a configuration acquisition unit that acquires configuration data indicating the configuration of the semiconductor device. The program may cause the computer to function as a characteristic acquisition unit that acquires characteristic data indicating the characteristics of the semiconductor device. The program may cause the computer to function as a control condition acquisition unit that acquires control condition data indicating the control conditions of the semiconductor device manufacturing apparatus. The program recommends the computer according to the input of the target characteristic data indicating the characteristics of the target semiconductor device by using the learning data including the acquired configuration data, characteristic data and control condition data. It may function as a third learning processing unit that executes the learning process of the third model that outputs at least one of the recommended configuration data indicating the configuration of the above and the recommended control condition data indicating the recommended control conditions.

In the eleventh aspect of the present invention, a program is provided. The program may allow the computer to function as a target characteristic acquisition unit that acquires target characteristic data indicating the characteristics of the target semiconductor device. The program supplies the target characteristic data acquired by the target characteristic acquisition unit to the first model that outputs the recommended configuration data indicating the configuration of the semiconductor device recommended according to the input of the target characteristic data to the computer. It may function as a first supply unit. The program may function as a recommended configuration acquisition unit that acquires the recommended configuration data output by the first model in response to the supply of the target characteristic data to the first model.

In a twelfth aspect of the invention, a program is provided. The program may function as a target characteristic acquisition unit that acquires target characteristic data indicating the characteristics of the target semiconductor device. The program outputs at least one of the recommended configuration data indicating the recommended semiconductor device configuration and the recommended control condition data indicating the recommended control conditions in response to the input of the target characteristic data to the computer. On the other hand, it may function as a first supply unit that supplies the target characteristic data acquired by the target characteristic acquisition unit. The program may function as an acquisition unit that acquires at least one of the recommended configuration data output by the third model and the recommended control condition data in response to the supply of the target characteristic data to the third model.

The outline of the above invention does not list all the necessary features of the present invention. Sub-combinations of these feature groups can also be inventions.

The apparatus 1 which concerns on this embodiment is shown. An example of the film forming apparatus 2 is shown. The gas supply equipment 2300 is shown. The operation of the device 1 is shown. The first learning process is shown. The second learning process is shown. The manufacturing process is shown. The film forming method using the first model 104 is shown. The film forming method using the second model 109 is shown. The layer structure of the semiconductor device 5 is shown. The apparatus 1A which concerns on the modification is shown. The apparatus 1B which concerns on the modification is shown. An example of a computer 2200 in which a plurality of aspects of the present invention may be embodied in whole or in part is shown.

Hereinafter, the present invention will be described through embodiments of the invention, but the following embodiments do not limit the inventions that fall within the scope of the claims. Also, not all combinations of features described in the embodiments are essential to the means of solving the invention.

[1. apparatus]
FIG. 1 shows the device 1 according to the present embodiment. The device 1 performs learning processing by machine learning regarding the manufacture of the semiconductor device 5 (see FIG. 10), and includes a configuration acquisition unit 101, a characteristic acquisition unit 102, a first learning processing unit 103, and a first model 104. A control condition acquisition unit 105, a state acquisition unit 107, a second learning processing unit 108, a second model 109, and a notification unit 100 are provided. Further, in the present embodiment, as an example, the apparatus 1 can manufacture the semiconductor device 5 by using the first model 104 and / or the second model 109, and the manufacturing unit 110, the target characteristic acquisition unit 111, and the second model It has one supply unit 112, a recommended configuration acquisition unit 113, a second supply unit 115, a recommended control condition acquisition unit 116, a control unit 117, and a determination unit 118.

[1-1. Semiconductor device]
The semiconductor device 5 may be a light emitting device, and although it is an LED as an example in this embodiment, it may be another light emitting device such as an LD. The semiconductor device 5 may emit light at a wavelength in the deep ultraviolet region. The semiconductor device 5 may have at least one layer. The semiconductor device 5 may include an electrode layer.

[1-2. Configuration acquisition department]
The configuration acquisition unit 101 acquires configuration data indicating the configuration of the semiconductor device 5. The configuration acquisition unit 101 may acquire configuration data of the semiconductor device 5 manufactured by the manufacturing unit 110. The configuration of the semiconductor device 5 may be at least one of the number of layers contained in the semiconductor device 5, the film thickness of each layer, the composition of each layer, the impurity concentration of each layer, and the material of the substrate. When the semiconductor device 5 manufactured by the manufacturing unit 110 includes an electrode layer, the configuration of the semiconductor device 5 may further include information about the electrode layer. The configuration data may include configuration parameters for each configuration. The composition may be the composition ratio of the constituent elements or the lattice constant. The composition may include a profile showing changes in the in-plane direction and / or the film thickness direction. The impurities may be n-type or p-type dopants, or may be other impurities. The configuration acquisition unit 101 inputs the configuration data to the operator, the manufacturing unit 110 (for example, the film forming apparatus 2 described later included in the manufacturing unit 110), the control unit 117, and a measuring device for measuring the configuration (not shown). It may be obtained from at least one of. The configuration data acquired from the manufacturing unit 110 and the control unit 117 may be the design data of the semiconductor device 5. The configuration data acquired from the measuring device may be the measurement data of the semiconductor device 5 after manufacturing. The measuring device may be arranged inside the manufacturing unit 110 or may be arranged outside. The configuration acquisition unit 101 may supply the acquired configuration data to the first learning processing unit 103 and the second learning processing unit 108.

[1-3. Characteristic acquisition section]
The characteristic acquisition unit 102 acquires characteristic data indicating the characteristics of the semiconductor device 5. The characteristic acquisition unit 102 may acquire characteristic data of the semiconductor device 5 manufactured by the manufacturing unit 110. The characteristics of the semiconductor device 5 are emission intensity (peak intensity, integrated intensity as an example), emission spectrum, emission peak wavelength, presence / absence of multi-peak, leakage current amount (IV characteristic), drop (efficiency decrease rate with respect to current increase), It may be at least one of the full width at half maximum of the emission spectrum, the color temperature, the emission efficiency, the exothermic property, the responsiveness, the directivity, the temperature dependence, the lifetime, and the driving voltage. The characteristic data may include characteristic parameters for each characteristic.

The characteristic acquisition unit 102 may acquire characteristic data from at least one of an operator and a measuring device (not shown) for measuring the characteristic. The measuring device may be arranged inside the manufacturing unit 110 or may be arranged outside. The characteristic acquisition unit 102 may supply the acquired characteristic data to the first learning processing unit 103 and the second learning processing unit 108.

[1-4. 1st learning processing department]
The first learning processing unit 103 executes the learning process of the first model 104 using the input learning data. The learning data used in the first learning processing unit 103 may include the configuration data and the characteristic data acquired from the configuration acquisition unit 101 and the characteristic acquisition unit 102. The first learning processing unit 103 may accumulate and store the learning data used in the learning processing.

The first learning processing unit 103 may have a second detection unit 1031 that detects the correlation between at least one characteristic parameter included in the characteristic data and at least one configuration parameter included in the configuration data. The second detection unit 1031 may detect the correlation between the designated characteristic parameter specified by the operator and at least one configuration parameter included in the configuration data among the plurality of characteristic parameters included in the characteristic data. The second detection unit 1031 may output a plurality of configuration parameters included in the configuration data in descending order of correlation with the designated characteristic parameters. As a result, the operator can easily grasp the configuration parameter that correlates with any characteristic parameter, and can acquire the semiconductor device 5 having the configuration excellent in the characteristic indicated by the characteristic parameter.

[1-5. 1st model]
The first model 104 outputs the recommended configuration data indicating the configuration of the recommended semiconductor device 5 in response to the input of the target characteristic data indicating the characteristics of the target semiconductor device 5. The target characteristic data may include a target value for each characteristic parameter, or may include a target range for at least one characteristic parameter. In the first model 104, the learning process may be executed using a plurality of training data including the values of at least one configuration parameter as the configuration data, and a plurality of recommended configuration data are supplied according to the supply of one target characteristic data. May be output. Each of these plurality of recommended configuration data may include the value of at least one recommended configuration parameter indicating the configuration of the recommended semiconductor device 5.

The first model 104 may be stored in a server outside the device 1. The first model 104 outputs the recommended configuration data to the recommended configuration acquisition unit 113, but may output the recommended configuration data to the outside of the device 1.

[1-6. Control condition acquisition unit]
The control condition acquisition unit 105 acquires control condition data indicating the control conditions of the manufacturing unit 110. The control conditions of the manufacturing unit 110 may be conditions that can be set via the control unit 117, although details will be described later. The control conditions shown in the control condition data may include conditions different from the conditions determined by the configuration of the semiconductor device 5, or may include conditions that more specifically define the conditions roughly determined by the configuration.

The control condition acquisition unit 105 may acquire control condition data from at least one of the operator, the manufacturing unit 110, and the control unit 117. The control condition acquisition unit 105 may supply the acquired control condition data to the second learning processing unit 108.

[1-7. Status acquisition unit]
The state acquisition unit 107 acquires state data indicating the state of the manufacturing unit 110. The state acquisition unit 107 may acquire state data from at least one of the operator and the manufacturing unit 110. Further, the state acquisition unit 107 may acquire the temperature, humidity, and the like of the environment in which the manufacturing unit 110 is installed as state data. The state acquisition unit 107 may supply the acquired state data to the second learning processing unit 108 and the second supply unit 115.

[1-8. 2nd learning processing department]
The second learning processing unit 108 executes the learning process of the second model 109 using the input learning data. The learning data used by the second learning processing unit 108 may include characteristic data, configuration data, and control condition data acquired from the characteristic acquisition unit 102, the configuration acquisition unit 101, and the control condition acquisition unit 105. This learning data may further include the state data acquired from the state acquisition unit 107. The second learning processing unit 108 may store and store the learning data used in the learning processing.

[1-9. 2nd model]
The second model 109 outputs recommended control condition data indicating recommended control conditions in response to input of target characteristic data and recommended configuration data. In addition to the target characteristic data and the recommended configuration data, the second model 109 may output the recommended control condition data according to the input of the state data. The second model 109 may be stored in a server outside the device 1. The second model 109 outputs the recommended control condition data to the recommended control condition acquisition unit 116, but may output the recommended control condition data to the outside of the device 1.

[1-10. Notification unit]
When the number of the plurality of learning data used in the learning process of the first model 104 is less than the number of the second reference number, the notification unit 100 notifies that fact. The notification unit 100 may notify that fact even when the number of the plurality of learning data used in the learning process of the second model 109 is less than the second reference number. The notification unit 100 may detect the number of learning data used in the learning process by referring to the learning data accumulated and stored in the first learning processing unit 103 and the second learning processing unit 108.

[1-11. Production department]
The manufacturing unit 110 is a manufacturing apparatus for the semiconductor device 5, and manufactures the semiconductor device 5 having the configuration shown in the recommended configuration data. The manufacturing unit 110 may manufacture the semiconductor device 5 using the recommended control condition data. The manufacturing unit 110 may have a film forming apparatus 2. The film forming apparatus 2 may be formed by a vapor phase growth method such as MOCVD, or may be formed by another method such as a molecular beam epitaxy method or a sputtering method. When the semiconductor device 5 is a light emitting device provided with a nitride semiconductor layer, the film forming apparatus 2 is a MOCVD apparatus from the viewpoints of crystallinity of the nitride semiconductor layer, controllability of carrier concentration, manufacturing stability, and the like. Is preferable. The manufacturing unit 110 may be controlled by the control unit 117.

[1-12. Target characteristic acquisition department]
The target characteristic acquisition unit 111 acquires target characteristic data indicating the characteristics of the target semiconductor device 5. The target characteristic acquisition unit 111 may acquire the target characteristic data from the operator. The target characteristic acquisition unit 111 may supply the acquired target characteristic data to the first supply unit 112 and the second supply unit 115.

[1-13. 1st supply unit]
The first supply unit 112 supplies the target characteristic data acquired by the target characteristic acquisition unit 111 to the first model 104.

[1-14. Recommended configuration acquisition unit]
The recommended configuration acquisition unit 113 acquires the recommended configuration data output by the first model 104 in response to the supply of the target characteristic data to the first model 104. The recommended configuration acquisition unit 113 may supply the acquired recommended configuration data to the control unit 117 and the second supply unit 115. The recommended configuration acquisition unit 113 may include a calculation unit 1131, a random configuration acquisition unit 1132, an interpolation configuration acquisition unit 1133, a first detection unit 1134, and a first extraction unit 1135.

The calculation unit 1131 is an example of the second calculation unit, and for each of the plurality of learning data used in the learning process (all the learning data used in the learning process as an example in the present embodiment), these plurality of learnings are performed. Among the data, the number of configuration approximation training data including the configuration data indicating the configuration in which the degree of approximation to the configuration indicated by the configuration data included in the training data is within the reference range is calculated. For example, the calculation unit 1131 constructs and approximates the number of other learning data in which the degree of closeness between the constituent data is within the reference range among the plurality of learning data for each of the plurality of learning data used in the learning process. It may be calculated as the number of training data.

The calculation unit 1131 refers to the learning data accumulated and stored in the first learning processing unit 103, and calculates the number of configuration approximation learning data for each learning data each time the first learning processing unit 103 acquires the learning data. You can fix it. The calculation unit 1131 may supply the number of configuration approximation learning data for each learning data to the interpolation configuration acquisition unit 1133.

Here, the degree of approximation between the configuration data (also referred to as the degree of approximation between the configurations) may be the sum of the differences for each configuration parameter between one configuration data and the other configuration data, or the configuration. It may be the mean, median or mode of the differences for each parameter. The reference range used by the calculation unit 1131 is an example of the second reference range, and may be any range.

The random configuration acquisition unit 1132 acquires recommended configuration data having random contents when the number of the plurality of learning data used in the learning process is smaller than the first reference number. The recommended configuration data having random contents may have contents irrelevant to the recommended configuration data output by the first model 104. The random configuration acquisition unit 1132 may randomly determine the recommended configuration data by itself without using the first model 104, or may acquire random recommended configuration data generated by another configuration. The recommended configuration data "generated in other configurations" may be, for example, appropriate data based on design of experiments. In design of experiments, recommended configuration data may be output using the concept of orthogonality, and the order of output may be random. The random configuration acquisition unit 1132 may acquire a determination result of whether or not the number of training data is less than the first reference number from the determination unit 118. The random configuration acquisition unit 1132 may supply the acquired recommended configuration data to the control unit 117. The first reference number may be any value.

The interpolation configuration acquisition unit 1133 acquires the configuration data obtained by interpolating any two of the reference number of learning data having the smallest number of configuration approximation learning data as the recommended configuration data. Interpolating two configuration data may mean generating new configuration data by taking a value between a value in one configuration data and a value in another configuration data for each configuration parameter. The interpolation configuration acquisition unit 1133 may generate recommended configuration data by interpolating any two configuration data with reference to the learning data stored and stored in the first learning processing unit 103. The interpolation configuration acquisition unit 1133 may acquire the recommended configuration data when the number of the plurality of learning data used in the learning process is larger than the first reference number and less than the second reference number. The interpolation configuration acquisition unit 1133 may acquire the magnitude determination results of the number of training data and each of the first reference number and the second reference number from the determination unit 118. The interpolation configuration acquisition unit 1133 may supply the acquired recommended configuration data to the control unit 117.

The reference number may be any value. The second reference number may be any value larger than the first reference number.

The first detection unit 1134 detects the distribution range of the constituent parameters in the plurality of learning data used in the learning process for the constituent parameters included in the constituent data of the learning data. The distribution range may be a range surrounded by the maximum value and the minimum value of the constituent parameters. When the configuration data includes a plurality of configuration parameters, the first detection unit 1134 may detect the distribution range of the values of the configuration parameters in the plurality of training data for each of the plurality of configuration parameters. .. The first detection unit 1134 may supply the distribution range for each detected configuration parameter to the first extraction unit 1135.

When one target characteristic data is supplied to the first model 104 and a plurality of recommended configuration data are acquired, the first extraction unit 1135 is detected by the first detection unit 1134 among these recommended configuration data. Extract recommended configuration data that contains only the recommended configuration parameters for the values that fall within the distribution range. For example, when each configuration data contains two configuration parameters, in one recommended configuration data, the first recommended configuration parameter is included in the distribution range of the first configuration parameter, and the second recommended configuration parameter is the second. When it is included in the distribution range of the constituent parameters of, the first extraction unit 1135 extracts the recommended constituent data of the one. Further, in the other recommended configuration data, when the first recommended configuration parameter is included in the distribution range of the first configuration parameter and the second recommended configuration parameter is not included in the distribution range of the second configuration parameter, The first extraction unit 1135 does not extract the other recommended configuration data.

In the first extraction unit 1135, when the value of any recommended configuration parameter is not included in the distribution range in each recommended configuration data, that is, the recommended configuration data including only the recommended configuration parameter of the value included in the distribution range is provided. If it does not exist, at least one recommended configuration data is extracted from the plurality of recommended configuration data in descending order of the number of recommended configuration parameters included in the distribution range. For example, when each configuration data contains a plurality of configuration parameters, the first recommended configuration data includes one recommended configuration parameter in the distribution range of the corresponding configuration parameter, and the second recommended configuration data includes each. When the recommended configuration parameter is not included in the distribution range of the corresponding configuration parameter, the first extraction unit 1135 may extract the first recommended configuration data. The first extraction unit 1135 may supply the extracted recommended configuration data to the control unit 117. In the present embodiment, as an example, the first extraction unit 1135 may extract a single recommended configuration data and supply it to the control unit 117.

[1-15. 2nd supply unit]
The second supply unit 115 supplies the target characteristic data acquired by the target characteristic acquisition unit 111 and the recommended configuration data acquired by the recommended configuration acquisition unit 113 to the second model 109. The second supply unit 115 may also supply the state data acquired by the state acquisition unit 107 to the second model 109.

[1-16. Recommended control condition acquisition unit]
The recommended control condition acquisition unit 116 acquires the recommended control condition data output by the second model 109 in response to the supply of the target characteristic data and the recommended configuration data to the second model 109. The recommended control condition acquisition unit 116 may further acquire the recommended control condition data output by the second model 109 in response to the supply of the state data to the second model 109. The recommended control condition acquisition unit 116 may supply the acquired recommended control condition data to the control unit 117.

[1-17. Control unit]
The control unit 117 operates the manufacturing unit 110 according to the configuration data and / or the control condition data to manufacture the semiconductor device 5 having the configuration indicated by the configuration data. For example, the control unit 117 may operate the manufacturing unit 110 according to the recommended control condition data and the recommended configuration data by supplying the recommended control condition data and the recommended configuration data to the manufacturing unit 110.

In addition, operating the manufacturing unit 110 according to the configuration data means that, for example, the number of layers, the film thickness of each layer, the composition of each layer, and / or the impurity concentration of each layer are realized. It may be to operate 110. Operating the manufacturing unit 110 according to the control condition data may mean operating the manufacturing unit 110 under the control conditions indicated by the control condition data.

[1-18. Judgment unit]
The determination unit 118 determines whether or not the difference between the characteristic data acquired by the characteristic acquisition unit 102 and the target characteristic data acquired by the target characteristic acquisition unit 111 is within the permissible range. In the present embodiment, as an example, the determination unit 118 may determine whether or not the difference is within the first permissible range and whether or not the difference is within the second permissible range narrower than the first permissible range.

When the difference is within the first permissible range, the determination unit 118 may disable the learning process of the first model 104 by the first learning processing unit 103, and in the present embodiment, as an example, the first learning A disable signal may be supplied to the processing unit 103. As a result, the manufacturing unit 110 and the first learning processing unit 103 supply the target characteristic data to the first model 104 and acquire the target characteristic data when the difference between the characteristic data and the target characteristic data is out of the first allowable range. The production of the semiconductor device 5 having the configuration shown by the recommended configuration data and the training process of the first model 104 using the training data including the characteristic data of the manufactured semiconductor device 5 may be repeated.

When the difference is within the second permissible range, the determination unit 118 may disable the learning process of the second model 109 by the second learning processing unit 108, and in the present embodiment, as an example, the second learning A disable signal may be supplied to the processing unit 108. As a result, the manufacturing unit 110 and the second learning processing unit 108 use the target characteristic data, the recommended configuration data, and the state data in the second model 109 when the difference between the characteristic data and the target characteristic data is out of the second allowable range. The manufacturing of the semiconductor device 5 having the configuration shown by the recommended control condition data supplied to and acquired by the second model 109 and the training process of the second model 109 using the training data including the characteristic data of the manufactured semiconductor device 5 are repeated. Good.

The first permissible range and the second permissible range of the difference may be arbitrarily set. When the characteristic data and the target characteristic data have a plurality of characteristic parameters, the determination unit 118 may determine whether or not the average value of the differences between the characteristic parameters is within the first permissible range and the second permissible range. ..

The determination unit 118 determines whether or not the difference between the characteristic data and the target characteristic data is within the permissible range when the number of the plurality of learning data used in the learning process is larger than the second reference number. You may go. In the present embodiment, as an example, the determination unit 118 refers to the number of learning data stored and stored in the first learning processing unit 103, and the number of learning data, the first reference number, and the second reference number, respectively. You may judge the magnitude of. The determination unit 118 may supply the determination result to the random configuration acquisition unit 1132 or the interpolation configuration acquisition unit 1133 of the recommended configuration acquisition unit 113.

According to the above device 1, the learning process of the first model 104 that outputs the recommended configuration data in response to the input of the target characteristic data is executed using the learning data including the configuration data and the characteristic data of the semiconductor device 5. Will be done. Therefore, the configuration of the semiconductor device 5 having the target characteristics can be obtained without the need for trial and error by a skilled worker.

Further, the training process of the second model 109, which outputs the recommended control condition data in response to the input of the state data, the target characteristic data, and the recommended configuration data, uses the training data including the characteristic data, the configuration data, and the control condition data. Therefore, it is possible to more reliably acquire the configuration of the semiconductor device 5 having the target characteristics.

Further, when the number of training data used in the learning process of the first model 104 and / or the second model 109 is less than the second reference number, a notification to that effect is given, so that the operator who received the notification It is possible to acquire the learning data according to the operation and execute the learning process of the first model 104 and the second model 109. Therefore, when the learning is insufficient, the learning data can be replenished and the learning process can be executed.

Further, the semiconductor device 5 having the configuration indicated by the recommended configuration data is manufactured, the configuration data and the characteristic data thereof are acquired, and the learning process of the first model 104 is performed using the training data including the configuration data and the characteristic data. Since it is executed, the learning process of the first model 104 and the manufacture of the semiconductor device 5 can be repeated. Therefore, the learning accuracy of the first model 104 can be improved, and the configuration of the semiconductor device 5 having the target characteristics can be acquired more reliably.

Further, when the number of the plurality of learning data used in the learning process is smaller than the first reference number, the recommended configuration data having random contents is acquired. Therefore, in the initial stage of learning, the configuration approximation learning data Recommended configuration data can be obtained without bias regardless of the number of. When the number of training data is larger than the first reference number and less than the second reference number, the configuration data obtained by interpolating any two configuration data having a small number of configuration approximation training data is obtained. Since it is acquired as the recommended configuration data, it is possible to acquire the recommended configuration data indicating the configuration in which the learning is not sufficiently performed at the stage where the learning is advanced.

Further, when the difference between the acquired characteristic data and the target characteristic data is out of the first permissible range, the manufacturing of the semiconductor device 5 and the learning process of the first model 104 are repeated, so that the semiconductor having the target characteristics The configuration of the device 5 can be acquired more reliably. Similarly, when the difference between the acquired characteristic data and the target characteristic data is out of the second permissible range, the manufacturing of the semiconductor device 5 and the learning process of the second model 109 are repeated, so that the target characteristic is obtained. The configuration of the semiconductor device 5 can be acquired more reliably.

Further, among the plurality of recommended configuration data acquired from the first model 104 in response to the supply of one target characteristic data, the recommended configuration data including only the recommended configuration parameters of the values included in the distribution range of the configuration parameters. Is extracted, so that the recommended configuration data in which all the recommended configuration parameters are included in the learning range is extracted. Therefore, it is possible to acquire the recommended configuration data within the learning range by excluding the recommended configuration data including the extraordinary configuration parameters outside the learning range.

If the value of any of the recommended configuration parameters is not included in the distribution range in each recommended configuration data, at least one recommended configuration data is extracted in descending order of the number of recommended configuration parameters included in the distribution range. , Even when each recommended configuration data is not included in the learning range, it is possible to acquire the recommended configuration data having a configuration close to the learning range.

Further, since the correlation between at least one characteristic parameter included in the characteristic data and at least one constituent parameter included in the configuration data is detected, an unknown correlation between the characteristic parameter and the configuration parameter can be found and more appropriate. The semiconductor device 5 having the configuration can be acquired.

[2. Film forming equipment]
FIG. 2 shows an example of the film forming apparatus 2. For example, the film forming apparatus 2 is an organic metal vapor phase growing apparatus, and Al, Ga, In, C, Si, Sn, B, Mg, Zn, Cd, Hg, N, As, Sb, O, S on the substrate 50. , Se, Te, Fe, Eu, Er, and one or more films containing at least one of Pr. As an example, the substrate 50 may be a compound semiconductor substrate containing at least one of Al, Ga and N (aluminum nitride as an example) or a sapphire substrate, having a diameter of 2 inches (50.8 mm) and a thickness of 0.5 mm. It may be disk-shaped. The film to be formed may be an n-type semiconductor layer, an undoped semiconductor layer, or a p-type semiconductor layer, or may be a non-uniform doped layer (modulated doped layer) such as δ-doped, or these. It may be any laminated body. The lattice constants may be different between the substrate 50 and the film formed on the substrate 50, and between the vertically adjacent films. The film forming apparatus 2 includes a film forming chamber 20, a susceptor 21, a substrate heater 22, a gas supply device 23, and a process pump 24.

[2-1. Film formation chamber]
The film forming chamber 20 is a closed reaction vessel for forming a film on the substrate 50 held inside. As an example, the film forming chamber 20 may have an annular side wall portion inside the housing. The housing may be made of SUS, and the side wall portion may be made of quartz.

[2-2. Suceptor]
The susceptor 21 supports the substrate 50. For example, a circular counterbore 210 for accommodating the substrate 50 may be provided on the upper surface of the susceptor 21. The diameter of the counterbore 210 may be larger than the diameter of the substrate 50, whereby a gap Δ1 may be provided between the inner wall of the counterbore 210 and the side peripheral surface of the substrate 50. As an example, the diameter of the substrate 50 may be 50.8 mm, and the diameter of the counterbore may be 51.6 mm. The interval Δ1 may be 0.4 mm over the entire circumferential direction of the substrate 50, but when the susceptor 21 rotates and the substrate 50 moves in the counterbore 210 due to centrifugal force as described later, the interval Δ1 is rotational. It may be 0.8 mm on the center side and 0 mm on the outer peripheral side. Further, the depth of the counterbore 210 may be different from the thickness of the substrate 50, whereby a step Δ2 may be provided between the upper surface of the substrate 50 in the counterbore 210 and the upper surface of the susceptor 21. For example, the depth of the counterbore 210 may be larger than the thickness of the substrate 50, and the upper surface of the substrate 50 in the counterbore 210 may be lower than the upper surface of the susceptor 21 by a step Δ2. As an example, the thickness of the substrate 50 may be 0.55 mm, the counterbore depth may be 0.7 mm, and the step Δ2 may be 0.15 mm. However, the upper surface of the substrate 50 may be flush with the upper surface of the susceptor 21, and the step Δ2 may be zero. The susceptor 21 is formed in a disk shape and may have a plurality of counterbore 210s (six as an example) in the circumferential direction, or may have a single counterbore 210 in the central portion. The susceptor 21 may be arranged concentrically with the film forming chamber 20, and may be rotatably provided by a shaft 211 hanging from a central portion. The rotation speed of the susceptor 21 may be adjustable. As an example, the susceptor 21 may be formed of carbon or silicon carbide, or may be formed of carbon coated with silicon carbide or the like.

[2-3. Board heater]
The substrate heater 22 heats the substrate 50. The substrate heater 22 may decompose the raw material compound in the vicinity of the substrate 50 (for example, the surface of the substrate 50) by heating the substrate 50. For example, the substrate heater 22 may be provided below the susceptor 21 and heat the substrate 50 via the susceptor 21. As an example, the substrate heater 22 may be maintained at 900 to 1300 ° C. and heated so that the upper surface of the substrate 50 is about 800 ° C. to 1200 ° C.

The substrate heater 22 may have a plurality of (three as an example) region-specific heaters (not shown) for heating different regions of the susceptor 21, and the power supplied to the region-specific heaters, and thus the heating amount by the region-specific heaters, may be It may be independently controllable. As an example, the region-specific heaters may be formed in an annular shape and arranged concentrically. When the film forming chamber 20 is translucent, the substrate heater 22 may heat the substrate 50 by another method such as heating the substrate 50 by radiation from the outside of the film forming chamber 20.

[2-4. Gas supply device]
The gas supply device 23 supplies gas into the film forming chamber 20. For example, the gas supply device 23 may supply the raw material gas as a raw material for film formation and the carrier gas for flowing the raw material gas into the film forming chamber 20 to the film forming chamber 20. The gas supply device 23 has a plurality of gas sources 230 and one or a plurality of raw material supply ports 232 for supplying gas from the gas source 230 into the film forming chamber 20.

[2-4 (1). Gas source]
The plurality of gas sources 230 supply the raw material gas and / or the carrier gas to the raw material supply port 232. Here, the raw materials contained in the raw material gas are Al, Ga, In, C, Si, Sn, B, Mg, Zn, Cd, Hg, N, As, Sb, O, S, Se, Te, Fe, Eu. , Er, and Pr, and may be a simple substance or a compound.

Si as a raw material may be supplied as monosilane. The monosilane may be supplied from the gas source 230 in a state of being diluted in advance with the carrier gas described later. The carrier gas is an inert gas having low reactivity with respect to the substrate 50 and the raw material, and may contain at least one _{of hydrogen (H 2} ) and nitrogen (N _{2) as an example.} Al, Ga, In, and Mg as raw materials may be supplied as compound organic metals, such organic metals as, for example, at least trimethylaluminum, trimethylgallium, triethylgallium, and cyclopentadienylmagnesium. There may be one. N as a raw material may be supplied as ammonia. When organometallics, monosilanes and ammonia are used as raw materials, they may be supplied from separate gas sources 230. The raw material gas may be a gas containing only the raw material, or may be a mixed gas further containing a bubbling gas for gasifying the liquid of the organic metal by bubbling. The bubbling gas is a gas that is inert to the substrate 50 and the raw material, and may be, for example, a gas of the same type as the carrier gas. As an example, when aluminum is contained as a raw material for film formation, the mixed gas may include a gas of trimethylaluminum as a raw material gas and a hydrogen (H ₂ ) gas as a bubbling gas. The raw material elements such as C, Sn, B, Zn, Cd, Hg, As, Sb, O, S, Se, Te, Fe, Eu, Er, and Pr are supplied in a conventionally known manner. May be done.

[2-4 (2). Raw material supply port]
Each raw material supply port 232 opens at a plurality of positions inside the film forming chamber 20. Each opening of the raw material supply port 232 may be dispersedly arranged so that the raw material gas is uniformly supplied to the upper surface of the substrate 50 rotating on the susceptor 21. In this figure, as an example of the raw material supply port, the raw material supply port 232 is provided on the ceiling surface of the film forming chamber 20. The distance between the opening of the raw material supply port 232 and the upper surface of the susceptor 21 is 5 mm to 20 mm, but the distance may be another size such as 10 cm or 20 cm, and the raw material supply port 232 is the film forming chamber 20. It may be provided on the side wall surface. The raw material supply port 232 may be a shower head as an example. The raw material supply port 232 may be maintained at 45 ° C. or the like by water cooling as an example, and the temperature of the gas supplied from the raw material supply port 232 into the film forming chamber 20 may be about 50 ° C. to 200 ° C. In the present embodiment, as an example, each raw material supply port 232 supplies gas separately into the film forming chamber 20, but the gas may be mixed in advance and then supplied into the film forming chamber 20.

[2-5. Process pump]
The process pump 24 sucks and discharges the gas in the film forming chamber 20. The process pump 24 may perform suction so that the pressure in the film forming chamber 20 becomes 50 mbar, for example, and the pressure in the film forming chamber 20 may be adjustable. An on-off valve 240 for adjusting the amount of gas discharged may be provided between the process pump 24 and the film forming chamber 20.

In the above film forming apparatus 2, for example, the substrate 50 is placed on the susceptor 21 and heated by the substrate heater 22, and the raw material gas containing an organic metal, silane and ammonia is mixed with the carrier gas in the film forming chamber while rotating the susceptor 21. By flowing into 20, the organic metal is decomposed and the metal element reacts with ammonia on the surface of the substrate 50 to form a film.

[2-6. Gas supply equipment]
FIG. 3 shows a gas supply facility 2300 that supplies a raw material gas of an organic metal together with a carrier gas to the film forming chamber 20. The gas supply device 23 may supply the raw material gas and the carrier gas of the organic metal to the film forming chamber 20 by the gas supply facility 2300.

The gas supply facility 2300 includes a gas source 230b for carrier gas, a main gas line 2301, a vent line 2302, a gas adjustment line 2303, a bubbling gas supply line 2304, and a gas source 230a for raw material gas. It has a raw material supply line 2305, a plurality of mass flow controllers 2306, and an on-off valve pair 2307, 2308.

The gas source 230b for the carrier gas is an example of the gas source 230, and supplies the carrier gas to the main gas line 2301. The main gas line 2301 communicates between the gas source 230b for the carrier gas and the film forming chamber 20. Almost all of the carrier gas supplied to the film forming chamber 20 flows through the main gas line 2301.

The venting line 2302 communicates between the gas source 230b for the carrier gas and the outside of the film forming apparatus 2. Carrier gas for exhaust is flowed through the venting line 2302.

The gas adjusting line 2303 communicates between the gas source 230b for the carrier gas and the main gas line 2301 and the venting line 2302. The adjusting carrier gas for adjusting the gas supply amount is flowed through the gas adjusting line 2303 so that the gas flow in the film forming chamber 20 does not change due to the gas supply from the raw material supply line 2305.

The bubbling gas supply line 2304 communicates between the gas source 230b for the carrier gas and the gas source 230a for the raw material gas of the organic metal. The carrier gas from the gas source 230b is flowed through the bubbling gas supply line 2304 as the bubbling gas at the gas source 230a.

The gas source 230a bubbles with the gas bubbler 231 using the bubbling gas (carrier gas in this embodiment) from the bubbling gas supply line 2304, and supplies the raw material gas and the mixed gas containing the carrier gas to the raw material supply line 2305. ..

The raw material supply line 2305 communicates between the gas bubbler 231 and the main gas line 2301 and the venting line 2302. A mixed gas from the gas source 230a flows through the raw material supply line 2305.

The plurality of mass flow controllers 2306 include mass flow controllers 2306 (1) to 2306 (6). The mass flow controller 2306 (1) is provided in the main gas line 2301 and adjusts the flow rate of the carrier gas flowing into the film forming chamber 20 via the main gas line 2301. The mass flow controller 2306 (2) is provided on the venting line 2302 and adjusts the flow rate of the carrier gas discharged to the outside of the film forming apparatus 2 via the venting line 2302. The mass flow controllers 2306 (3) and 2306 (5) are provided in the bubbling gas supply line 2304 and adjust the flow rate of the bubbling gas (carrier gas in this embodiment) flowing through the bubbling gas supply line 2304. The mass flow controllers 2306 (4) and 2306 (6) are provided on the gas adjusting line 2303 and adjust the flow rate of the adjusting carrier gas flowing through the gas adjusting line 2303.

The on-off valve pair 2307 has a pair of valves provided at a connection portion between the gas adjusting line 2303 and the main gas line 2301 and a connecting portion between the gas adjusting line 2303 and the venting line 2302, respectively. As a result, the on-off valve pair 2307 adjusts the injection amount of the adjusting carrier gas from the gas adjusting line 2303 to each of the main gas line 2301 and the venting line 2302. For example, the on-off valve pair 2307 is adjusted according to the change in the gas supply amount from the raw material supply line 2305 to the main gas line 2301 so that the gas supply amount from the main gas line 2301 to the film forming chamber 20 is kept constant. Carrier gas is injected into the main gas line 2301 and / or the venting line 2302.

The on-off valve pair 2308 has a pair of valves provided at a connection portion between the raw material supply line 2305 and the main gas line 2301 and a connection portion between the raw material supply line 2305 and the venting line 2302, respectively. As a result, the on-off valve pair 2308 adjusts the injection amount of the mixed gas (mixed gas of the raw material gas and the carrier gas in this embodiment) from the raw material supply line 2305 to the main gas line 2301 and the venting line 2302, respectively.

The gas source 230a, the gas adjusting line 2303, the bubbling gas supply line 2304, the raw material supply line 2305, and the on-off valve pairs 2307 and 2308 may be provided for each raw material of the organic metal. In the present embodiment, as an example, the gas supply facility 2300 includes a gas source 230a (Ga) for an organic metal of gallium, a gas adjustment line 2303 (Ga), a bubbling gas supply line 2304 (Ga), and a raw material supply line 2305 (Ga). ), And open / close valve pair 2307 (Ga), 2308 (Ga), gas source 230a (Al) for organic metal of aluminum, gas adjustment line 2303 (Al), bubbling gas supply line 2304 (Al), raw material. It may have a supply line 2305 (Al) and an on-off valve pair 2307 (Al), 2308 (Al).

In the above gas supply equipment 2300, in the preparation stage for film formation, the mass flow controller 2306 (1) flows a certain amount of carrier gas from the main gas line 2301 to the film formation chamber 20. Further, the mass flow controller 2306 (2) causes a constant amount of carrier gas to flow through the venting line 2302. The flow rates in the main gas line 2301 and the venting line 2302 may be the same or different.

When the raw material gas is supplied into the film forming chamber 20, the mass flow controller 2306 (3) uses the bubbling gas supply lines 2304 (in this embodiment, as an example, the bubbling gas supply lines 2304 (Ga)) and 2304 (as an example in the present embodiment). A carrier gas is passed through Al)). Each gas source 230a performs bubbling to supply a mixed gas containing a raw material gas and a carrier gas to the raw material supply line 2305. The on-off valve pair 2307 for raw materials that is not used for film formation (for example, the on-off valve pair 2307 (Al) for forming a GaN film) causes a mixed gas to flow through the venting line 2302 to be exhausted. Opening / closing valve pair 2307 (for example, opening / closing valve pair 2307 (Ga) when forming a GaN film, and opening / closing valve pair 2307 (Ga), 2307 when forming an AlGaN film) for raw materials used for film formation. In (Al)), the mixed gas is allowed to flow through the venting line 2302 to be exhausted until the flow rate of each mixed gas becomes the target flow rate and the bubbling becomes stable. Here, the target flow rate of the mixed gas may be about 1/100 of the flow rate of the carrier gas flowing through the main gas line 2301 by the mass flow controller 2306 (1). For example, the volume flow rate ratio of the carrier gas and the mixed gas may be 10 L / min: 100 to 1000 sccm.

The mass flow controller 2306 (4) flows the adjusting carrier gas through the gas adjusting line 2303 for each raw material, and the on-off valve pair 2307 of the gas adjusting line 2303 injects the adjusting carrier gas into the main gas line 2301. Here, the mass flow controller 2306 (4) and the on-off valve pair 2307 cooperate so that the injection amount of the adjusting carrier gas from each gas adjusting line 2303 to the main gas line 2301 becomes the target flow rate of the adjusting carrier gas. It may be adjusted using feedback control. The target flow rate of the adjusting carrier gas may be the maximum value of the target flow rate of the mixed gas of the corresponding raw material, or the flow rate obtained by adding a margin to the maximum value. As an example, when the maximum value of the target flow rate of the mixed gas from the gas source 230a (Ga) is 100 sccm, the target flow rate of the adjustment carrier gas by the gas adjustment line 2303 (Ga) may be 100 sccm. Similarly, when the maximum value of the target flow rate of the mixed gas from the gas source 230a (Al) is 100 sccm, the target flow rate of the adjusting carrier gas by the gas adjusting line 2303 (Al) may be 100 sccm. The adjusting carrier gas may be injected into the main gas line 2301 from a reference time (for example, 5 minutes) before the start of supply of the mixed gas of the raw materials.

When the flow rate of the mixed gas matches the target flow rate and the bubbling is stable, among the raw materials, the on-off valve pair 2308 provided in the raw material supply line 2305 for the raw material used for film formation is the raw material. The injection destination of the mixed gas from the supply line 2305 is changed from the venting line 2302 to the main gas line 2301. Further, among the raw materials, the on-off valve pair 2307 provided in the gas adjusting line 2303 for the raw material used for film formation maintains a constant gas supply amount from the main gas line 2301 to the film forming chamber 20. As described above, the injection destination of at least a part of the adjusting carrier gas is changed from the main gas line 2301 to the vent line 2302 in accordance with the change in the amount of gas supplied from the raw material supply line 2305 to the main gas line 2301. As a result, the uniformity of the amount of supplied gas over time when the supply of the raw material gas is started is improved, and the change in the gas flow in the film forming chamber 20 is prevented. A differential pressure gauge may be used to determine whether or not the flow rate of the mixed gas matches the target flow rate. For example, when the adjusting carrier gas is flowed through the gas adjusting line 2303 for the raw material used for film formation at a flow rate equal to the target flow rate of the mixed gas of the raw material, a differential pressure gauge is used in the main gas line 2301. By monitoring the differential pressure between the pressure of the gas and the pressure in the venting line 2302, it is possible to determine whether or not the flow rate of the mixed gas matches the target flow rate.

The injection destination from the raw material supply line 2305 and the gas adjustment line 2303 is changed instantaneously, but it may be gradually changed. When the supply of the raw material gas to the film forming chamber 20 is stopped, the on-off valve pair 2308 supplies the raw material so that the flow rate of the gas supplied from the main gas line 2301 to the film forming chamber 20 is kept constant. The injection destination of the mixed gas from the line 2305 is changed from the main gas line 2301 to the vent line 2302, and the on-off valve pair 2307 injects the adjustment carrier gas from the gas adjustment line 2303 from the vent line 2302 to the main gas. Change to line 2301.

[2-7. Control conditions for film forming apparatus 2]
The film forming apparatus 2 performs film forming according to the control conditions. The control condition is an input condition input to the device and is set by, for example, an operator. The control condition is not limited to the condition that can be directly controlled by the film forming apparatus 2, but may be a condition that can be indirectly controlled. As an example, the control conditions are the amount of raw material gas supplied into the film forming chamber 20, the partial pressure of the raw material gas, the amount of carrier gas supplied into the film forming chamber 20, and the type of gas existing in the film forming chamber 20. , Divided pressure ratio of each gas, temperature of gas bubbler 231, pressure of gas bubbler 231, vapor pressure of organic metal in gas bubbler 231, supply amount of bubbling gas to gas bubbler 231, rotation speed of susceptor 21, pressure in film forming chamber 20, substrate The temperature of 50, the temperature of the substrate heater 22, the power supplied to the substrate heater 22, the balance of the power supplied to the heaters for each region in the substrate heater 22, the gain in the feedback control of the amount of gas supplied to the film forming chamber 20, and the film forming. It may be at least one of the time and the pretreatment conditions performed before the film formation in the film forming apparatus 2. Of these, the amounts of the raw material gas and the carrier gas supplied into the film forming chamber 20 may be the flow rate of the gas supplied per unit time (for example, a volume flow rate or a mass flow rate), or the film forming starts and ends. It may be the total supply amount (volume or mass as an example) within the period up to. The pressure in the film forming chamber 20 may be the total pressure in the film forming chamber 20 or the partial pressure of each gas. As the temperature of the substrate 50, for example, the temperature of a thermocouple installed between the substrate heater 201 and the substrate 50 in order to control the power supplied to the substrate heater 201 can be used. The gain in the feedback control of the supply gas amount to the film forming chamber 20 is, for example, the gain in the feedback control when the injection amount of the adjustment carrier gas from the gas adjustment line 2303 to the main gas line 2301 is set as the target flow rate. As an example, at least one of P gain, I gain, and D gain in PID control may be used. The pretreatment performed before film formation in the film forming apparatus 2 includes the temperature of the annealing process for the substrate 50, the annealing time, and the type of gas filled in the film forming chamber 20 (for example, H ₂ , N ₂ and / Alternatively, it may be at least one of _{NH 3).} The control conditions as described above may be set over time, or may be set uniformly regardless of the passage of time (for example, set as a maximum value, a minimum value, an average value, etc.). May be good. Since the pressure in the film forming chamber 20 is affected by the operating condition of the process pump 24 and the like, it is state data indicating the state of the film forming apparatus 2, but the pressure in the film forming chamber 20 is higher than a certain threshold value. When the film forming operation is performed in a good case, it can be used as a control condition. Similarly, at least a part of the state data described later indicating the state of the film forming apparatus 2 may be used as a control condition.

Of the elements of the above-mentioned control conditions, the power supplied to the substrate heater 201 may indirectly control the substrate temperature. As a result of changes in the ease of movement of raw materials on the surface of the substrate 50 when the substrate temperature changes, the degree of crystallization in the film to be formed, the surface shape, and the characteristics of the film (for example, flatness, crystallinity, etc.) Changes.

Further, the pressure of the film forming chamber 20, the type of gas existing in the film forming chamber 20, and the amount of gas may directly or indirectly control the pressure and the quality of the atmosphere in the film forming chamber 20. .. When the pressure in the film forming chamber 20 and the quality of the atmosphere change, the state of the growth surface changes, and as a result, the characteristics of the film to be formed change.

[2-8. maintenance]
Various maintenance is performed on the film forming apparatus 2. For example, the maintenance may be performed by opening the film forming chamber 20 to the atmosphere, or may be performed while keeping the film forming chamber 20 closed. Maintenance can also be performed with nitrogen vented or purged. Maintenance may be performed on a regular basis (for example, every 3 days), depending on the characteristics of the film formed, or depending on the life of the parts used, failure, etc. Good.

[3. motion]
FIG. 4 shows the operation of the device 1. The device 1 manufactures the semiconductor device 5 having the target characteristics by the processing of steps S1 to S10.

In step S1, the device 1 executes the learning process of the first model 104 by the first learning processing unit 103 or the like.

In step S3, the apparatus 1 manufactures the semiconductor device 5 by the manufacturing unit 110 or the like using the recommended configuration data acquired from the first model 104. Of the control conditions of the manufacturing unit 110, the control conditions that are not affected by the recommended configuration data may be set to the specified standard contents, and the control conditions whose range is roughly determined by the recommended configuration data include the median value within the range. May be set to. When one target characteristic data is supplied to the first model 104 and a plurality of recommended configuration data are acquired, the recommended configuration data extracted by the first extraction unit 1135 (as an example, the value included in the distribution range). The semiconductor device 5 may be manufactured using (recommended configuration data) containing only the recommended configuration parameters.

In the present embodiment, the first model 104 is a regression model based on a linear model such as multiple regression, Lasso regression, or Ridge regression, or a neural network such as a recurrent type or a time delay type, as an example. Other machine learning algorithms may be included, including gradient boosting, logistic regression, and support vector machines (SVMs). For example, the first model 104 may include a node corresponding to each element of the training data in the input layer and a node corresponding to each element of the recommended control condition in the output layer. There may be one node or a plurality of nodes in the input layer for one element of the training data. An intermediate layer (hidden layer) containing one or more nodes may be interposed between the input layer and the output layer. The first learning processing unit 103 may execute the learning processing by adjusting the weight of the edge connecting the nodes and the bias value of the output node.

In step S4, the apparatus 1 determines whether or not the number of the plurality of learning data used in the learning process is larger than the second reference number by the determination unit 118. When the number of training data is equal to or less than the second reference number (step S4; No), the apparatus 1 shifts the process to step S1. As a result, until the number of training data used in the learning process reaches the second reference number, the manufacturing unit 110, the first learning processing unit 103, and the like manufacture the semiconductor device 5 having the configuration indicated by the recommended configuration data. , The learning process of the first model 104 using the learning data including the characteristic data of the manufactured semiconductor device 5 is repeated. When the number of training data is equal to or less than the second reference number (step S4; No), the notification unit 100 may notify the operator. When the number of training data is larger than the second reference number (step S4; Yes), the apparatus 1 shifts the process to step S5.

In step S5, in the apparatus 1, is the difference between the characteristic data acquired by the characteristic acquisition unit 102 from the manufactured semiconductor device 5 and the target characteristic data acquired by the target characteristic acquisition unit 111 within the first permissible range? Whether or not it is determined by the determination unit 118. If the difference is outside the first permissible range (step S5; No), the apparatus 1 shifts the process to step S1. As a result, the manufacturing unit 110, the first learning processing unit 103, and the like manufacture the semiconductor device 5 having the configuration indicated by the recommended configuration data, and the first model using the learning data including the characteristic data of the manufactured semiconductor device 5. The learning process of 104 is repeated. If the difference is within the first permissible range (step S5; Yes), the apparatus 1 shifts the process to step S7.

In step S7, the device 1 executes the learning process of the second model 109 by the second learning processing unit 108 or the like. The process of step S7 may be performed a plurality of times. As in the case of the first model 104, the second model 109 is, as an example, a regression model based on a linear model such as multiple regression, Lasso regression, or Ridge regression, or a neural network such as a recurrent type or a time delay type. Although it is a network, it may be another machine learning algorithm including random forest, gradient boosting, logistic regression, and support vector machine (SVM).

In step S9, the apparatus 1 manufactures the semiconductor device 5 by the manufacturing unit 110 or the like using the recommended configuration data acquired from the first model 104 and the recommended control condition data acquired from the second model 109.

In step S10, in the apparatus 1, is the difference between the characteristic data acquired by the characteristic acquisition unit 102 from the manufactured semiconductor device 5 and the target characteristic data acquired by the target characteristic acquisition unit 111 within the second permissible range? Whether or not it is determined by the determination unit 118. If the difference is outside the second permissible range (step S10; No), the apparatus 1 shifts the process to step S7. If the difference is within the second permissible range (step S10; Yes), the apparatus 1 ends the process. In this case, the apparatus 1 may repeatedly manufacture the semiconductor device 5 using the recommended configuration data and the recommended control condition data used in the manufacturing.

According to the above operation, after the accuracy of the recommended configuration data indicating the recommended configuration of the semiconductor device 5 having the target characteristics is improved by the first learning process, the recommended manufacturing conditions of the semiconductor device 5 having the target characteristics and the recommended configuration are set. The accuracy of the recommended control condition data shown is enhanced by the second learning process. Therefore, the recommended configuration data and the recommended control condition data can be obtained by surely converging the first learning process and the second learning process, so that the semiconductor device 5 having the target characteristics can be reliably manufactured.

[3-1. 1st learning process]
FIG. 5 shows the first learning process. The device 1 learns the first model 104 by the processing of steps S11 to S15.

In step S11, the configuration acquisition unit 101 acquires configuration data indicating the configuration of the semiconductor device 5. The semiconductor device 5 may be manufactured by the manufacturing unit 110. When the manufacturing unit 110 manufactures the semiconductor device 5 according to the recommended configuration data, the configuration acquisition unit 101 may acquire the recommended configuration data.

In step S13, the characteristic acquisition unit 102 acquires characteristic data indicating the characteristics of the semiconductor device 5. The semiconductor device 5 may be manufactured by the manufacturing unit 110. When a plurality of semiconductor devices 5 (also referred to as a plurality of semiconductor devices 5 in the same batch) are collectively manufactured in the manufacturing unit 110 in one manufacturing operation, that is, when a plurality of substrates 50 are held by the susceptor 21. The characteristic acquisition unit 102 may acquire the characteristics of all the semiconductor devices 5 among the plurality of semiconductor devices 5 in the same batch by 100% inspection, or the characteristics of some of the semiconductor devices 5 may be sampled and inspected. May be obtained by.
The processes of steps S11 and S13 do not have to be performed in this order.

In step S15, the first learning processing unit 103 executes the learning process of the first model 104 using the learning data including the acquired configuration data and characteristic data.

Here, in the learning process by the first learning processing unit 103, the second detection unit 1031 of the first learning processing unit 103 has at least one characteristic parameter included in the characteristic data and at least one configuration included in the configuration data. Correlation with parameters may be detected. In addition, the second detection unit 1031 may detect the correlation between at least two constituent parameters with respect to the value of any characteristic parameter. For example, when improving the value of the characteristic parameter for the optical output of the semiconductor device 5, if the thickness of one layer included in the semiconductor device 5 is increased and the thickness of the other layer is increased, these two thicknesses are used. Has a positive correlation. The second detection unit 1031 is based on the weight of the term calculated by regression analysis of the value of any characteristic parameter included in the characteristic data and the term having the value of at least two constituent parameters included in the constituent data. The correlation between at least two configuration parameters may be detected.

As an example, the second detection unit 1031 may perform regression analysis using the following Lasso regression equation (1). In the formula (1), y is the value of the characteristic parameter, and i is the identification number of the semiconductor device 5. k and l are integers and are numbers assigned to each configuration parameter. For example, when there are three constituent parameters, the film thickness and doping concentration of the first layer and the film thickness of the second layer, these constituent parameters include k = 1,2,3 and l = 1,2, respectively. It will be numbered 3. k and l are synonymous. x is a value of a configuration parameter, x _i ^k is the value of the k-th configuration parameters in the i-th semiconductor device 5. Equation (1) is a machine learning model that describes the characteristic parameter y by adding the quadratic term of the value x of each constituent parameter and the term represented by the product of two different constituent parameters. Here, w is a weight, _{w k} is the weight for term ^x k ^{x l} of the product of the _{weights, w kl} configuration parameters ^{x k} and configuration parameters ^{x l} of the configuration parameters of ^{x k.} When three constituent parameters are present as shown in the above example, w ₁₂ x _i ¹ x _i ² is the product of the film thickness of the first layer and the doping concentration for the i-th semiconductor device 5. The number of weights w may be determined according to the number of configuration parameters included in the configuration data. λ is the penalty parameter of the penalty term. In the Lasso regression analysis, the weight w of each term is determined so as to satisfy the equation (1).

Using the above regression equation (1), the second detection unit 1031 may detect the correlation between the configuration parameter x ^k and the configuration parameter x ^l _{based on the weight w kl.} The second detection unit 1031 may output a set of highly correlated configuration parameters x ^k and x ^l.

In the present embodiment, as an example, since the configuration data in the learning data includes a configuration parameter indicating the number of layers of the semiconductor device 5, the first learning processing unit 103 includes the number of layers in the learning target. The learning process of one model 104 may be executed. For example, the plurality of configuration data used in the learning process may include configuration parameters indicating a larger number of layers and / or a smaller number of layers than the semiconductor device 5 that can be actually manufactured.

According to the above operation, the first model 104 to which the recommended configuration data is output can be generated by inputting the target characteristic data.

Further, a correlation between the two constituent parameters is detected based on the weight of the term calculated by regression analysis between the value of one of the characteristic parameters and the term having the values of at least two constituent parameters. Therefore, it is possible to discover the correlation between the constituent parameters (correlation of the layer thickness as an example), which has not been considered to be correlated until now. Further, since a set of highly correlated configuration parameters is output, for example, the operator can obtain a more appropriate layer configuration by repeating the first learning process using the learning data in which the configuration parameters of the set are adjusted. Can be done.

When the first learning process according to steps S11 to S15 is performed a plurality of times, at least a part of the elements included in the configuration data and the characteristic data acquired in steps S11 and S13, that is, the configuration parameter and the type of the characteristic parameter. The learning process may be performed with the learning data of different elements in the learning process of step S15. _{For example, when N times of learning (where N is a natural number of 2 or more) is performed, the elements (c 1} ) in the constituent data and the characteristic data are included in the learning data from the first time to the M times (where M is a natural number smaller than N). _, c 2, ... c _n) contains, M + 1-th new element in the n-th configuration data in the learning data and characteristic data from (c n _{+ 1)} may be added. In this case, in the learning after the M + 1th time, the learning process may be performed using the learning data including the _{element (cn + 1).} Further, since the learning data before the Mth time _{does not originally include an element (cn + 1} ), an element having a random value (cn _{+ 1} ) may be added to each of these learning data. Accordingly, elements in the training data _{(c 1, c 2, ...} c n) for the relationship between can be effectively utilized learning results up to M-th. Further, the value of the element _{(c n + 1)} by a random value, it is possible to eliminate the influence of the value of the element _{(c n + 1)} of other elements _{(c 1, c 2, ...} c n) applied to.

[3-2. Second learning process]
FIG. 6 shows the second learning process. The device 1 learns the second model 109 by the processing of steps S21 to S29.

In steps S21 to S23, the configuration acquisition unit 101 and the characteristic acquisition unit 102 acquire the configuration data indicating the configuration of the semiconductor device 5 and the characteristic data indicating the characteristics of the semiconductor device 5 in the same manner as in steps S11 to S13 described above. To do.

In step S25, the control condition acquisition unit 31 acquires control condition data indicating the control conditions of the manufacturing unit 110 when the semiconductor device 5 is manufactured.

In step S27, the state acquisition unit 107 acquires state data indicating the state of the manufacturing unit 110 (the film forming apparatus 2 as an example in this embodiment). The state data is an actual measurement value of a controlled object when the film forming apparatus 2 operates under the control conditions indicated by the control condition data (for example, an actual measurement value in chronological order measured at a plurality of time points, or a measurement value at a certain time point. Actual measurement value) may be included. For example, the state data includes the temperature of the substrate 50, the temperature of the substrate heater 201, the power supply to the substrate heater 201, the balance of the power supply to the heaters for each region of the substrate heater 22, the step Δ2 between the upper surface of the substrate 50 and the upper surface of the susceptor, and the seat. The distance between the inner wall of the drill 210 and the side peripheral surface of the substrate 50 Δ1, the amount of warpage of the substrate 50, the film forming speed, the contents of the pretreatment performed before the film formation in the film forming apparatus 2, and the film forming apparatus 2. Details of processing applied to the substrate 50 before charging, temperature of gas bubbler 231, pressure of gas bubbler 231, vapor pressure of organic metal in gas bubbler 231, supply amount of bubbling gas to gas bubbler 231, pressure of film forming chamber 20, opening / closing valve It may include data on an opening degree of 240 and at least one of the rotational speeds of the susceptor 200. Of these, the temperature of the substrate heater 201 may be measured by a thermocouple. The temperature of the substrate 50 may be measured by a radiation thermometer arranged inside or outside the film forming chamber 20. Alternatively, the temperature of the substrate 50 may be measured and calculated based on the band edge absorption and transmission spectrum of the substrate 50. The amount of warpage of the substrate 50 may be the height difference between the side peripheral portion and the central portion of the substrate 50, and may be caused by heat treatment or the like on the substrate 50. The film forming rate is the film thickness formed per unit time, and may be measured by an optical interferometry method. The contents of the pretreatment performed before the film formation in the film forming apparatus 2 include, for example, the temperature of the annealing process applied to the substrate 50 as the pretreatment, the annealing time, and the gas filled in the film forming chamber 20. At least one of the types is sufficient. The processing content applied to the substrate 50 before being charged into the film forming apparatus 2 may be, for example, the content of the cleaning processing of the substrate 50. The pressure in the film forming chamber 20 may be measured by a vacuum gauge (for example, a pyrani gauge or a ballatron) arranged in the film forming chamber 20. The state data may include the total film forming time after maintenance of the film forming apparatus 2 (the elapsed time from the start of the so-called campaign) and the characteristics of the base surface (characteristics of the substrate 50 as an example). The state acquisition unit 107 may acquire state data related to the substrate temperature for each substrate 50, or may acquire status data for some of the substrates 50.

Further, the state data may include operation history data indicating the operation history of the film forming apparatus 2. The operation history data includes data on at least one of the number and contents of maintenance performed on the film forming apparatus 2 (for example, a certain part is replaced or washed), and the number of times at least one component of the film forming apparatus 2 is used. It may include at least one of data relating to, data relating to the weight of the raw material charged to the gas bubbler 231 during maintenance, data relating to the number of depositions of the film forming apparatus 2, and data relating to films formed in the past. The data relating to at least one of the number and contents of maintenance may be data indicating the history of maintenance. The data regarding the number of film formations may be the number of film formations after maintenance or the total number of film formations regardless of maintenance. The data related to the film formed in the past may be historical data showing the type and characteristics of the film formed in the past. These operation history data may be related to the state of adhesion of raw materials inside the film forming chamber 20, and thus the heat capacity of each member and the state of heat conduction.

The processes of steps S21 to S27 do not have to be performed in this order.
In step S29, the second learning processing unit 108 executes the learning process of the second model 109 using the learned data including the acquired configuration data, characteristic data, control condition data, and state data.

According to the above operation, it is possible to generate a second model in which the recommended control condition data is output by inputting the target characteristic data, the recommended configuration data, and the state data. Further, since the state data includes the operation history data, it is possible to obtain more accurate control conditions for forming the film having the target characteristics. Further, since the operation history data includes data relating to at least one of the number of maintenances and the contents, the second model 35 can learn the number of maintenances and the relationship between the contents and the characteristics. From the viewpoint of obtaining such an effect, it is preferable that the characteristic data is acquired from the film formed by the film forming apparatus 2 which has been maintained twice or more.

Similar to the first learning process, when the second learning process by the processes of steps S21 to S29 is performed a plurality of times, the configuration data, characteristic data, control condition data, and state data acquired in steps S21 to S27 are performed. The types of elements included in the above may be different at least at some times, and in the learning process of step S29, the learning process may be performed with the learning data of different elements.

[3-3. Manufacture of compound semiconductors]
FIG. 7 shows a manufacturing method of the semiconductor device 5. First, in step S31, the operator prepares the substrate 50. For example, the operator sets the substrate 50 in the film forming chamber 20 of the film forming apparatus 2. In step S33, the apparatus 1 forms at least one film to be included in the semiconductor device 5 on the substrate 50. As a result, the semiconductor device 5 in which one or more films are formed on the substrate 50 is manufactured. In the process of step S13, film formation is performed using at least one of the first model 104 and the second model 109. For example, in the process of step S13 in the manufacturing process of step S3 described above, film formation may be performed using the first model 104, and in the process of step S13 in the manufacturing process of step S9 described above, at least the second model 109 The film may be formed using.

[3-3-1. Film formation using the first model]
FIG. 8 shows a film forming method using the first model 104.

In step S41, the target characteristic acquisition unit 36 acquires target characteristic data indicating the target characteristics of the semiconductor device 5 to be manufactured, and in step S43, the first supply unit 112 supplies the target characteristic data to the first model 104. .. As a result, the recommended configuration data of the semiconductor device 5 is output from the first model 104.

In step S45, the recommended configuration acquisition unit 113 acquires the recommended configuration data.
When the number of the plurality of learning data used in the learning process is smaller than the first reference number, the random configuration acquisition unit 1132 of the recommended configuration acquisition unit 113 may acquire the recommended configuration data having random contents. The random configuration acquisition unit 1132 may acquire recommended configuration data having random contents even when the number of training data is the same as the first reference number.

When the number of the plurality of training data used in the learning process is larger than the first reference number and less than the second reference number, the interpolation configuration acquisition unit 1133 of the recommended configuration acquisition unit 113 is subjected to the calculation unit 1131. The configuration data obtained by interpolating any two configuration data of the reference number of training data having the smallest number of calculated configuration approximation training data may be acquired as the recommended configuration data. The interpolation configuration acquisition unit 1133 may acquire the configuration data obtained by interpolation as the recommended configuration data even when the number of training data is the same as the second reference number.

When the number of the plurality of training data used in the learning process is larger than the number of the second reference number, the recommended configuration acquisition unit 113 supplies the target characteristic data to the first model 104, and the first model The recommended configuration data output from 104 may be acquired. When one target characteristic data is supplied to the first model 104 and a plurality of recommended configuration data are acquired, they are included in the distribution range detected by the first detection unit 1134 among these recommended configuration data. Recommended configuration data containing only the recommended configuration parameters of values may be extracted by the first extraction unit 1135.

In step S47, the control unit 117 operates the manufacturing unit 110 according to the recommended configuration data. As a result, at least one film having the configuration shown in the recommended configuration data or a configuration similar thereto is formed. When the film is formed as described above, the control condition data and the state data at the time of film formation and the configuration data and the characteristic data of the manufactured semiconductor device 5 are input to the second model 109 as learning data. Then, the learning process of the second model 109 may be performed by performing the second learning process of FIG. 6 described above.

[3-3-2. Film formation using the second model]
FIG. 9 shows a film forming method using the second model 109.

In step S51, the target characteristic acquisition unit 36 and the recommended configuration acquisition unit 113 acquire the target characteristic data and the recommended configuration data for the semiconductor device 5 to be manufactured, and the state acquisition unit 107 acquires the state data indicating the state of the manufacturing unit 110. To do. The recommended configuration acquisition unit 113 may acquire recommended configuration data from the first model 104.

In step S53, the second supply unit 115 supplies the target characteristic data, the recommended configuration data, and the state data to the second model 109. As a result, the recommended control condition data regarding the control conditions of the manufacturing unit 110 is output from the second model 109.

In step S55, the recommended control condition acquisition unit 116 acquires the recommended control condition data output from the second model 109, and in step S57, the control unit 117 operates the control unit 117 under the control conditions indicated by the recommended control condition data. As a result, the semiconductor device 5 having the target characteristics or characteristics similar to the target characteristics is formed.

[4. Specific examples of semiconductor devices]
The manufactured semiconductor device 5 may be used, for example, in a light emitting device (for example, an LED that generates ultraviolet C wave light having a wavelength of 200 to 280 nm). Such a light emitting device is provided in, for example, a germicidal lamp or a DNA analysis device. In general, the layer configuration of a light emitting device is complicated, and the configuration parameters included in the configuration data are enormous. Therefore, it is often impractical to experimentally optimize all layers. Therefore, when the semiconductor device 5 is a light emitting device, it is preferable to apply the device 1 according to the present embodiment because optimization can be efficiently executed. Specifically, it is possible to obtain an optimum layer structure while significantly reducing the number of experiments. In addition, the correlation between non-adjacent layers, which was extremely difficult to predict from the findings so far, will be clarified, and the discovery of an epoch-making new layer structure can be expected.

FIG. 10 shows the layer structure of the semiconductor device 5. The semiconductor device 5 includes an aluminum nitride monocrystal layer 51 (thickness 500 nm as an example), an aluminum gallium nitride n- layer 52 (thickness 500 nm as an example), and nitridation on an aluminum nitride substrate 50 (thickness 500 μm as an example). Quantum well layer 53 and electron block layer (EBL) 54 having a barrier layer 530 (thickness 6 nm as an example) of aluminum gallium (Al _{0.75 GaN) and a well layer 531 (thickness 3 nm as an example) of aluminum gallium nitride Al 0.55 GaN.} It has a graded layer 55 (thickness 30 nm as an example) with a continuously reduced composition (thickness 15 nm as an example), and a gallium nitride p-layer 56 (thickness 10 nm as an example). The quantum well layer 53 may be laminated in multiple layers (five layers as an example). The film formed by the film forming method of FIG. 6 may be at least one of layers 51 to 56. Here, the graded layer 55 is a layer that functions as a hole generation layer, and is formed by reducing the aluminum content from 75% to 25% as the distance from the substrate 50 side increases. Further, the p-layer 56 functions as a contact layer. In this semiconductor device 5, since an aluminum nitride substrate is used as the substrate 50, unlike the case where a sapphire substrate is used, it is possible to align the lattice constant with the film to be formed and reduce the dislocation density.

[5. Modification example]
FIG. 11 shows the device 1A according to the modified example. The recommended configuration acquisition unit 113A of the apparatus 1A replaces / adds the calculation unit 1131, the random configuration acquisition unit 1132, the interpolation configuration acquisition unit 1133, the first detection unit 1134, and the first extraction unit 1135, and the calculation unit 1138 and the second unit. It has an extraction unit 1139.

The calculation unit 1138 is an example of the first calculation unit, and the recommended configuration acquisition unit 113 is a first model for each of a plurality of recommended configuration data acquired by supplying one target characteristic data to the first model 104. Of the plurality of training data used in the training process in 104, the approximate configuration learning data in which the degree of approximation to the configuration indicated by the recommended configuration data is within the reference range (that is, the configuration in which the degree of approximation is within the reference range) Calculate the number of learning data) having the indicated configuration data. The reference range used by the calculation unit 1138 is an example of the first reference range, and may be any range. The reference range used by the calculation unit 1138 and the reference range used by the calculation unit 1131 may be the same or different.

The second extraction unit 1139 extracts at least one recommended configuration data from the plurality of recommended configuration data based on the number of configuration approximation learning data calculated by the calculation unit 1138. For example, the second extraction unit 1139 supplies at least one of the recommended configuration data acquired by supplying one target characteristic data to the first model 104 in descending order of the number of configuration approximation training data. One recommended configuration data may be extracted.

In this modification, as an example, the recommended configuration acquisition unit 113A has a calculation unit 1131, a random configuration acquisition unit 1132, and an interpolation configuration acquisition unit 1133. Therefore, when the number of the plurality of learning data used in the learning process is smaller than the first reference number, the recommended configuration acquisition unit 113A acquires the recommended configuration data having random contents by the random configuration acquisition unit 1132. You can do it. Further, when the number of the plurality of learning data used in the learning process is larger than the first reference number and less than the second reference number, the recommended configuration acquisition unit 113A is calculated by the calculation unit 1131. The configuration data obtained by interpolating any two configuration data of the reference number of training data having the smallest number of approximate training data by the interpolation configuration acquisition unit 1133 may be acquired as the recommended configuration data.

According to the above device 1A, for each of the plurality of recommended configuration data acquired by supplying one target characteristic data to the first model 104, the number of approximation configuration learning data indicating a configuration in which the degree of approximation is within the reference range. Is calculated, and at least one recommended configuration data is extracted based on the calculation result. Therefore, since the recommended configuration data is extracted based on the number of the approximate configuration learning data, it is possible to acquire the recommended configuration data having more / less learning data of the approximate configuration.

In the above modification, the recommended configuration acquisition unit 113A has a calculation unit 1131, a random configuration acquisition unit 1132, and an interpolation configuration acquisition unit 1133, and the number of training data used in the learning process is the reference number (second reference). When the number is less than the number), it has been described as acquiring the recommended configuration data having random contents or the contents obtained by interpolating the two configuration data. However, when the recommended configuration acquisition unit 113A does not have the calculation unit 1131, the random configuration acquisition unit 1132, and the interpolation configuration acquisition unit 1133, the number of training data used in the learning process is the reference number (second reference number). If the number is smaller, at least one recommended configuration data is extracted in ascending order of the number of constitutive approximation training data, and if the number of training data is larger than the reference number, at least one is extracted in descending order of the number of constitutive approximation training data. Recommended configuration data may be extracted.

Further, the second extraction unit 1139 has characteristics from the semiconductor device 5 manufactured by the manufacturing unit 110 instead of switching the extraction method of the recommended configuration data according to the number of training data used in the training process in the first model 104. The extraction method may be switched depending on whether or not the difference between the characteristic data acquired by the acquisition unit 102 and the target characteristic data acquired by the target characteristic acquisition unit 111 is within the reference range. For example, the second extraction unit 1139 extracts at least one recommended configuration data in descending order of the number of configuration approximation training data when the difference is within the reference range, and when the difference is outside the reference range, the second extraction unit 1139 extracts at least one recommended configuration data. At least one recommended configuration data may be extracted in ascending order of the number of configuration approximation training data. Here, the determination unit 118 may determine whether or not the difference is within the reference range. Further, the reference range used for determining whether or not to switch the recommended configuration data extraction method is the learning processing of the first model 104 and the second model 109 by the first learning processing unit 103 and the second learning processing unit 108. It may be a range wider than the above-mentioned first permissible range and second permissible range used for determining whether or not to disable.

[6. Modification example]
FIG. 12 shows the device 1B according to another modified example. The device 1B has a third model 120 in place of the first model 104 and the second model 109, and has a third learning processing unit 121 in place of the first learning processing unit 103 and the second learning processing unit 108.

[6-1. 3rd model]
The third model 120 is a model in which the first model 104 and the second model 109 are integrated, and outputs at least one of the recommended configuration data and the recommended control condition data in response to the input of the target characteristic data. For example, the third model 120 may supply the recommended configuration data to the recommended configuration acquisition unit 113 in response to the input of the target characteristic data from the first supply unit 112. Further, the third model 120 may supply the recommended control condition data to the recommended control condition acquisition unit 116 in response to the input of the target characteristic data and the recommended configuration data from the second supply unit 115. Instead, the third model 120 generates recommended configuration data in response to input of target characteristic data from the first supply unit 112 and supplies the recommended configuration data to the recommended configuration acquisition unit 113, and also inputs the input target characteristics. The recommended configuration data corresponding to the data and the internally generated recommended configuration data may be generated and supplied to the recommended control condition acquisition unit 116. State data may be input to the third model 120 in addition to the target characteristic data and the recommended configuration data. In this modification, the recommended configuration acquisition unit 113 and / or the recommended control condition acquisition unit 116 may be an example of the acquisition unit.

In the third model 120, the learning process may be executed using a plurality of training data including the values of at least one configuration parameter as the configuration data, and a plurality of recommended configuration data are supplied according to the supply of one target characteristic data. May be output. Each of these plurality of recommended configuration data may include the value of at least one recommended configuration parameter indicating the configuration of the recommended semiconductor device 5.

[6-2. Third learning processing department]
The third learning processing unit 121 may execute the learning process of the third model 120 by using the learning data including the acquired configuration data, characteristic data, and control condition data. This learning data may further include the state data acquired from the state acquisition unit 107. The third learning processing unit 121 may have a second detection unit 1031 similar to the first learning processing unit 103.

The device 1B may execute the learning process of the third model 120 in the same manner as the second learning process described with reference to FIG. According to the above device 1B, the configuration of the semiconductor device 5 and the control conditions of the manufacturing unit 110 can be optimized in parallel.

In the above modification, the device 1B has been described as including the recommended configuration acquisition unit 113, but the recommended configuration acquisition unit 113A may be provided.

[7. Other variants]
In the above-described embodiment and modification, the devices 1 and 1A include the configuration acquisition unit 101, the characteristic acquisition unit 102, the first learning processing unit 103, the control condition acquisition unit 105, the state acquisition unit 107, and the first device. Although it has been described that the two learning processing units 108 and the second model 109 are provided, at least one of them may not be provided. In this case, the devices 1 and 1A include at least the target characteristic acquisition unit 111, the first supply unit 112, and the recommended configuration acquisition unit 113 (or the recommended configuration acquisition unit 113A), and the trained first model 104 The recommended configuration data may be obtained using. Further, the devices 1 and 1A include a target characteristic acquisition unit 111, a first supply unit 112, a recommended configuration acquisition unit 113 (or a recommended configuration acquisition unit 113A), a second supply unit 115, and a recommended control condition acquisition unit 116. Although it has been described that the control unit 117 and the determination unit 118 are provided, at least one of these may not be provided. In this case, the devices 1 and 1A may include at least the configuration acquisition unit 101, the characteristic acquisition unit 102, and the first learning processing unit 103, and perform the learning processing of the first model 104. Further, although the devices 1 and 1A have been described as including the second model 109 in addition to the first model 104, the second model 109 may not be provided. In this case, the devices 1 and 1A may not include the control condition acquisition unit 105, the state acquisition unit 107, the second learning processing unit 108, the second supply unit 115, the recommended control condition acquisition unit 116, and the like. Further, although the devices 1, 1A and 1B have been described as including the notification unit 100, the devices 1, 1A and 1B may not be provided.

Further, although the devices 1, 1A and 1B have been described as including one manufacturing unit 110, a plurality of devices 1, 1A and 1B may be provided. In this case, the learning data obtained from each manufacturing unit 110 may be supplied to the first learning processing unit 103, the second learning processing unit 108 and / or the third learning processing unit 121 to improve the learning efficiency, or one. The first model 104, the second model 109, and / or the third model 120, which have been trained based on the learning data obtained from the manufacturing unit 110, may be shared by the plurality of manufacturing units 110 to improve the manufacturing efficiency.

Further, although the film forming apparatus 2 has been described as the configuration of FIG. 2, another configuration may be used as the organic metal vapor phase growth apparatus. Further, the manufacturing unit 110 may have a processing device or another film forming apparatus in addition to / instead of the film forming apparatus 2. When the manufacturing unit 110 has a vapor deposition apparatus in addition to the film forming apparatus 2, the manufacturing unit 110 may form an electrode film by the vapor deposition apparatus. In this case, the first learning processing unit 103 and the second learning processing unit 108 may execute the learning process by including the configuration of the electrode film, the control conditions of the vapor deposition apparatus, and the like as the learning target.

Further, when the value of any of the recommended configuration parameters is not included in the distribution range in each of the plurality of recommended configuration data obtained by supplying one target characteristic data to the first model 104, the plurality of recommendations Although it has been described that the first extraction unit 1135 extracts at least one recommended configuration data from the configuration data in descending order of the number of recommended configuration parameters included in the distribution range, other processing may be performed. For example, the first supply unit 112 excludes a characteristic parameter having a lower priority from each characteristic parameter included in the target characteristic data, or relaxes the value of the characteristic parameter and supplies the characteristic parameter to the first model 104 to provide a recommended configuration. Data may be output. As a result, the recommended configuration of the semiconductor device 5 having the target characteristics can be obtained for the characteristics having high priority (for example, the optical output is 50 mW or more and the drive voltage is 9 V or less).

Further, although the recommended configuration acquisition units 113 and 113A have been described as having the calculation unit 1131, the random configuration acquisition unit 1132, and the interpolation configuration acquisition unit 1133, they may not be included. Further, although it has been described that the recommended configuration acquisition units 113 and 113A have the first detection unit 1134, the first extraction unit 1135, the calculation unit 1138, and the second extraction unit 1139, in addition to / instead of any of these, It may have other configurations. For example, the recommended configuration acquisition unit 113 is a layer having an earlier formation order among the plurality of layers included in the semiconductor device 5 from a plurality of recommended configuration data acquired by supplying one target characteristic data to the first model 104. It may have a third extraction unit that extracts some recommended configuration data in which the recommended configuration parameters are equal to each other. In this case, since the configuration can be determined from the layer having the earliest formation order, the configuration of the semiconductor device 5 having the target characteristics can be efficiently obtained.

In addition, various embodiments of the present invention may be described with reference to flowcharts and block diagrams, wherein the block serves (1) the stage of the process in which the operation is performed or (2) the role of performing the operation. It may represent a section of the device it has. Specific stages and sections are implemented by dedicated circuits, programmable circuits supplied with computer-readable instructions stored on a computer-readable medium, and / or processors supplied with computer-readable instructions stored on a computer-readable medium. You can. Dedicated circuits may include digital and / or analog hardware circuits, and may include integrated circuits (ICs) and / or discrete circuits. Programmable circuits are memory elements such as logical AND, logical OR, logical XOR, logical NAND, logical NOR, and other logical operations, flip-flops, registers, field programmable gate arrays (FPGA), programmable logic arrays (PLA), etc. May include reconfigurable hardware circuits, including, etc.

The computer readable medium may include any tangible device capable of storing instructions executed by the appropriate device, so that the computer readable medium having the instructions stored therein is specified in a flowchart or block diagram. It will be equipped with a product that contains instructions that can be executed to create means for performing the operation. Examples of computer-readable media may include electronic storage media, magnetic storage media, optical storage media, electromagnetic storage media, semiconductor storage media, and the like. More specific examples of computer-readable media include floppy (registered trademark) disks, diskettes, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), Electrically erasable programmable read-only memory (EEPROM), static random access memory (SRAM), compact disk read-only memory (CD-ROM), digital versatile disk (DVD), Blu-ray (RTM) disk, memory stick, integrated A circuit card or the like may be included.

Computer-readable instructions are assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state-setting data, or object-oriented programming such as Smalltalk, JAVA®, C ++, etc. Contains either source code or object code written in any combination of one or more programming languages, including languages and traditional procedural programming languages such as the "C" programming language or similar programming languages. Good.

Computer-readable instructions are used locally or on a processor or programmable circuit of a general purpose computer, special purpose computer, or other programmable data processing unit, or in a wide area network (WAN) such as the local area network (LAN), the Internet, etc. ) May be executed to create a means for performing the operation specified in the flowchart or block diagram. Examples of processors include computer processors, processing units, microprocessors, digital signal processors, controllers, microcontrollers and the like.

FIG. 13 shows an example of a computer 2200 in which a plurality of aspects of the present invention may be embodied in whole or in part. The program installed on the computer 2200 can cause the computer 2200 to function as an operation associated with the device according to an embodiment of the present invention or as one or more sections of the device, or the operation or the one or more. Sections can be run and / or the computer 2200 can be run a process according to an embodiment of the invention or a stage of such process. Such a program may be run by the CPU 2212 to cause the computer 2200 to perform certain operations associated with some or all of the blocks in the flowcharts and block diagrams described herein.

The computer 2200 according to this embodiment includes a CPU 2212, a RAM 2214, a graphic controller 2216, and a display device 2218, which are interconnected by a host controller 2210. The computer 2200 also includes input / output units such as a communication interface 2222, a hard disk drive 2224, a DVD-ROM drive 2226, and an IC card drive, which are connected to the host controller 2210 via the input / output controller 2220. There is. The computer also includes legacy input / output units such as the ROM 2230 and keyboard 2242, which are connected to the input / output controller 2220 via an input / output chip 2240.

The CPU 2212 operates according to the programs stored in the ROM 2230 and the RAM 2214, thereby controlling each unit. The graphic controller 2216 acquires the image data generated by the CPU 2212 in a frame buffer or the like provided in the RAM 2214 or itself so that the image data is displayed on the display device 2218.

The communication interface 2222 communicates with other electronic devices via the network. The hard disk drive 2224 stores programs and data used by the CPU 2212 in the computer 2200. The DVD-ROM drive 2226 reads the program or data from the DVD-ROM 2201 and provides the program or data to the hard disk drive 2224 via the RAM 2214. The IC card drive reads programs and data from the IC card and / or writes programs and data to the IC card.

The ROM 2230 contains a boot program or the like executed by the computer 2200 at the time of activation, and / or a program depending on the hardware of the computer 2200. The input / output chip 2240 may also connect various input / output units to the input / output controller 2220 via a parallel port, serial port, keyboard port, mouse port, and the like.

The program is provided by a computer-readable medium such as a DVD-ROM 2201 or an IC card. The program is read from a computer-readable medium, installed on a hard disk drive 2224, RAM 2214, or ROM 2230, which is also an example of a computer-readable medium, and executed by the CPU 2212. The information processing described in these programs is read by the computer 2200 and provides a link between the program and the various types of hardware resources described above. The device or method may be configured by implementing manipulation or processing of information in accordance with the use of computer 2200.

For example, when communication is executed between the computer 2200 and an external device, the CPU 2212 executes a communication program loaded in the RAM 2214, and performs communication processing on the communication interface 2222 based on the processing described in the communication program. You may order. Under the control of the CPU 2212, the communication interface 2222 reads and reads transmission data stored in a transmission buffer processing area provided in a recording medium such as a RAM 2214, a hard disk drive 2224, a DVD-ROM 2201, or an IC card. The data is transmitted to the network, or the received data received from the network is written to the reception buffer processing area or the like provided on the recording medium.

Further, the CPU 2212 causes the RAM 2214 to read all or necessary parts of a file or database stored in an external recording medium such as a hard disk drive 2224, a DVD-ROM drive 2226 (DVD-ROM2201), or an IC card. Various types of processing may be performed on the data on the RAM 2214. The CPU 2212 then writes back the processed data to an external recording medium.

Various types of information, such as various types of programs, data, tables, and databases, may be stored on the recording medium and processed. The CPU 2212 describes various types of operations, information processing, conditional judgment, conditional branching, unconditional branching, and information retrieval described in various parts of the present disclosure with respect to the data read from the RAM 2214. Various types of processing may be performed, including / replacement, etc., and the results are written back to RAM 2214. Further, the CPU 2212 may search for information in a file, a database, or the like in the recording medium. For example, when a plurality of entries each having an attribute value of the first attribute associated with the attribute value of the second attribute are stored in the recording medium, the CPU 2212 specifies the attribute value of the first attribute. Search for an entry that matches the condition from the plurality of entries, read the attribute value of the second attribute stored in the entry, and associate it with the first attribute that satisfies the predetermined condition. The attribute value of the second attribute obtained may be acquired.

The program or software module described above may be stored on or near a computer 2200 on a computer-readable medium. Also, a recording medium such as a hard disk or RAM provided within a dedicated communication network or a server system connected to the Internet can be used as a computer-readable medium, thereby providing the program to the computer 2200 over the network. To do.

Although the present invention has been described above using the embodiments, the technical scope of the present invention is not limited to the scope described in the above embodiments. It will be apparent to those skilled in the art that various changes or improvements can be made to the above embodiments. It is clear from the description of the claims that such modified or improved forms may also be included in the technical scope of the present invention.

The order of execution of operations, procedures, steps, steps, etc. in the devices, systems, programs, and methods shown in the claims, specification, and drawings is particularly "before" and "prior to". It should be noted that it can be realized in any order unless the output of the previous process is used in the subsequent process. Even if the scope of claims, the specification, and the operation flow in the drawings are explained using "first", "next", etc. for convenience, it means that it is essential to carry out in this order. It's not a thing.

1 device, 2 film forming device, 5 semiconductor device, 20 film forming chamber, 21 susceptor, 22 substrate heater, 23 gas supply device, 24 process pump, 50 substrate, 51 single crystal layer, 52 n- layer, 53 quantum well layer , 55 graded layer, 56 p-layer, 100 notification unit, 101 configuration acquisition unit, 102 characteristic acquisition unit, 103 first learning processing unit, 104 first model, 105 control condition acquisition unit, 107 state acquisition unit, 108th 2 Learning processing unit, 109 2nd model, 110 Manufacturing unit, 111 Target characteristic acquisition unit, 112 1st supply unit, 113 Recommended configuration acquisition unit, 115 2nd supply unit, 116 Recommended control condition acquisition unit, 117 Control unit, 118 Judgment unit, 200 susceptor, 201 substrate heater, 211 shaft, 230 gas source, 231 gas bubbler, 232 raw material supply port, 240 open / close valve, 530 barrier layer, 513 well layer, 1031 second detection unit, 1131 calculation unit, 1132 random configuration Acquisition unit, 1133 Interpolation configuration acquisition unit, 1134 1st detection unit, 1135 1st extraction unit, 1138 calculation unit, 1139 2nd extraction unit, 2200 computer, 2201 DVD-ROM, 2210 host controller, 2212 CPU, 2214 RAM, 2216 Graphic controller, 2218 display device, 2220 input / output controller, 2222 communication interface, 2224 hard disk drive, 2226 DVD-ROM drive, 2230 ROM, 2240 input / output chip, 2242 keyboard, 2300 gas supply equipment, 2301 main gas line, 2302 Vent line, 2303 gas adjustment line, 2304 bubbling gas supply line, 2305 raw material supply line, 2306 mass flow controller, 2307 open / close valve pair, 2308 open / close valve pair

Claims (35)

A configuration acquisition unit that acquires configuration data indicating the configuration of a semiconductor device,
A characteristic acquisition unit that acquires characteristic data indicating the characteristics of the semiconductor device, and
Using the acquired configuration data and learning data including the characteristic data, output the recommended configuration data indicating the recommended semiconductor device configuration according to the input of the target characteristic data indicating the characteristics of the target semiconductor device. The first learning processing unit that executes the learning processing of the first model
A device equipped with. A control condition acquisition unit that acquires control condition data indicating control conditions of the semiconductor device manufacturing apparatus, and a control condition acquisition unit.
Using the characteristic data, the configuration data, and the learning data including the control condition data, the recommended control condition data indicating the recommended control conditions according to the input of the target characteristic data and the recommended configuration data is output. The second learning processing unit that executes the learning processing of the two models, and
The device according to claim 1, further comprising. A configuration acquisition unit that acquires configuration data indicating the configuration of a semiconductor device,
A characteristic acquisition unit that acquires characteristic data indicating the characteristics of the semiconductor device, and
A control condition acquisition unit that acquires control condition data indicating control conditions of the semiconductor device manufacturing apparatus, and a control condition acquisition unit.
Using the acquired training data including the configuration data, the characteristic data, and the control condition data, the recommended semiconductor device configuration is determined according to the input of the target characteristic data indicating the characteristics of the target semiconductor device. A third learning processing unit that executes the training processing of the third model that outputs at least one of the recommended configuration data shown and the recommended control condition data indicating the recommended control conditions, and
A device equipped with. The target characteristic acquisition unit that acquires the target characteristic data and
A first supply unit that supplies the target characteristic data acquired by the target characteristic acquisition unit to the first model, and a first supply unit.
A recommended configuration acquisition unit that acquires the recommended configuration data output by the first model in response to supplying the target characteristic data to the first model.
The device according to claim 1, further comprising. Acquired by the target characteristic acquisition unit for the second model that outputs the recommended control condition data indicating the control conditions of the semiconductor device manufacturing apparatus, which is recommended according to the input of the target characteristic data and the recommended configuration data. A second supply unit that supplies the target characteristic data obtained and the recommended configuration data acquired by the recommended configuration acquisition unit.
A recommended control condition acquisition unit that acquires the recommended control condition data output by the second model in response to supplying the target characteristic data and the recommended configuration data to the second model.
4. The apparatus according to claim 4. The target characteristic acquisition unit that acquires the target characteristic data and
A first supply unit that supplies the target characteristic data acquired by the target characteristic acquisition unit to the third model, and a first supply unit.
An acquisition unit that acquires at least one of the recommended configuration data output by the third model in response to the supply of the target characteristic data to the third model and the recommended control condition data.
3. The apparatus according to claim 3. A target characteristic acquisition unit that acquires target characteristic data indicating the characteristics of the target semiconductor device,
The first supply that supplies the target characteristic data acquired by the target characteristic acquisition unit to the first model that outputs the recommended configuration data indicating the configuration of the semiconductor device recommended in response to the input of the target characteristic data. Department and
A recommended configuration acquisition unit that acquires the recommended configuration data output by the first model in response to supplying the target characteristic data to the first model.
A device equipped with. Acquired by the target characteristic acquisition unit for the second model that outputs the recommended control condition data indicating the control conditions of the semiconductor device manufacturing apparatus, which is recommended according to the input of the target characteristic data and the recommended configuration data. A second supply unit that supplies the target characteristic data obtained and the recommended configuration data acquired by the recommended configuration acquisition unit.
A recommended control condition acquisition unit that acquires the recommended control condition data output by the second model in response to supplying the target characteristic data and the recommended configuration data to the second model.
7. The apparatus according to claim 7. A target characteristic acquisition unit that acquires target characteristic data indicating the characteristics of the target semiconductor device,
With respect to the third model, which outputs at least one of the recommended configuration data indicating the recommended semiconductor device configuration and the recommended control condition data indicating the recommended control conditions in response to the input of the target characteristic data. The first supply unit that supplies the target characteristic data acquired by the target characteristic acquisition unit, and
An acquisition unit that acquires at least one of the recommended configuration data output by the third model in response to the supply of the target characteristic data to the third model and the recommended control condition data.
A device equipped with. In the first model, the training process is executed using a plurality of training data including the values of at least one configuration parameter indicating the configuration of the semiconductor device, and in response to the supply of the target characteristic data. A plurality of the recommended configuration data including the values of at least one recommended configuration parameter indicating the recommended semiconductor device configuration are output.
The recommended configuration acquisition unit is
For each of the at least one configuration parameter, a first detection unit that detects a distribution range of the values of the configuration parameter in the plurality of training data, and a first detection unit.
A first extraction unit that extracts recommended configuration data including only the recommended configuration parameters of values included in the distribution range among the plurality of recommended configuration data, and a first extraction unit.
The device according to any one of claims 4, 5, 7 and 8. In the third model, the training process is executed using a plurality of training data including the values of at least one configuration parameter indicating the configuration of the semiconductor device, and the third model is supplied with one target characteristic data. A plurality of the recommended configuration data including the values of at least one recommended configuration parameter indicating the recommended semiconductor device configuration are output.
The acquisition unit
For each of the at least one configuration parameter, a first detection unit that detects a distribution range of the values of the configuration parameter in the plurality of training data, and a first detection unit.
A first extraction unit that extracts recommended configuration data including only the recommended configuration parameters of values included in the distribution range among the plurality of recommended configuration data, and a first extraction unit.
The device according to claim 6 or 9. When the value of any of the recommended configuration parameters is not included in the distribution range in each recommended configuration data, the first extraction unit has a recommended configuration included in the distribution range from the plurality of recommended configuration data. The apparatus according to claim 10, wherein at least one recommended configuration data is extracted in descending order of the number of parameters. When the value of any of the recommended configuration parameters is not included in the distribution range in each recommended configuration data, the first extraction unit has a recommended configuration included in the distribution range from the plurality of recommended configuration data. The device according to claim 11, wherein at least one recommended configuration data is extracted in descending order of the number of parameters. In the first model, the training process is executed using a plurality of training data including the values of at least one configuration parameter indicating the configuration of the semiconductor device, and in response to the supply of the target characteristic data. A plurality of the recommended configuration data including the values of at least one recommended configuration parameter indicating the recommended semiconductor device configuration are output.
The recommended configuration acquisition unit is
For each of the plurality of recommended configuration data, among the plurality of training data used in the learning process, the configuration approximation learning data indicating that the degree of approximation to the configuration indicated by the recommended configuration data is within the first reference range. The first calculation unit that calculates the number of
A second extraction unit that extracts at least one recommended configuration data from the plurality of recommended configuration data based on the calculation result by the first calculation unit.
4. The device according to any one of claims 4, 5, 7, 8 and 12. In the third model, the training process is executed using a plurality of training data including the values of at least one configuration parameter indicating the configuration of the semiconductor device, and the third model is supplied with one target characteristic data. A plurality of the recommended configuration data including the values of at least one recommended configuration parameter indicating the recommended semiconductor device configuration are output.
The acquisition unit
For each of the plurality of recommended configuration data, among the plurality of training data used in the learning process, the configuration approximation learning data indicating that the degree of approximation to the configuration indicated by the recommended configuration data is within the first reference range. The first calculation unit that calculates the number of
A second extraction unit that extracts at least one recommended configuration data from the plurality of recommended configuration data based on the calculation result by the first calculation unit.
The device according to any one of claims 6, 9, 11 and 13. A manufacturing unit that manufactures the semiconductor device having the configuration shown in the recommended configuration data using the recommended control condition data, and a manufacturing unit.
A configuration acquisition unit that acquires configuration data indicating the configuration of the semiconductor device manufactured by the manufacturing unit, and a configuration acquisition unit.
A characteristic acquisition unit that acquires characteristic data indicating the characteristics of the semiconductor device manufactured by the manufacturing unit, and a characteristic acquisition unit.
A first learning processing unit that executes learning processing of the first model using the configuration data and learning data including the characteristic data, and
The device according to claim 5 or 8, further comprising. A manufacturing unit that manufactures the semiconductor device having the configuration shown in the recommended configuration data using the recommended control condition data, and a manufacturing unit.
A configuration acquisition unit that acquires configuration data indicating the configuration of the semiconductor device manufactured by the manufacturing unit, and a configuration acquisition unit.
A characteristic acquisition unit that acquires characteristic data indicating the characteristics of the semiconductor device manufactured by the manufacturing unit, and a characteristic acquisition unit.
A third learning processing unit that executes the learning processing of the third model using the configuration data and the learning data including the characteristic data, and
The device according to claim 6 or 9. In the first model, a learning process is executed using a plurality of learning data, and the learning process is executed.
The recommended configuration acquisition unit is
For each of the plurality of training data used in the learning process, among the plurality of learning data, the degree of approximation to the configuration indicated by the configuration data included in the learning data indicates a configuration within the second reference range. A second calculation unit that calculates the number of configuration approximation learning data including the configuration data, and
When the number of the plurality of learning data used in the learning process is less than the first reference number, a random configuration acquisition unit that acquires the recommended configuration data having random contents, and a random configuration acquisition unit.
When the number of the plurality of training data used in the learning process is larger than the first reference number and less than the second reference number, the number calculated by the second calculation unit is the smallest reference number. An interpolation configuration acquisition unit that acquires configuration data obtained by interpolating any two of the training data as the recommended configuration data, and an interpolation configuration acquisition unit.
The device according to claim 16. In the third model, the learning process is executed using a plurality of learning data, and the learning process is executed.
The acquisition unit
For each of the plurality of training data used in the learning process, among the plurality of learning data, the degree of approximation to the configuration indicated by the configuration data included in the learning data indicates a configuration within the second reference range. A second calculation unit that calculates the number of configuration approximation learning data including the configuration data, and
When the number of the plurality of learning data used in the learning process is smaller than the first reference number, the random configuration acquisition unit that acquires the recommended configuration data and the recommended control condition data having random contents, and the random configuration acquisition unit.
When the number of the plurality of training data used in the learning process is larger than the first reference number and less than the second reference number, the number calculated by the second calculation unit is the smallest reference number. An interpolation acquisition unit that acquires configuration data obtained by interpolating any two of the training data as the recommended configuration data, and an interpolation acquisition unit.
The device according to claim 17. The manufacturing unit and the first learning processing unit have the semiconductor having the configuration shown in the recommended configuration data until the number of the plurality of learning data used in the learning process reaches at least the second reference number. The apparatus according to claim 18, wherein the manufacture of the device and the learning process of the first model using the learning data including the characteristic data of the manufactured semiconductor device are repeated. When the number of the plurality of learning data used in the learning process is larger than the second reference number, the difference between the characteristic data acquired by the characteristic acquisition unit and the target characteristic data is within an allowable range. Further equipped with a judgment unit for determining whether or not
The manufacturing unit and the first learning processing unit are configured as shown by the recommended configuration data acquired by supplying the target characteristic data to the first model when the difference is out of the permissible range. The apparatus according to claim 18 or 20, wherein the manufacture of the semiconductor device and the training process of the first model using the training data including the characteristic data of the manufactured semiconductor device are repeated. The manufacturing unit and the third learning processing unit have the semiconductor having the configuration shown in the recommended configuration data until the number of the plurality of learning data used in the learning process reaches at least the second reference number. The apparatus according to claim 19, wherein the manufacture of the device and the learning process of the third model using the learning data including the characteristic data of the manufactured semiconductor device are repeated. When the number of the plurality of learning data used in the learning process is larger than the second reference number, the difference between the characteristic data acquired by the characteristic acquisition unit and the target characteristic data is within an allowable range. Further equipped with a judgment unit for determining whether or not
The manufacturing unit and the third learning processing unit are configured as shown by the recommended configuration data acquired by supplying the target characteristic data to the third model when the difference is out of the permissible range. 19. The apparatus according to claim 19 or 22, wherein the manufacturing of the semiconductor device and the learning process of the third model using the learning data including the characteristic data of the manufactured semiconductor device are repeated. The first learning processing unit has a second detection unit that detects a correlation between at least one characteristic parameter included in the characteristic data and at least one configuration parameter included in the configuration data, claims 1 and 2. , 16 and 18. The second detection unit calculates the weight of the term by regression analysis of the value of any characteristic parameter included in the characteristic data and the term having the value of at least two constituent parameters included in the constituent data. 24. The apparatus of claim 24, which detects the correlation between the at least two configuration parameters based on. The configuration of the semiconductor device is any one of claims 1 to 25, which is at least one of the number of layers contained in the semiconductor device, the film thickness of each layer, the composition of each layer, the impurity concentration of each layer, and the material of the substrate. The device according to item 1. The device according to any one of claims 1 to 26, wherein the semiconductor device is a light emitting device. The configuration acquisition stage for acquiring configuration data indicating the configuration of a semiconductor device, and
A characteristic acquisition stage for acquiring characteristic data indicating the characteristics of the semiconductor device, and
Using the acquired configuration data and learning data including the characteristic data, output the recommended configuration data indicating the recommended semiconductor device configuration according to the input of the target characteristic data indicating the characteristics of the target semiconductor device. The first learning process stage that executes the learning process of the first model to be performed, and
How to prepare. The configuration acquisition stage for acquiring configuration data indicating the configuration of a semiconductor device, and
A characteristic acquisition stage for acquiring characteristic data indicating the characteristics of the semiconductor device, and
A control condition acquisition stage for acquiring control condition data indicating the control conditions of the semiconductor device manufacturing apparatus, and
Using the acquired training data including the configuration data, the characteristic data, and the control condition data, the recommended semiconductor device configuration is determined according to the input of the target characteristic data indicating the characteristics of the target semiconductor device. A third learning process stage that executes the training process of the third model that outputs at least one of the recommended configuration data shown and the recommended control condition data indicating the recommended control conditions, and
How to prepare. The target characteristic acquisition stage for acquiring target characteristic data indicating the characteristics of the target semiconductor device, and
The first supply that supplies the target characteristic data acquired in the target characteristic acquisition step to the first model that outputs the recommended configuration data indicating the configuration of the semiconductor device recommended in response to the input of the target characteristic data. Stages and
A recommended configuration acquisition stage for acquiring the recommended configuration data output by the first model in response to supplying the target characteristic data to the first model, and
How to prepare. The target characteristic acquisition stage for acquiring target characteristic data indicating the characteristics of the target semiconductor device, and
With respect to the third model, which outputs at least one of the recommended configuration data indicating the recommended semiconductor device configuration and the recommended control condition data indicating the recommended control conditions in response to the input of the target characteristic data. The first supply stage for supplying the target characteristic data acquired by the target characteristic acquisition stage, and
An acquisition step of acquiring at least one of the recommended configuration data output by the third model and the recommended control condition data in response to the supply of the target characteristic data to the third model.
How to prepare. Computer,
A configuration acquisition unit that acquires configuration data indicating the configuration of a semiconductor device,
A characteristic acquisition unit that acquires characteristic data indicating the characteristics of the semiconductor device, and
Using the acquired configuration data and learning data including the characteristic data, output the recommended configuration data indicating the recommended semiconductor device configuration according to the input of the target characteristic data indicating the characteristics of the target semiconductor device. A program that functions as a first learning processing unit that executes the learning processing of the first model. Computer,
A configuration acquisition unit that acquires configuration data indicating the configuration of a semiconductor device,
A characteristic acquisition unit that acquires characteristic data indicating the characteristics of the semiconductor device, and
A control condition acquisition unit that acquires control condition data indicating control conditions of the semiconductor device manufacturing apparatus, and a control condition acquisition unit.
Using the acquired training data including the configuration data, the characteristic data, and the control condition data, the recommended semiconductor device configuration is determined according to the input of the target characteristic data indicating the characteristics of the target semiconductor device. A program that functions as a third learning processing unit that executes the learning processing of the third model that outputs at least one of the recommended configuration data shown and the recommended control condition data indicating the recommended control conditions. Computer,
A target characteristic acquisition unit that acquires target characteristic data indicating the characteristics of the target semiconductor device,
The first supply that supplies the target characteristic data acquired by the target characteristic acquisition unit to the first model that outputs the recommended configuration data indicating the configuration of the semiconductor device recommended in response to the input of the target characteristic data. Department and
A program that functions as a recommended configuration acquisition unit that acquires the recommended configuration data output by the first model in response to supplying the target characteristic data to the first model. Computer,
A target characteristic acquisition unit that acquires target characteristic data indicating the characteristics of the target semiconductor device,
With respect to the third model, which outputs at least one of the recommended configuration data indicating the recommended semiconductor device configuration and the recommended control condition data indicating the recommended control conditions in response to the input of the target characteristic data. The first supply unit that supplies the target characteristic data acquired by the target characteristic acquisition unit, and
A program that functions as an acquisition unit that acquires at least one of the recommended configuration data output by the third model and the recommended control condition data in response to the supply of the target characteristic data to the third model.

Images