Lifetime prediction device for semiconductor light emitting devices

Abstract

To provide an apparatus for predicting the lifetime of a semiconductor light emitting element capable of predicting the lifetime of the semiconductor light emitting element and a manufacturing system for the semiconductor light emitting element.SOLUTION: It has a data acquisition unit 21 for acquiring data on at least one of the composition parameters that define the composition of the layers of a semiconductor light emitting element 100 or the physical property parameters that define the physical properties of the layers, and the lifetime data from outside the device, linking the acquired data of at least one of the compositional or physical property parameters to the lifetime data to create the learning date 31; a model creation unit 22 for using the learning data 31 as teacher data to create a learned model 32 by machine learning with composition and physical property parameters as explanatory variables and lifetime as the objective variable; a lifetime prediction unit 23 for predicting the lifetime corresponding to any prediction source data 33 using the learned model 32 and makes it into the prediction data 34; and a prediction results presentation unit 24 for presenting the prediction data 34.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present invention relates to a semiconductor light emitting device. Place Regarding. [Background technology]

[0002] Group III nitride semiconductors made of compounds of aluminum (Al), gallium (Ga), indium (In) and nitrogen (N) are used as materials for ultraviolet light emitting devices. Among them, group III nitride semiconductors made of AlGaN with a high Al composition are used for ultraviolet light emitting devices and deep ultraviolet light emitting devices (for example, see Patent Document 1).

[0003] Prior art document information related to the invention of this application includes Patent Document 2. Patent Document 2 proposes a prediction method using machine learning to efficiently predict the properties of a film formed by a film forming apparatus. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 6001756 [Patent Document 2] Patent No. 6959191 Summary of the Invention [Problem to be solved by the invention]

[0005] Nitride semiconductor light-emitting devices require chipping and packaging after film formation, and it takes a long time to produce a product. For this reason, there is a problem that, for example, when reviewing the design at the time of film formation, it takes a very long time to repeat prototyping and evaluation. Therefore, there is a demand for predicting the lifetime of nitride semiconductor light-emitting devices before prototyping. However, nitride semiconductor light-emitting devices require control of a large number of parameters, and it was unclear what parameters should be used to predict the lifetime.

[0006] Therefore, the present invention provides a semiconductor light emitting element lifetime prediction device capable of predicting the lifetime of the semiconductor light emitting element. Place The purpose is to provide. [Means for solving the problem]

[0007] A lifetime prediction device for a semiconductor light-emitting element according to one embodiment of the present invention is an apparatus for predicting a lifetime of a semiconductor light-emitting element, and includes: a data acquisition unit that acquires, from outside the device, data corresponding to at least one of composition parameters, which are parameters that define a composition of a layer constituting the semiconductor light-emitting element, or physical parameters, which are parameters that define a physical property of a layer constituting the semiconductor light-emitting element, and lifetime data of the semiconductor light-emitting element, and creates learning data by linking the acquired data corresponding to at least one of the composition parameters or the physical parameters with the lifetime data; a model creation unit that uses the learning data as teacher data, performs machine learning on a correlation between the explanatory variables and the objective variable, with at least one of the composition parameters or the physical parameters as an explanatory variable and the lifetime as an objective variable, and creates a trained model representing the correlation; a lifetime prediction unit that uses the trained model to predict the lifetime corresponding to prediction source data, which is data of any explanatory variable, and generates predicted data; and a prediction result presentation unit that presents the predicted data.

[0008] In addition, a manufacturing system for a semiconductor light-emitting element according to one embodiment of the present invention includes a film-forming apparatus for a semiconductor light-emitting element, a management terminal for managing the film-forming apparatus, and a life prediction device for the semiconductor light-emitting element, and the data acquisition unit acquires data to be used for the learning data from the film-forming apparatus or the management terminal. Effect of the Invention

[0009] According to the present invention, there is provided a semiconductor light emitting element lifetime prediction device capable of predicting the lifetime of a semiconductor light emitting element. Place We can provide it. [Brief description of the drawings]

[0010] [Figure 1] 1 is a schematic diagram illustrating a configuration of a nitride semiconductor light-emitting device. [Diagram 2] 1 is a schematic configuration diagram of a lifetime prediction device for a nitride semiconductor light emitting element according to an embodiment of the present invention. [Figure 3A] FIG. 11 is a diagram illustrating an example of learning data. [Figure 3B] FIG. 11 is a diagram illustrating an example of learning data. [Figure 4] FIG. 11 is a graph showing the change in the film thickness of the p-type semiconductor layer while changing the film formation temperature of the p-type semiconductor layer while maintaining the film thickness of the p-type semiconductor layer constant, where (a) shows the change in the film thickness of the p-type semiconductor layer, and (b) shows the change in the residual light output rate after 1000 hours of current application. [Diagram 5] FIG. 1 is a diagram illustrating a model creation process. [Figure 6] FIG. 13 is a diagram illustrating a life prediction process. [Figure 7] 4 is a diagram illustrating a control flow of a lifetime prediction method for a nitride semiconductor light emitting device according to an embodiment of the present invention. FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] [Embodiment Mode] Hereinafter, an embodiment of the present invention will be described with reference to the accompanying drawings.

[0012] (Nitride semiconductor light emitting element 100) First, a nitride semiconductor light-emitting element 100 (hereinafter, simply referred to as "light-emitting element 100") whose lifetime is to be predicted in this embodiment will be described. Fig. 1 is a schematic diagram showing the configuration of the nitride semiconductor light-emitting element 100. Note that in Fig. 1, the dimensional ratio of each layer in the stacking direction of the light-emitting element 100 does not necessarily match the actual one, and each layer may have a multi-layer structure.

[0013] The light emitting element 100 is a light emitting diode (LED) and emits light with a wavelength in the ultraviolet region in this embodiment. The light emitting element 100 is, for example, a deep ultraviolet LED that emits ultraviolet light with a central wavelength of 200 nm or more and 365 nm or less, and is used, for example, for sterilizing water or air.

[0014] 1, the light emitting device 100 includes a buffer layer 102, an n-type semiconductor layer 103, an active layer 104, an electron blocking layer 105, and a p-type semiconductor layer 106, which are arranged in this order on a substrate 101. A p-type electrode 107 and a p-side pad electrode 108 are arranged in this order on the p-type semiconductor layer 106, and an n-type electrode 109 and an n-side pad electrode 110 are arranged in this order on the n-type semiconductor layer 103. A passivation film (protective film) 111 is arranged on the side surfaces of the p-type electrode 107 and the n-type electrode 109, and on the surface of the light emitting device 100 between the p-type electrode 107 and the n-type electrode 109. The active layer 104 has a multi-quantum well structure (Multi Quantum Well: MQW) formed by alternately stacking barrier layers 104a and well layers 104b.

[0015] Each of the layers 102 to 106 on the substrate 101 can be formed using a well-known epitaxial growth method such as Metal Organic Chemical Vapor Deposition (MOCVD), Molecular Beam Epitaxy (MBE), or Hydride Vapor Phase Epitaxy (HVPE).

[0016] The semiconductor constituting the light emitting device 100 is, for example, Al. x Ga y In 1-x-y A binary to quaternary III-N nitride semiconductor represented by 0≦x≦1, 0≦y≦1, 0≦x+y≦1 can be used. In deep ultraviolet LEDs, Al containing no indium is used. z Ga 1-zIn many cases, N-based (0≦z≦1) is used. In addition, a part of the group III elements of the semiconductors constituting the light emitting device 100 may be replaced with boron (B), thallium (Tl), etc. In addition, a part of the nitrogen may be replaced with phosphorus (P), arsenic (As), antimony (Sb), bismuth (Bi), etc. In this embodiment, each of the layers 102 to 106 is made of Al. z Ga 1-z It is composed of N (0≦z≦1).

[0017] It should be noted that the structure in FIG. 1 is merely an example, and the specific structure of the light-emitting element 100 is not limited to that shown in the figure, and can be modified as appropriate.

[0018] (Lifetime prediction device for nitride semiconductor light emitting devices 1) FIG. 2 is a schematic diagram of a lifetime prediction device 1 for a nitride semiconductor light-emitting element (hereinafter, simply referred to as "lifetime prediction device 1"). FIG. 2 shows a manufacturing device 10 (film formation device) for a nitride semiconductor light-emitting element and a management terminal 11 together with the lifetime prediction device 1. The manufacturing device 10 for a nitride semiconductor light-emitting element is a device for forming layers 102 to 106 of a light-emitting element 100 by using a well-known epitaxial growth method such as metal organic chemical vapor deposition, molecular beam epitaxy, or halide vapor phase epitaxy. The management terminal 11 is a terminal device for managing the manufacturing device 10 for a nitride semiconductor light-emitting element, and is composed of, for example, a personal computer. The management terminal 11 can be omitted, and the life prediction device 1 may be equipped with a function as the management terminal 11.

[0019] 2, the life prediction device 1 includes a control unit 2, a storage unit 3, a display unit 4, and an input device 5. The life prediction device 1 is configured from a computing device such as a personal computer or a server device.

[0020] The control unit 2 has a learning data acquisition unit 21, a model creation unit 22, a life prediction unit 23, and a prediction result presentation unit 24. Details of each unit will be described later. The control unit 2 is realized by appropriately combining a computing element, a memory, an interface, a storage device, etc. The storage unit 3 is realized by a predetermined storage area of ​​a memory or a storage device, and stores data etc. used for various controls by the control unit 2 described later. The input device 5 is composed of, for example, a keyboard, a mouse, etc. The display unit 4 is composed of, for example, a liquid crystal display, etc.

[0021] (Learning data acquisition unit 21) The learning data acquisition unit 21 performs a process of acquiring various data used for learning (machine learning) from the outside and storing it in the storage unit 3 as learning data 31. The various data may be acquired directly from the nitride semiconductor light-emitting device manufacturing apparatus 10 via wired or wireless communication, or may be acquired from the management terminal 11 via wired or wireless communication. The various data may be input from the input device 5, or may be input using a medium such as a USB memory. In this way, the method of acquiring the learning data 31 is not particularly limited.

[0022] (Learning data 31 and parameters used in machine learning) Here, an example of the learning data 31 used in machine learning will be described. FIGS. 3A and 3B are diagrams showing an example of the learning data 31. As shown in FIGS. 3A and 3B, the learning data 31 stores data corresponding to the composition parameters, physical property parameters, and manufacturing condition parameters of each of the layers 102 to 106, and data on the lifetime, which are linked to each other. The composition parameters are parameters that define the composition of each of the layers 102 to 106 constituting the light-emitting device 100. The manufacturing condition parameters are parameters that define the manufacturing conditions of the light-emitting device 100.

[0023] The specific parameters of the composition parameters, physical property parameters, and manufacturing condition parameters are not limited to those shown in the drawings, and other parameters may be used. The learning data 31 may include parameters other than the composition parameters, physical property parameters, and manufacturing condition parameters, and may include, for example, equipment state parameters representing the state of the manufacturing equipment 10 of the nitride semiconductor light-emitting device. Examples of the equipment state parameters include the height of the pile on the tray and the number of pockets, the temperature and flow rate of the chiller and cooling water, the number of film formations, and furnace dimensions. The learning data 31 may also include, in addition to the above, substrate parameters representing the state of the substrate 101, electrode parameters representing the states of the electrodes 107 to 110, and the like. The manufacturing condition parameters are not essential and can be omitted. The learning data 31 may also include only one of the composition parameters and the physical property parameters.

[0024] Although details will be described later, in this embodiment, the relationship between each parameter of the composition parameter, the physical property parameter, and the manufacturing condition parameter and the life span is machine-learned. Therefore, it is desirable to select parameters that have a large effect on the life span as the composition parameter and the physical property parameter used in the machine learning. When the manufacturing condition parameter is used in the machine learning, it is desirable to select the parameter that is changed when adjusting the value of the composition parameter or the physical property parameter used in the machine learning.

[0025] More specifically, in the buffer layer 102, since the defect density, which is a physical property parameter, is considered to be related to the lifetime, it is desirable to use it for machine learning when the data has a degree of variation that cannot be considered constant. The mix value described in the physical parameters in FIG. 3A is the half-width (arcsec) of the X-ray rocking curve obtained by ω-scanning of X-ray diffraction for the (10-12) plane (Mixed plane) of the crystal, and is a representative index indicating the crystal quality of each layer in a nitride semiconductor light-emitting device. When the manufacturing condition parameters of the buffer layer 102 are used for machine learning, it is desirable to use the film formation temperature (heater temperature or substrate temperature) and the TMA (trimethylaluminum) flow rate as the manufacturing condition parameters.

[0026] In the n-type semiconductor layer 103, it is preferable to use the film resistance as a physical property parameter. Since the film resistance of the n-type semiconductor layer 103 has a large effect on the life span, it is preferable to use it for machine learning when the data has a variation that cannot be considered constant. In addition, in the n-type semiconductor layer 103, similarly to the buffer layer 102, it is considered that the defect density, which is a physical property parameter, is related to the life span, it is preferable to use it for machine learning when the data has a variation that cannot be considered constant. In addition, when the manufacturing condition parameters of the n-type semiconductor layer 103 are used for machine learning, it is preferable to use the film formation temperature (heater temperature or substrate temperature), the TMA flow rate, the TMG (trimethylgallium) flow rate, and the TMSi (tetramethylsilane) flow rate as the manufacturing condition parameters.

[0027] In the barrier layer 104a of the active layer 104, similarly to the buffer layer 102 and the n-type semiconductor layer 103 described above, the defect density, which is a physical property parameter, is thought to be related to the lifetime, so when there is variation in the data to a degree that cannot be considered constant, it is desirable to use it for machine learning.

[0028] It is preferable to use defect density as a physical property parameter for the well layer 104b of the active layer 104. Since the defect density of the well layer 104b has a large effect on the lifetime, it is preferable to use it for machine learning when the data has a variation that cannot be considered constant.

[0029] In the electron block layer 105, it is preferable to use the doping concentration as the composition parameter, and the film thickness and defect density as the physical property parameters. Since a doping concentration of 0 means that no doping is performed, it can be said that the doping concentration also includes information on the presence or absence of doping. Since the doping concentration, film thickness, and defect density parameters of the electron block layer 105 have a large effect on the life span, it is preferable to use them for machine learning when the data has a variation that cannot be considered constant. When the manufacturing condition parameters of the electron block layer 105 are used for machine learning, it is preferable to use the film formation time, Cp2Mg (biscyclopentadienyl magnesium) flow rate, and TMSi flow rate as the manufacturing condition parameters.

[0030] In the p-type semiconductor layer 106, it is preferable to use the doping concentration as the composition parameter, and the film thickness and defect density as the physical property parameters. Since the parameters of the doping concentration, film thickness, and defect density of the p-type semiconductor layer 106 have a large effect on the life span, it is preferable to use them for machine learning when the data has a degree of variation that cannot be considered constant. Furthermore, in the p-type semiconductor layer 106, since the growth rate, which is a manufacturing condition parameter, affects the life span, it is preferable to use them for machine learning when the data has a degree of variation that cannot be considered constant. As other manufacturing condition parameters of the p-type semiconductor layer 106, it is preferable to use the film formation temperature (heater temperature or substrate temperature), TMA flow rate, TMG flow rate, Cp2Mg flow rate, and NH3 flow rate as manufacturing condition parameters.

[0031] In this specification, "lifetime" refers to the time elapsed from the start of use until the light output of the light-emitting element 100 drops to a predetermined value (light output determined as the end of life). Here, the remaining light output rate obtained by dividing the remaining light output after a current of 350 mA is passed for 1000 hours by the initial light output is used as a parameter representing the lifespan. In addition, the lifespan also varies depending on various conditions in the chipping and packaging processes, but here, the chipping and packaging are performed under the same conditions, and are not used in machine learning. However, it is also possible to perform machine learning including the conditions during chipping and packaging.

[0032] (Regarding combined use of manufacturing condition parameters) Furthermore, the inventors have found through their research that even when the composition parameters and physical property parameters are constant, the life may change if the manufacturing condition parameters change. This will be specifically described below.

[0033] While maintaining the thickness of the p-type semiconductor layer 106 constant, the deposition temperature of the p-type semiconductor layer 106 was changed. As described above, the thickness of the p-type semiconductor layer 106 has a large effect on the lifespan, and changing the thickness of the p-type semiconductor layer 106 changes the lifespan. The change in the thickness of the p-type semiconductor layer 106 at this time is shown in FIG. 4(a). As shown in FIG. 4(a), the thickness of the p-type semiconductor layer 106 becomes almost constant, and the thickness of the p-type semiconductor layer 106 is almost constant regardless of the deposition temperature. Here, measurements were performed using samples at two locations, the outer edge (relatively close to the outer edge) and the center of the wafer. The change in the residual light output rate after 1000 hours of current application at this time is shown in FIG. 4(b). As shown in FIG. 4(b), in the sample at the outer edge of the wafer, the residual light output rate increases and the lifespan becomes longer as the deposition temperature of the p-type semiconductor layer 106 increases, and in the sample at the center of the wafer, the residual light output rate decreases and the lifespan becomes shorter as the deposition temperature of the p-type semiconductor layer 106 increases. Thus, it is found that the lifetime changes depending on the deposition temperature of p-type semiconductor layer 106, even though the film thickness, which is a physical parameter of p-type semiconductor layer 106, is approximately constant.

[0034] Thus, there are cases where the lifetime cannot be predicted with sufficient accuracy only by the physical property parameters and the composition parameters. In this case, in addition to at least one of the physical property parameters and the composition parameters, it is necessary to predict the lifetime by further considering the manufacturing condition parameters that are changed when adjusting the value of the composition parameter or the physical property parameter. From the result of FIG. 4, when at least one of the composition parameter or the physical property parameters of a predetermined layer (the p-type semiconductor layer 106 in the example of FIG. 4) of the nitride semiconductor light-emitting device is used for learning, it is possible to predict the lifetime of the nitride semiconductor light-emitting device 100 with high accuracy by using the deposition temperature of the predetermined layer (the p-type semiconductor layer 106 in the example of FIG. 4) as a manufacturing condition parameter for learning. The active layer 104 (the barrier layer 104a and the well layer 104b) and the electron block layer 105 are generally thin, having a thickness of 100 nm or less, and it is difficult to predict the lifetime only by the composition parameters and the physical property parameters, so it is desirable to include the deposition temperature in the explanatory variables. In addition, it is more preferable to use the substrate temperature as the deposition temperature. The substrate temperature can be calculated by measuring infrared rays generated from the substrate surface with a pyrometer, for example.

[0035] From the results of Figures 4(a) and (b), it can be said that it is desirable to use position parameters that represent the position on the wafer for learning. As position parameters, distance from the center, coordinates from a preset reference position, or numbers assigned to multiple areas that have been previously divided can be used.

[0036] (Model Creation Part 22) The model creation unit 22 performs a process of creating a trained model 32 using the training data 31. The process performed by the model creation unit 22 corresponds to a model creation step of the present invention. In this embodiment, the model creation unit 22 creates the trained model 32 by learning (machine learning) a correlation between at least one of the composition parameters or physical property parameters of the layers 102 to 106 constituting the light-emitting element 100, the manufacturing condition parameters changed when adjusting the values ​​of the composition parameters or physical property parameters, and the lifetime of the light-emitting element 100. In this embodiment, the model creation unit 22 is configured to create the trained model 32 by using both the composition parameters and the physical property parameters for machine learning.

[0037] The model creation unit 22 may be configured to perform machine learning of correlations between the composition parameters, physical property parameters, and manufacturing condition parameters and the lifetime of the nitride semiconductor light-emitting device to create the trained model 32. This enables prediction of the lifetime taking into account the manufacturing conditions in addition to the composition and physical properties, and can further improve the accuracy of the lifetime prediction.

[0038] The model creation unit 22 includes a learning algorithm that uses preset composition parameters, physical property parameters, and manufacturing condition parameters of each layer 102-106 as explanatory variables and a lifetime as an objective variable, and learns by itself the correlation between each parameter of the explanatory variables and the objective variable by machine learning. The learning algorithm is not particularly limited, and a known learning algorithm called, for example, deep forest or deep neural network can be used. As described above, a position parameter indicating a position on the wafer may be further used as an explanatory variable.

[0039] As shown in Fig. 5, learning data 31 is input to the model creation unit 22. Using the input learning data 31, the model creation unit 22 repeatedly executes learning based on a data set of parameters used as explanatory variables and a lifespan that is a target variable, and automatically interprets the correlation between the two to create a learned model 32. The model creation unit 22 stores the created learned model 32 in the storage unit 3. The model creation unit 22 may be configured to allow the user to appropriately select parameters (composition parameters, physical property parameters, manufacturing condition parameters) to be used as explanatory variables.

[0040] (Lifespan Prediction Unit 23) The life prediction unit 23 performs a process of predicting a life using the trained model 32. The process performed by the life prediction unit 23 corresponds to a life prediction step of the present invention. As shown in FIG. 6, the trained model 32 created by the model creation unit 22 and prediction source data 33 are input to the life prediction unit 23. The prediction source data 33 is the value of each parameter used as an explanatory variable, and is input by, for example, the input device 5. The life prediction unit 23 predicts a life corresponding to the prediction source data 33 by applying the value of each parameter of the prediction source data 33 to the trained model 32. The predicted life is stored in the storage unit 3 as prediction data 34.

[0041] In addition, in the case where the trained model 32 is created using a position parameter representing a position on the wafer as an explanatory variable, the prediction source data 33 may not include the position parameter, and the life prediction unit 23 may use the prediction source data 33 and data of a plurality of positions on the wafer that have been set in advance (e.g., the center of the wafer, the outer edge, etc.) to calculate the life at each of the plurality of positions on the wafer that have been set in advance. In this case, the life prediction unit 23 may calculate and output an index value for life evaluation on the entire wafer, for example, by calculating an average value or a median value of the predicted values ​​of the life at the plurality of positions on the wafer. Furthermore, the life prediction unit 23 may output a ratio of predicted values ​​of the life at the plurality of positions on the wafer that is equal to or greater than a predetermined threshold value (e.g., a ratio of the optical output remaining rate after 1000 hours of power supply being 70% or more) as an index value for life evaluation on the entire wafer. Furthermore, the life prediction unit 23 may output a maximum value and a minimum value of predicted values ​​of the life at the plurality of positions on the wafer as an index value for life evaluation on the entire wafer.

[0042] (Prediction result presentation unit 24) The prediction result presentation unit 24 performs a process of presenting the prediction data 34 predicted by the life prediction unit 23. The prediction result presentation unit 24 presents the prediction data 34, for example, by displaying the prediction data 34 on the display 4. The format of the presentation is not particularly limited, and the data may be presented in an appropriate format such as numerical values ​​or graphs. However, the present invention is not limited to this, and the prediction data 34 may be presented, for example, by outputting the prediction data 34 to an external device.

[0043] (Method for predicting lifetime of nitride semiconductor light emitting device) Fig. 7 is a flow diagram showing a control flow of the lifetime prediction method for a nitride semiconductor light emitting device according to this embodiment. The control flow of Fig. 7 is executed when predicting a lifetime. Prior to the control flow of Fig. 7, the learning data acquisition unit 21 acquires learning data 31, and the acquired learning data 31 is stored in the storage unit 3 as needed.

[0044] As shown in FIG. 7, in the lifetime prediction method for a nitride semiconductor light-emitting element according to this embodiment, a model creation step in step S1, a lifetime prediction step in step S2, and a prediction result presentation step in step S3 are performed in this order.

[0045] In the model creation process of step S1, first, in step S11, the model creation unit 22 determines whether the learning data 31 has been updated since the previous time the trained model 32 was created. The determination in step S11 can be made by comparing the creation date and time of the trained model 32 with the update date and time of the training data 31. Note that in step S11, if the trained model 32 has not been created (first time), it is determined that the learning data 31 has been updated (Yes). If the determination in step S11 is No (N), the process proceeds to the life prediction process of step S2. If the determination in step S11 is Yes (Y), in step S12, the model creation unit 22 machine-learns the correlation between each parameter of the explanatory variable set in advance and the lifespan as the objective variable based on the updated training data 31, and creates the trained model 32.

[0046] In this embodiment, the parameters used as explanatory variables in the machine learning in step S11 include at least one of the composition parameters and physical property parameters of the layers 102 to 106 constituting the light emitting device 100. More preferably, the following parameters are used: Defect density in each layer 102-106 Film resistance of the n-type semiconductor layer 103 The doping concentration and thickness of the p-type semiconductor layer 106 and the electron blocking layer 105 The deposition temperature and growth rate of the p-type semiconductor layer 106 Regarding the parameter, since the influence on the life span is very large, when the data has a variation that cannot be regarded as constant, it is desirable to include it in the explanatory variables. In addition, it is advisable to use the deposition temperature of the layer using the composition parameter or the physical property parameter as a manufacturing parameter. As explained in FIG. 4, when at least the film thickness of the p-type semiconductor layer 106 is used as an explanatory variable, it is desirable to include the deposition temperature of the p-type semiconductor layer 106 in the explanatory variables. And it is more desirable to include a position parameter representing the position on the wafer in the explanatory variable. Although not shown in FIG. 7, a step of selecting which parameter to use as the explanatory variable may be added before step S12.

[0047] After creating the trained model 32 in step S12, in step S13, the model creation unit 22 stores the created trained model 32 in the memory unit 3, and the process proceeds to the life prediction step of step S2.

[0048] In the life prediction process of step S2, first, in step S21, the prediction source data 33 is input. At this time, for example, an input screen for the prediction source data 33 may be displayed on the display 4 so that the prediction source data 33 can be input by the input device 5. Also, when the prediction source data 33 is input in a file format, a display prompting the input of the file may be displayed on the display 4. Thereafter, in step S22, the life prediction unit 23 predicts a life corresponding to the prediction source data 33 using the trained model 32 stored in the storage unit 3. Thereafter, in step S23, the life prediction unit 23 stores the predicted life value in the storage unit 3 as the predicted data 34, and proceeds to the prediction result presentation process of step S3. Note that, in the case where the trained model 32 is created using a position parameter representing a position on the wafer as an explanatory variable, in the life prediction process, the prediction source data 33 and data of a plurality of positions on the wafer that have been set in advance (for example, the center of the wafer, the outer edge, etc.) may be used to obtain a life at each of the plurality of positions on the wafer that have been set in advance. In this case, the lifetime prediction process may calculate an index value for lifetime evaluation over the entire wafer, for example by calculating the average or median of predicted lifetime values ​​at multiple positions on the wafer, and this index value may be included in the prediction data 34.

[0049] In the prediction result presentation process of step S3, in step S31, the prediction result presentation unit 24 of step S3 presents the prediction result of the lifespan by presenting the prediction data 34 stored in the storage unit 3 on the display 4. Then, the process ends.

[0050] In the present embodiment, a case has been described in which the trained model 32 is updated when predicting a life span, but the present invention is not limited to this. The model creation unit 22 may be configured to monitor the update status of the training data 31 and update the trained model 32 every time the training data 31 is updated. The model creation unit 22 may also be configured to update the trained model 32 at predetermined intervals (for example, every week or every month).

[0051] (Functions and Effects of the Embodiments) According to the embodiment described above, the following actions and effects can be obtained.

[0052] (1) By using a trained model 32 that has been machine-learned to correlate at least one of the composition parameters or physical property parameters of the layers constituting the light-emitting element 100 with the manufacturing condition parameters that are changed when adjusting the values ​​of the composition parameters or physical property parameters, and by using the film formation temperature of a predetermined layer used for training the composition parameters or physical property parameters as a manufacturing condition parameter for training, it becomes possible to accurately predict the lifetime of the light-emitting element 100. As a result, it becomes possible to predict the lifetime of the light-emitting element 100 without making a prototype, and it becomes possible to carry out development in a short period of time without the need for trial and error through repeated prototyping and evaluation over a long period of time.

[0053] (2) In particular, when the thickness of the p-type semiconductor layer 106 is used as a composition parameter, the accuracy of predicting the lifetime is improved by further using the deposition temperature of the p-type semiconductor layer 106 as a manufacturing condition parameter for learning.

[0054] (3) Furthermore, by using position parameters representing the position on the wafer as explanatory variables, the accuracy of lifetime prediction is further improved.

[0055] (4) When a position parameter representing the position on the wafer is used as an explanatory variable, the lifetime at each of a number of positions on the set wafer is predicted, and an index value for lifetime evaluation of the entire wafer is calculated from the predicted values. This makes it possible to evaluate the lifetime of the entire wafer, taking into account the variation in lifetime due to position on the wafer.

[0056] (Summary of the embodiment) Next, the technical ideas grasped from the above-described embodiment will be described by using the reference numerals and the like in the embodiment.

[0057] [1] A method for predicting a lifetime of a nitride semiconductor light-emitting element (100), comprising: a model creation step of learning a correlation between at least one of a composition parameter, which is a parameter defining a composition of layers (102-106) constituting the nitride semiconductor light-emitting element (100), or a physical property parameter, which is a parameter defining a physical property of the layers (102-106) constituting the nitride semiconductor light-emitting element (100), a manufacturing condition parameter, which is a manufacturing condition of the nitride semiconductor light-emitting element (100) that is changed when adjusting the value of the composition parameter or the physical property parameter, and the lifetime of the nitride semiconductor light-emitting element (100), to create a trained model (32); and a lifetime prediction step of predicting a lifetime using the trained model (32), wherein in the model creation step, at least one of the composition parameter or the physical property parameter of a predetermined layer of the nitride semiconductor light-emitting element (100) is used in the learning, and a film formation temperature of the predetermined layer is used as the manufacturing condition parameter in the learning.

[0058] [2] The method for predicting a lifetime of a nitride semiconductor light-emitting element according to [1], wherein the specified layer is a p-type semiconductor layer (106) made of AlGaN, and at least a thickness of the p-type semiconductor layer (106), which is the physical property parameter, and a deposition temperature of the p-type semiconductor layer (106), which is the manufacturing condition parameter, are used in the learning.

[0059] [3] The lifetime prediction method for a nitride semiconductor light-emitting element according to [2], further comprising using a position parameter representing a position on a wafer for the learning.

[0060] [4] The method for predicting a lifetime of a nitride semiconductor light-emitting element described in [3], wherein the lifetime prediction step predicts a lifetime at each of a plurality of previously set positions on the wafer using data on the plurality of positions on the wafer, and calculates an index value for lifetime evaluation of the entire wafer using the predicted lifetime values ​​at the plurality of positions on the wafer.

[0061] [5] A device for predicting a lifetime of a nitride semiconductor light-emitting element (100), comprising: a model creation unit (22) that learns a correlation between at least one of a composition parameter that is a parameter defining a composition of layers (102 to 106) constituting the nitride semiconductor light-emitting element (100) or a physical property parameter that is a parameter defining a physical property of the layers (102 to 106) constituting the nitride semiconductor light-emitting element (100), a manufacturing condition parameter that is a manufacturing condition of the nitride semiconductor light-emitting element (100) that is changed when adjusting a value of the composition parameter or the physical property parameter, and the lifetime of the nitride semiconductor light-emitting element (100) to create a trained model (32); and a lifetime prediction unit (23) that predicts a lifetime using the trained model (32), wherein the model creation unit (22) uses at least one of the composition parameter or the physical property parameter of a predetermined layer of the nitride semiconductor light-emitting element (100) in the learning, and uses a film formation temperature of the predetermined layer as the manufacturing condition parameter in the learning.

[0062] (Additional Note) Although the embodiment of the present invention has been described above, the invention according to the claims is not limited to the embodiment described above. It should be noted that not all of the combinations of features described in the embodiment are essential to the means for solving the problems of the invention. The present invention can be modified appropriately without departing from the spirit of the invention. [Explanation of symbols]

[0063] 1. Lifetime prediction device for nitride semiconductor light-emitting element 2...Control section 21…Learning data acquisition section 22…Model Creation Section 23...Life expectancy prediction section 24…Prediction result display section 3...Storage section 31…Learning data 32…Trained model 33…Source data for prediction 34…Prediction data 100...Nitride semiconductor light emitting element (light emitting element) 101...Substrate 102...Buffer layer 103...n-type semiconductor layer 104…Active layer 104a...Barrier layer 104b...Well layer 105...Electron blocking layer 106...p-type semiconductor layer

Claims

1. An apparatus for predicting the lifetime of a semiconductor light emitting element, comprising: a data acquisition unit that acquires data corresponding to at least one of composition parameters, which are parameters that define the composition of a layer constituting the semiconductor light emitting device, or physical property parameters, which are parameters that define the physical properties of a layer constituting the semiconductor light emitting device, and data on the lifetime of the semiconductor light emitting device from outside the device, and creates learning data by linking the acquired data corresponding to at least one of the composition parameters or the physical property parameters with the lifetime data; a model creation unit that uses the learning data as teacher data, and performs machine learning on a correlation between the explanatory variables and the objective variable, with at least one of the composition parameters and the physical property parameters as explanatory variables and the lifetime as an objective variable, to create a learned model that represents the correlation; a life prediction unit that predicts the life corresponding to prediction source data, which is data of any explanatory variable, by using the trained model, and sets the predicted data as prediction data; A prediction result presentation unit that presents the prediction data, the model creation unit creates the trained model by using position parameters representing positions on a wafer as the explanatory variables; In the life prediction unit, the prediction source data not including the position parameters is input, and the prediction source data and a plurality of the position parameters set in advance are input to the trained model, so as to automatically predict the lives at a plurality of positions on the wafer. A life prediction device for semiconductor light emitting devices.

2. The life prediction unit calculates an index value for evaluating a life of the entire wafer using the predicted values ​​of the life at a plurality of positions on the wafer, and includes the index value in the prediction data. The device for predicting the lifetime of a semiconductor light-emitting element according to claim 1 .

Images