Information processing device, information processing method and program

Abstract

To provide an information processing device, an information processing method and a program capable of reducing cost of time, effort, expense, or the like for acquiring an optimal condition about thin film formation.SOLUTION: An information processing device is provided with one or more transposed convolution layers, includes one or more activation layers, inputs a condition for thin film formation to a generation model 100 formed by a decoder 120 in which at least one activation coefficient in the one or more activation layers is a ReLU (Rectified LinearUnit), and outputs a characteristic prediction result of a thin film predicted from the condition.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present invention relates to an information processing device, an information processing method, and a program, and more particularly to an information processing device, a method, and a program that can be used for, for example, an AI combinatorial physical property data prediction system. [Background technology]

[0002] Conventionally, combinatorial methods have been known as a method for searching for materials, compounds, or products that are optimal for a certain application (hereinafter, also referred to as optimal solutions). For example, in the field of chemistry, combinatorial chemistry exists in which a wide variety of compound groups (libraries) are efficiently synthesized by combining multiple raw materials based on combinatorial theory, and the compound libraries thus created are utilized for various purposes. The method of synthesizing a wide variety of compound groups at once based on combinatorial theory is called combinatorial synthesis, and is positioned as the core technology of combinatorial chemistry.

[0003] Moreover, in recent years, research has been conducted into the use of AI (Artificial Intelligence) technology to find materials that are optimal for a particular application.

[0004] For example, Patent Document 1 shown below discloses a technology for generating compounds that have high binding affinity with a target protein by training a conditional variational autoencoder (CVAE) using a set of pairs of known proteins and compounds that are known to have high binding affinity, and inputting the target protein (distributed representation) and latent variables (sampled with random noise) into the decoder of the trained conditional variational autoencoder.

[0005] Furthermore, Patent Document 2 discloses a technique for generating compounds having activity against a selected target using a generative model trained using a training set including compounds and biological or chemical information related to the compounds.

[0006] Furthermore, Patent Document 3 discloses a technology for identifying a desired dye material that satisfies all of multiple physical properties, using a VAE encoder trained to receive dye material information expressed in a predetermined notation as input and output a latent variable in a latent space equivalent to the dye material information, a VAE decoder trained to receive an arbitrary latent variable in the latent space as input and output dye material information expressed in a predetermined notation, and data related to multiple physical properties of the dye material. [Prior art documents] [Patent documents]

[0007] [Patent Document 1] Patent Publication No. 2021-68410 [Patent Document 2] Patent Publication No. 2021-121927 [Patent Document 3] Patent No. 7180806 Summary of the Invention [Problem to be solved by the invention]

[0008] However, the above conventional techniques require a huge amount of teacher data to train the generative model, and it takes a long time to learn the generative model, which poses the problem that it takes a huge amount of time, effort, and cost to obtain an optimal solution.

[0009] Therefore, an object of the present invention is to provide an information processing device, an information processing method, and a program that make it possible to reduce the cost required to obtain an optimal solution. [Means for solving the problem]

[0010] An information processing apparatus according to an embodiment of the present invention includes a generative model that receives conditions for thin film formation and outputs a prediction result of the thin film characteristics predicted from the conditions. [Brief description of the drawings]

[0011] [Figure 1] FIG. 1 is a diagram illustrating an example of a schematic configuration of a generative model according to an embodiment of the present invention. [Diagram 2] FIG. 2 illustrates an example of a neural network structure of a decoder according to an embodiment of the present invention. [Diagram 3] FIG. 1 is a diagram for explaining learning of a generative model according to one embodiment of the present invention. [Figure 4] 11 is a flowchart illustrating an example of the operation of a generative model according to the present embodiment. [Diagram 5] FIG. 2 is a diagram showing an example of measurement points (16 points) used as input to a generative model in characteristic prediction using the generative model according to one embodiment of the present invention. [Figure 6] FIG. 2 is a diagram showing an example of a three-dimensional combinatorial library obtained by a generative model according to an embodiment of the present invention. [Figure 7] FIG. 7 is a diagram showing an example of a combinatorial library in a direction of a certain angle (θ1) extracted from the three-dimensional combinatorial library shown in FIG. 6. [Figure 8] FIG. 8 is a diagram showing the combinatorial library shown in FIG. 7 using contour lines. [Figure 9] FIG. 13 is a diagram showing an example of a combinatorial library obtained by Bayesian optimization as a comparative example. [Figure 10] FIG. 2 shows an example of a portion of a combinatorial library that can be obtained with a generative model according to an embodiment of the present invention. [Figure 11] 1 is a hardware configuration diagram showing an example of an information processing device according to an embodiment of the present invention; DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings.

[0013] <0. Introduction> In recent years, with the advancement of AI (artificial intelligence) technology, various generative models using neural networks have been developed.

[0014] For example, an autoencoder with a typical hourglass network structure has an input-side encoder and an output-side decoder, and the decoder can be used as a generative model after learning.

[0015] The input-side encoder has, for example, a convolutional neural network structure, inputs a high-dimensional observable variable (hereinafter, simply referred to as a variable) x (hereinafter, the input-side variable x is also referred to as an input x), and outputs a low-dimensional latent variable z that is uniquely determined from this input x.

[0016] On the other hand, the decoder on the output side has, for example, a transposed convolutional neural network structure, and outputs a reconstructed high-dimensional variable x for an input of a low-dimensional latent variable z (hereinafter, the variable x on the output side is also referred to as the output x).

[0017] Autoencoders with such structures are typically used in unsupervised learning, i.e., they can be trained without a teacher signal by training the output x to approach the input x.

[0018] In training an autoencoder, the trainer learns how to summarize the input x to obtain a latent variable z that can recover the input x as the output x.

[0019] After the autoencoder has been trained, new data with the same number of dimensions as the output x can be generated by inputting random values ​​with the same number of dimensions as the latent variable z to the decoder. In other words, the encoder can be omitted and the decoder alone can function as a generative model.

[0020] Therefore, in this embodiment, we will explain by way of examples an information processing device, information processing method, and program that, unlike conventional techniques, uses a decoder in a neural network generative model such as an autoencoder as a standalone generative model, thereby making it possible to obtain high-dimensional output from low-dimensional information while reducing the time and cost required to obtain an optimal solution.

[0021] Note that the neural network generation model in this description is not limited to the structure shown here as an example, and can use various operations used in neural networks, such as CNN (Convolutional Neural Network), RNN (Recurrent Neural Network), Transformer, Attention, etc. For clarification, the following description will be given by way of example of a case where a decoder using a transposed convolution operation is used as a generation model.

[0022] <1. Generative model> Next, the generative model according to this embodiment will be described in detail with reference to the drawings. In this description, the generative model (also called a generative model) may be a statistical model based on the joint probability distribution p(x|z) of an observable variable x and a latent variable (also called a target variable) z, and may be a neural network capable of outputting new data by performing an operation using the latent variable on input data (for example, a transpose convolution operation to be described later). There is no limit to the latent variable, and various values ​​can be used.

[0023] Fig. 1 is a diagram showing a schematic configuration example of a generative model according to the present embodiment. As shown in Fig. 1, the generative model 100 according to the present embodiment is configured with a decoder (also called a generator) 120, in other words, it has a configuration of only the decoder 120 without an encoder.

[0024] In this embodiment, by having the generative model 100 learn an X-ray diffraction image, it is possible to generate an X-ray diffraction image (also referred to as X-ray data) of a thin film (hereinafter also referred to as a measured film) that can be formed under certain conditions without requiring an encoder. Here, the measured film may be a thin film that is a target for characteristic prediction.

[0025] In this description, an X-ray diffraction image that can be obtained by X-ray diffraction (XRD (X-ray diffraction)) is exemplified as a predicted characteristic, but the present invention is not limited to this and can be configured to predict various characteristics. In addition, in this description, the conditions used when forming the measurement film are also referred to as experimental conditions, and the combination of these experimental conditions and experimental data (in this example, X-ray diffraction images) obtained by measuring the measurement film formed under these experimental conditions is also referred to as measurement points.

[0026] In this embodiment, the input z is directly input to the decoder 120, which is the generative model 100. Therefore, the input z to the generative model 100 can be data of the same dimension as the latent variable z, which is lower in dimension than the output x.

[0027] <1.1. Specific configuration example of the decoder 120> Next, a specific configuration example of the decoder 120 in the generative model 100 will be described in detail with reference to the drawings.

[0028] Fig. 2 is a diagram showing an example of a neural network structure of a decoder according to this embodiment. As shown in Fig. 2, the decoder 120 is composed of a fully connected layer 121 and one or more transposed convolution layers 123 to 126, in order from the input side.

[0029] The input z input to the decoder 120 is first converted into 224-dimensional x 1-channel data by being input to the fully connected layer 121. The input z input to the decoder 120 may be multimodal information, such as the experimental conditions when forming the measurement film, and may be sufficiently lower-dimensional information than the image data obtained as the output. In addition, the 224-dimensional x 1-channel data is reshaped into 7-dimensional x 32-channel data by being reshaped.

[0030] Each of the transposed convolution layers 123-126 performs a transposed convolution operation that is the apparent inverse of a convolution operation, thereby expanding the input data. That is, each of the transposed convolution layers 123-126 executes the same arithmetic processing as that performed by a convolution layer in a neural network such as a CNN, but since upsampling is also performed simultaneously during the arithmetic processing, the signal dimension is expanded. In the learning of the decoder 120, each coefficient of a filter coefficient matrix (hereinafter simply referred to as a filter) called a kernel used in this transposed convolution operation is updated by learning.

[0031] The decoder 120 performs such a transposed convolution operation one or more times to synthesize a high-dimensional output x from a low-dimensional input z. For example, as shown in Fig. 2, when four transposed convolution layers 123 to 126 are provided, it is possible to finally obtain a 2401-dimensional output x. In this case, by inputting the experimental conditions as the input z, a 2401-dimensional X-ray diffraction image is synthesized.

[0032] Note that the activation functions of all or one or more activation layers in the decoder 120 may be ReLU (Rectified Linear Unit). However, this is not limited thereto, and various activation functions such as SiLU (Sigmoid-weighted Linear Unit) may be used.

[0033] <2. Input to the generative model> Next, the input z to the generative model 100 according to this embodiment will be described.

[0034] In this embodiment, the experimental data can be obtained by measuring a measurement film formed under certain experimental conditions. This suggests that the experimental conditions completely satisfy the properties of the latent variable z. That is, in this embodiment, the experimental conditions for forming the measurement film can be used as the input z.

[0035] Moreover, the experimental conditions used when forming a certain measurement film may include information such as layer structure, film formation method for each layer, film formation temperature, film formation time, composition, film thickness, chamber pressure, plasma excitation frequency, dose amount, etc. Furthermore, the characteristics (here, X-ray diffraction image) of a measurement film formed under the same experimental conditions are considered to be the same if the variations in film formation conditions, etc. are ignored.

[0036] In this way, the fact that the experimental data (characteristics) of the measurement film obtained from a certain experimental condition is substantially uniquely determined suggests that the decoder 120 according to this embodiment can be trained alone (the first main feature of the generative model according to this embodiment). That is, in the learning of a normal generative model, as described above, the encoder and the decoder are trained as a set. Specifically, in unsupervised learning when an autoencoder having a conventional hourglass-type network structure is used as the generative model, an input x is input to the encoder, and the latent variable z of the constriction is trained so that the output x output from the decoder matches the input x. On the other hand, in the learning of the generative model 100 according to this embodiment, the experimental conditions of each measurement point are directly input to the decoder 120, and the decoder 120 is trained so that the output corresponding to the input approaches the experimental data of each measurement point (here, X-ray diffraction image), thereby training the generative model 100.

[0037] Furthermore, the fact that the experimental data (characteristics) of the measurement film obtained under certain experimental conditions are substantially uniquely determined suggests that even if the generative model 100 is trained with a very small number of data items, such as a few to a dozen or so, sufficient estimation accuracy can be achieved (the second main feature of the generative model according to this embodiment). That is, a huge number of data items, such as big data, are used to train a normal generative model. This is because, in order to improve the estimation accuracy using deep learning, it is necessary to sufficiently train the neural network using a huge number of data items. On the other hand, in this embodiment, as described above, the experimental data (characteristics) of the measurement film obtained under certain experimental conditions are substantially uniquely determined, so that it is possible to train the generative model with a very small number of data items, such as a few to a dozen or so.

[0038] In this embodiment, the experimental conditions used as the input z do not necessarily have to be all of the experimental conditions listed above (layer structure, deposition method for each layer, deposition temperature, deposition time, composition, film thickness, pressure inside the chamber, plasma excitation frequency, dose amount, etc.), but may be at least one or a combination of two or more of these experimental conditions.

[0039] <3. Learning the generative model> Next, the learning of the generative model 100 (that is, the decoder 120) according to this embodiment will be described with reference to the drawings. Fig. 3 is a diagram for explaining the learning of the generative model according to the present embodiment. As shown in Fig. 3, the learning of the generative model 100 uses a combination (i.e., measurement points) of experimental conditions used in forming a certain measurement film and experimental data (here, X-ray diffraction images) obtained by measuring the measurement film formed under the experimental conditions.

[0040] In the independent learning of the generative model 100, i.e., the decoder 120, an experimental condition (input z) is input to the decoder 120. The learning unit (also called an optimizer) 140 learns so that the experimental data (output x) output from the decoder 120 for this input z approaches the experimental data actually obtained from a measurement film formed under the experimental conditions.

[0041] In addition, various optimization algorithms 140 such as Stochastic Gradient Descent (SGD), MomentumSGD, Nesterov's Acceleration Algorithm (NAG), AdaGrad, RMSprop, AdaDelta, Adam, RMSpropGraves, SMORMS3, AdaMax, Nadam, Eve, Santa, GD by GD, AdaSecant, AMSGrad, AdaBound, and AMSBound can be used for training the decoder 120 by the training unit 140.

[0042] <4. Creation of combinatorial libraries> Next, the creation of a combinatorial library using the generative model 100 according to this embodiment will be described.

[0043] As described above, the trained generative model 100 can predict experimental data of a measurement film formed under certain experimental conditions by inputting the experimental conditions. Therefore, according to this embodiment, by inputting all experimental conditions in the prediction range into the generative model 100, it is possible to interpolate experimental data of measurement points where no experimental data has actually been obtained, and create a combinatorial library for the entire prediction range.

[0044] 4 is a flowchart showing an example of the operation of the generative model according to the present embodiment when creating a combinatorial library for the entire prediction range. Note that the operation described below may be performed under the control of a processor that executes the operation of the learning unit 140 (for example, the CPU 1100 described later with reference to FIG. 11).

[0045] 4, in this operation, first, the generative model 100 is trained (step S101), and then one experimental condition within the prediction range is input to the trained generative model 100 (step S102). In response, the generative model 100 outputs experimental data predicted for a measurement film formed under the input experimental conditions (step S103).

[0046] Next, it is determined whether all experimental conditions within the prediction range have been input into the generative model 100 (step S104), and if there are experimental conditions that have not yet been input (NO in step S104), the operation returns to step S102 and the subsequent operations are repeated.

[0047] In this way, when all the experimental conditions within the prediction range are input to the generative model 100 and the experimental data at each measurement point is predicted (YES in step S104), the experimental data at all the measurement points are collected to create a combinatorial library for the entire prediction range (step S105). After that, this operation ends.

[0048] <5. Property prediction using the generative model according to this embodiment> Next, characteristic prediction using the generative model 100 according to this embodiment will be described.

[0049] <5.1. XRD search as a concrete example> First, characteristic prediction using the generative model of this embodiment will be explained using an example in which the generative model of this embodiment is used to search for combinations of luminescent materials to be used in semiconductor lasers, LEDs (Light Emitting Diodes), etc.

[0050] In the development of semiconductor lasers, the crystal structure of the light-emitting layer formed from a combination of various materials is usually analyzed by X-ray diffraction (XRD). Therefore, it is easy to obtain a combination of the experimental conditions during film formation (one or more of the following information: layer structure, film formation method for each layer, film formation temperature, film formation time, composition, film thickness, chamber pressure, plasma excitation frequency, dose amount, etc.) and the experimental data obtained by performing X-ray diffraction on the light-emitting layer formed under these experimental conditions (hereinafter also referred to as X-ray diffraction image or X-ray data).

[0051] Conventionally, when searching for a combination of luminescent materials using a combinatorial method, for example, one or more substrates (e.g., semiconductor wafers, hereinafter also referred to as samples) whose film thickness, composition, etc. vary depending on the position are prepared, and X-ray diffraction images are measured for each position on the sample to identify the dependency of the X-ray peak on the film thickness, composition, etc. of the luminescent material. Then, the combination of materials that is considered to be optimal is determined based on the identified dependency.

[0052] However, in such conventional methods, it is necessary to prepare and measure samples of all or many combinations that can be candidates, and therefore it takes a lot of time and effort (e.g., the number of trials) and costs to prepare a combinatorial library. In addition, for combinations other than the combinations of materials prepared as samples, it is necessary to search for them using a method such as Bayesian optimization that is commonly used in fields such as MI (Material Informatics). However, in order to perform a sufficiently reliable search (prediction) using these conventional search methods, many measurement results are required, and therefore it is difficult to sufficiently reduce costs such as time and effort (e.g., the number of trials) and costs.

[0053] In contrast, the generative model according to the present embodiment is capable of generating highly accurate measurement data from a small number of discrete measurement points, which makes it possible to significantly reduce the cost required for creating a combinatorial library, and therefore the cost required for obtaining an optimal material combination.

[0054] FIG. 5 is a diagram showing an example of 16 measurement points (experimental conditions and experimental data) used as input for characteristic prediction. FIG. 6 is a diagram showing an example of a characteristic prediction result (three-dimensional combinatorial library) obtained by inputting the 16 measurement points shown in FIG. 5 into the generation model according to this embodiment. FIG. 7 is a diagram showing an example of a combinatorial library in a direction of a certain angle (θ1) extracted from the three-dimensional combinatorial library shown in FIG. 6. FIG. 8 is a diagram showing the combinatorial library generated at a certain angle (θ1) shown in FIG. 7 using isopleths. FIG. 9 is a diagram showing an example of a combinatorial library obtained by Bayesian optimization from the experimental conditions shown in FIG. 5. Note that, even with Bayesian optimization, it is possible to generate a three-dimensional combinatorial library in the same way as with the generation model 100 according to this embodiment. However, for the sake of simplicity of explanation, FIG. 9 shows an example of a combinatorial library in a direction of a certain angle (θ1) extracted from the three-dimensional combinatorial library.

[0055] In Fig. 6, the X-axis (corresponding to the horizontal axis in Figs. 7 to 9) indicates temperature, the Y-axis (corresponding to the vertical axis in Figs. 7 to 9) indicates pressure, and the Z-axis indicates the observation angle (θ) of the X-ray diffraction image. In Figs. 7 to 9, the horizontal axis indicates temperature and the vertical axis indicates pressure. In Figs. 7 and 9, the 16 measurement points shown in Fig. 5 are indicated by "◯", and in Fig. 8, the 16 measurement points shown in Fig. 5 are indicated by "X". It should be noted that a combinatorial library of pressure against temperature as exemplified in Figs. 6 to 8 has not been possible to create in the past.

[0056] As shown in Fig. 5, in this embodiment, it is possible to input a small number of discrete experimental conditions (for example, 16 measurement points that are widely and discretely distributed) to the generative model 100. As can be seen by comparing Fig. 7 and Fig. 8 with Fig. 9, the characteristic prediction results (Figs. 7 and 8) obtained by the generative model 100 according to this embodiment when a small number of discrete experimental conditions are input show a band-like dependence of pressure on temperature.

[0057] In contrast, in the characteristic prediction results obtained by Bayesian optimization for the same input (Figure 9), the dependence of pressure on temperature appears only in the vicinity of the measurement points. This suggests that while the generative model 100 can obtain characteristic prediction results over a wide range with high accuracy even with a small number of discrete measurement points, Bayesian optimization cannot predict characteristics in areas where there are no measurement points, and therefore cannot obtain accurate characteristic prediction results overall.

[0058] <5.2. Examples of combinatorial libraries that can be generated> Next, several examples of combinatorial libraries that can be generated by the generative model 100 according to this embodiment will be described. The combinatorial libraries exemplified below are examples of some of the combinatorial libraries that can be generated by the generative model 100 according to this embodiment. The generative model 100 according to this embodiment is not limited to these examples, and can generate various combinatorial libraries by changing the learning, input, etc.

[0059] Fig. 10 is a diagram showing an example of a combinatorial library obtained by a generative model according to this embodiment. For the sake of simplicity, (b) to (e) of Fig. 10 show a combinatorial library at a certain angle (θ2), but in reality, a three-dimensional combinatorial library with the angle (θ) as the z-axis may be generated.

[0060] Fig. 10(a) shows an example of measurement points input to the generative model, and Fig. 10(b) to (e) show examples of combinatorial libraries at a certain angle (θ2) obtained for the input of Fig. 10(a). Fig. 10(b) is a graph of the temperature gradient versus the composition gradient, Fig. 10(c) is a graph of the composition gradient versus the composition gradient, Fig. 10(d) is a graph of the pressure gradient versus the target-substrate distance, and Fig. 10(e) is a graph of the pressure gradient versus the composition gradient.

[0061] As shown in Figs. 10(a) and (b), according to the generation model of this embodiment, by acquiring X-ray diffraction images of several measurement points that are discretely scattered in the search space (see Fig. 10(a)), it is possible to create a combinatorial library of temperature gradients for composition gradients (see Fig. 10(b)). Also, as shown in Figs. 10(a) and (c), a combinatorial library of double composition gradients (composition gradients for composition gradients), which is a specialty of conventional combinatorial methods, can be estimated from a small number of discrete X-ray diffraction images. Furthermore, as shown in Figs. 10(a), (d) and (e), it is also possible to create a peak intensity map for pressure gradients, which was difficult to do with conventional combinatorial methods. For example, it is also possible to create a combinatorial library in which the vertical axis is the pressure gradient and the horizontal axis is the target-substrate distance gradient, as shown in Fig. 10(d), or a combinatorial library in which the vertical axis is the pressure gradient and the horizontal axis is the composition gradient, as shown in Fig. 10(e). In addition, it is possible to create combinatorial libraries in which the vertical axis represents a pressure gradient and the horizontal axis represents a temperature gradient, which was not possible with conventional combinatorial methods, and to create peak intensity mappings in which the target-substrate distance gradient and the pressure gradient are changed simultaneously.

[0062] As described above, by using the generation model according to this embodiment, a technique for mapping peak intensities in an X-ray diffraction image obtained by XRD can be established.

[0063] <6. Summary> As described above, according to this embodiment, it is possible to realize a generative model 100 capable of generating highly accurate measurement data from a small number of discrete measurement points. This makes it possible to significantly reduce the time, effort (e.g., the number of trials) and costs required for creating a combinatorial library, and therefore the cost required for obtaining an optimal material combination can be significantly reduced. In addition, according to this embodiment, it is possible to train the decoder 120 without requiring an encoder, and therefore the cost required for training the generative model 100 can also be reduced.

[0064] In the above embodiment, a combination of light-emitting materials used in semiconductor lasers and LEDs is given as a specific example of the measurement film, but the present invention is not limited thereto, and the generation model 100 according to the present embodiment can be applied to the search for combinations of various materials. For example, for fluorescent materials used in radiation detectors, a combinatorial library of combinations can be generated using the generation model 100 according to the present embodiment by using a fluorescence spectrum obtained by a fluorescence spectrophotometer or the like as experimental data instead of an X-ray diffraction image. As described above, the information processing device of the present invention can be called an AI combinatorial property data prediction system.

[0065] <7. Hardware configuration> The generative model 100 and the like according to the above-described embodiments and their modified examples can be realized by, for example, an information processing device 1000 having a configuration as shown in Fig. 11. Fig. 11 is a hardware configuration diagram showing an example of the information processing device 1000 that realizes the functions of the generative model 100 and the like. The information processing device 1000 has a CPU 1100, a read only memory (ROM) 1200, a random access memory (RAM) 1300, a recording device 1400, an input / output interface (I / F) 1500, and a communication unit 1600. The various units of the information processing device 1000 are connected by a bus 1700.

[0066] The CPU 1100 operates based on a program stored in the ROM 1200 or the recording device 1400, and controls each unit. For example, the CPU 1100 loads the program stored in the ROM 1200 or the recording device 1400 into the RAM 1300, and executes processing corresponding to the various programs.

[0067] The ROM 1200 stores a boot program such as a basic input output system (BIOS) executed by the CPU 1100 when the information processing device 1000 is started, and programs that depend on the hardware of the information processing device 1000, and the like.

[0068] Recording device 1400 is a computer-readable recording medium that non-temporarily records programs executed by CPU 1100 and data used by such programs, etc. Specifically, recording device 1400 is a recording medium that records programs for executing each operation according to the present invention, which is an example of program data.

[0069] The communication unit 1600 is an interface for connecting the information processing device 1000 to an external network 1650 (e.g., the Internet). For example, the CPU 1100 receives data from other devices and transmits data generated by the CPU 1100 to other devices via the communication unit 1600.

[0070] The input / output I / F 1500 is an interface for connecting the input / output device 1650 and the information processing device 1000. For example, the CPU 1100 receives data from an input device such as a keyboard or a mouse via the input / output I / F 1500. The CPU 1100 also transmits data to an output device such as a display, a speaker or a printer via the input / output I / F 1500. The input / output I / F 1500 may also function as a media interface that reads a program or the like recorded on a predetermined recording medium.

[0071] For example, when the information processing device 1000 realizes the functions of the generative model 100 according to the above-mentioned embodiment, the CPU 1100 of the information processing device 1000 realizes the functions of the generative model 100 by executing a program loaded on the RAM 1300. The recording device 1400 stores the programs according to the present invention. The CPU 1100 reads and executes the program data from the recording device 1400, but as another example, the CPU 1100 may obtain these programs from other devices via the external network 1650.

[0072] Although the embodiment and its modifications of the present invention have been described above, the technical scope of the present invention is not limited to the above-mentioned embodiment or its modifications, and various modifications are possible without departing from the gist of the present invention. In addition, components of different embodiments and modifications may be appropriately combined. [Explanation of symbols]

[0073] 100 Generative Models 120 Decoder 140 Optimization Algorithms

Claims

1. An information processing device comprising a generative model for inputting conditions for thin film formation and outputting predicted results of characteristics of the thin film predicted from the conditions.

2. The information processing apparatus according to claim 1 , wherein the generative model comprises a decoder that receives the condition as an input and outputs the characteristic prediction result.

3. the decoder comprises one or more transposed convolutional layers; the decoder includes one or more activation layers; At least one activation function of the one or more activation layers is a Rectified Linear Unit (ReLU); The information processing device according to claim 2 .

4. The information processing apparatus according to claim 1 , wherein the conditions include at least one of a layer structure, a film formation method for each layer, a film formation temperature, a film formation time, a composition, a film thickness, a pressure inside a chamber, a plasma excitation frequency, and a dose amount.

5. The information processing apparatus according to claim 1 , wherein the characteristic prediction result includes at least one of a temperature gradient with respect to a composition gradient, a composition gradient with respect to a composition gradient, a pressure gradient with respect to a target-substrate distance gradient, a pressure gradient with respect to a composition gradient, and a pressure gradient with respect to a temperature gradient.

6. An information processing method executed in an information processing device, Enter the conditions for thin film formation, Converting the conditions into one-dimensional data; performing one or more transpose convolution operations on the one-dimensional data to generate a property prediction result of the thin film; An information processing method comprising the steps of:

7. A program for causing a processor implemented in an information processing device to function, inputting conditions for thin film formation; converting the conditions into one-dimensional data; performing one or more transpose convolution operations on the one-dimensional data to generate a predicted thin film characteristic result; A program for causing the processor to execute the above.

Images