Characteristic prediction system

Abstract

To provide a characteristic prediction system.SOLUTION: A characteristic prediction system includes an input unit, a processing unit, an arithmetic unit, and an output unit, and the input unit has a function of supplying the structure of a light-emitting device or the characteristics of the light-emitting device, the processing unit has a function of generating a learning dataset or data used for predicting the characteristics and a function of quantifying the molecular structure of an organic compound, the arithmetic unit has a function of performing supervised learning on the basis of the learning dataset and a function of inferring the characteristics of the light-emitting device from the data on the basis of the learning results of the supervised learning, and the output unit has a function of providing the results of the inference, thereby predicting the characteristics of the light-emitting device having a layer containing the organic compound.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] One aspect of the present invention relates to a method for predicting characteristics of a light-emitting device and a characteristic prediction system for predicting characteristics of a light-emitting device. [Background technology]

[0002] Light-emitting devices (also called light-emitting elements) that have an EL (Electro Luminescence) layer sandwiched between a pair of electrodes have characteristics such as being thin and lightweight, having fast response to input signals, and low power consumption, and displays that use these devices are attracting attention as next-generation flat panel displays.

[0003] Light-emitting devices that use organic compounds in the EL layer (also called organic EL devices or organic EL elements) exhibit a variety of characteristics by stacking various organic compounds. However, there are more than one million known types of organic compounds. Furthermore, when a light-emitting device has a stacked structure, the film thickness of each stacked layer and the concentration ratio of the organic compounds also affect the characteristics of the light-emitting device. Therefore, a great deal of effort is required to optimize the structure of the light-emitting device (also simply called the element structure).

[0004] In recent years, methods for classification, estimation, prediction, etc., using methods such as machine learning have undergone great evolution. The performance of selection and prediction using neural networks (especially deep learning) has improved significantly, and excellent results have been achieved in various fields. Patent Document 1 discloses a method for discovering new substances using machine learning and an apparatus for the same. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] JP 2017-91526 A Summary of the Invention [Problem to be solved by the invention]

[0006] Optimizing the structure of an organic electroluminescence (EL) device depends heavily on skilled personnel. The technology of organic electroluminescence (EL) devices spans multiple academic fields, and requires knowledge of organic chemistry, semiconductor physics, electrical engineering, and other fields, so a high level of skill is required. In addition, optimization takes many years, so the process must be memorized.

[0007] In view of the above, an object of one embodiment of the present invention is to provide a method for predicting characteristics of a light-emitting device.Another object of one embodiment of the present invention is to provide a system for predicting characteristics of a light-emitting device.Another object of one embodiment of the present invention is to provide a method for generating a learning dataset.

[0008] Note that the description of these problems does not preclude the existence of other problems. Note that one embodiment of the present invention does not necessarily solve all of these problems. Note that problems other than these will become apparent from the description of the specification, drawings, claims, etc., and it is possible to extract problems other than these from the description of the specification, drawings, claims, etc. [Means for solving the problem]

[0009] In view of the above problems, one embodiment of the present invention provides a system for predicting characteristics of a light-emitting device from information about the light-emitting device by using information about the light-emitting device and characteristics of the light-emitting device for machine learning.

[0010] One aspect of the present invention is a characteristic prediction system that predicts characteristics of a light-emitting device having a layer containing an organic compound. The characteristic prediction system has an input unit, a processing unit, a calculation unit, and an output unit. The input unit has a function of supplying the structure of the light-emitting device or the characteristics of the light-emitting device. The processing unit has a function of generating a learning data set or data used for predicting the characteristics, and a function of quantifying the molecular structure of the organic compound. The calculation unit has a function of performing supervised learning based on the learning data set, and a function of inferring the characteristics of the light-emitting device from the data based on the learning result of the supervised learning. The output unit has a function of providing the result of the inference.

[0011] In the above characteristic prediction system, it is preferable that the light-emitting device has a plurality of layers, one or more of the plurality of layers having one or more organic compounds, the training data set has a plurality of training data, each of the plurality of training data having input data and teacher data for the input data, the input data having a stacking order of the plurality of layers, a molecular structure of the one or more organic compounds in one or more of the plurality of layers, a film thickness of each of the plurality of layers, and a concentration ratio of the plurality of organic compounds in the layer having the plurality of organic compounds, and the teacher data has characteristics of the light-emitting device for the input data.

[0012] In addition, in the above-mentioned characteristic prediction system, the characteristic of the light-emitting device is preferably one or more of luminance-current density characteristic, current efficiency-luminance characteristic, luminance-voltage characteristic, current-voltage characteristic, external quantum efficiency-luminance characteristic, chromaticity-luminance characteristic, emission spectrum, and reliability.

[0013] Another aspect of the present invention is a characteristic prediction system for predicting characteristics of a light-emitting device having a layer containing an organic compound. The characteristic prediction system includes an input unit, a processing unit, a calculation unit, and an output unit. A learning data set and data used for predicting the characteristics are input to the input unit. The processing unit has a function of quantifying the molecular structure of the organic compound. The calculation unit has a function of performing supervised learning based on the learning data set, and a function of inferring the reliability of the light-emitting device from the data based on the learning result of the supervised learning. The output unit has a function of providing the result of the inference. The learning data set includes a plurality of learning data, and each of the plurality of learning data includes input data and a plurality of teacher data for the input data. The input data includes a stacking order of the plurality of layers, a molecular structure of one or more organic compounds included in each of the plurality of layers, a film thickness of each of the plurality of layers, a concentration ratio of the plurality of organic compounds in the layer including the plurality of organic compounds, and an external quantum efficiency-luminance characteristic of the light-emitting device. The teacher data includes the reliability of the light-emitting device for the input data.

[0014] In the above-mentioned property prediction system, the molecular structure of the organic compound is preferably quantified using quantitative structure-activity relationship or fingerprint method.

[0015] In the above characteristic prediction system, it is preferable that a neural network is used for supervised learning, and that the neural network has two or more hidden layers between an input layer and an output layer.

[0016] In addition, it is preferable that the above characteristic prediction system further includes a storage unit, in which the trained model generated by supervised learning is stored. Effect of the Invention

[0017] According to an aspect of the present invention, a method for predicting characteristics of a light-emitting device can be provided. Also, according to an aspect of the present invention, a system for predicting characteristics of a light-emitting device can be provided. Also, according to an aspect of the present invention, a method for generating a training dataset can be provided.

[0018] The effects of one embodiment of the present invention are not limited to the effects listed above. The effects listed above do not preclude the existence of other effects. The other effects are effects not mentioned in this section, which will be described below. Effects not mentioned in this section can be derived by a person skilled in the art from descriptions in the specification, drawings, etc., and can be appropriately extracted from these descriptions. One embodiment of the present invention has at least one of the effects listed above and / or other effects. Therefore, one embodiment of the present invention may not have the effects listed above in some cases. [Brief description of the drawings]

[0019] [Figure 1] 1A and 1B are diagrams illustrating a characteristic prediction system. [Diagram 2] FIG. 2 is a flow diagram showing a method for predicting the characteristics of a light-emitting device. [Diagram 3] FIG. 3 is a flow diagram showing a method for predicting the characteristics of a light-emitting device. [Figure 4] 4A and 4B are diagrams for explaining the configuration of a neural network. [Diagram 5] FIG. 5 is a diagram showing the structure of a light-emitting device. [Figure 6] 6A to 6D are diagrams illustrating the learning data set. [Figure 7] FIG. 7 is a diagram showing a method of converting a molecular structure by the fingerprint method. [Figure 8] 8A to 8D are diagrams explaining types of fingerprinting. [Figure 9] FIG. 9 is a diagram illustrating the conversion from SMILES notation to notation based on the fingerprint method. [Figure 10] FIG. 10 is a diagram for explaining types of fingerprinting and overlapping notations. [Figure 11] 11A and 11B are diagrams for explaining an example in which a molecular structure is represented using multiple fingerprint methods. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0020] The embodiments will be described in detail with reference to the drawings. However, the present invention is not limited to the following description, and it will be easily understood by those skilled in the art that the modes and details of the present invention can be modified in various ways without departing from the spirit and scope of the present invention. Therefore, the present invention should not be interpreted as being limited to the description of the embodiments shown below.

[0021] In the configuration of the invention described below, the same parts or parts having similar functions are denoted by the same reference numerals in different drawings, and the repeated explanations are omitted. In addition, when referring to similar functions, the same hatch pattern may be used and no particular reference numeral may be used.

[0022] In addition, for ease of understanding, the position, size, range, etc. of each component shown in the drawings may not represent the actual position, size, range, etc. Therefore, the disclosed invention is not necessarily limited to the position, size, range, etc. disclosed in the drawings.

[0023] It should also be noted that the ordinal numbers "first," "second," and "third" used in this specification are used to avoid confusion of components and are not intended to limit the numbers.

[0024] (Embodiment 1) In this embodiment, a characteristics prediction system for predicting characteristics of a light-emitting device and a method for predicting characteristics of a light-emitting device will be described.

[0025] <Example of property prediction system configuration> An example of the configuration of a characteristic prediction system will be described with reference to FIGS. 1A and 1B.

[0026] 1A is a diagram showing an example of the configuration of a characteristic prediction system 100. The characteristic prediction system 100 includes an input unit 101, a processing unit 102, a calculation unit 103, and an output unit 104.

[0027] The input unit 101 and the processing unit 102 are connected via a transmission line. Moreover, the processing unit 102 and the calculation unit 103 are connected via a transmission line. Moreover, the calculation unit 103 and the output unit 104 are connected via a transmission line. Note that the input unit 101, the processing unit 102, the calculation unit 103, and the output unit 104 may each be connected via a transmission line.

[0028] The above-mentioned transmission path includes networks such as a local area network (LAN) and the Internet, etc. Furthermore, the network can use either or both of wired and wireless communication.

[0029] In addition, when wireless communication is used in the above network, in addition to short-range communication means such as Wi-Fi (registered trademark) and Bluetooth (registered trademark), various communication means can be used, such as communication means compliant with the third generation mobile communication system (3G), communication means compliant with LTE (sometimes called 3.9G), communication means compliant with the fourth generation mobile communication system (4G), or communication means compliant with the fifth generation mobile communication system (5G).

[0030] The input unit 101 has a function of supplying data IN1 and data IN2 to the processing unit 102. Each of the data IN1 and data IN2 includes data such as information on the light-emitting device and characteristics of the light-emitting device. The data IN2 includes at least information on the light-emitting device. The data IN2 may also include data such as characteristics of the light-emitting device.

[0031] The data IN1 and the data IN2 may be supplied to the processing unit 102 at the same timing or at different timings. When the data IN1 and the data IN2 are supplied at different timings, it is preferable that the data IN1 is supplied to the processing unit 102, and then the data IN2 is supplied to the processing unit 102.

[0032] The processing unit 102 has a function of generating a learning dataset DS from the data IN1. The processing unit 102 also has a function of generating data DI used for predicting characteristics from the data IN2. The processing unit 102 also has a function of digitizing the molecular structure of an organic compound. The processing unit 102 may also have a function of digitizing the structure of an inorganic compound.

[0033] The calculation unit 103 has a function of performing machine learning. For example, the calculation unit 103 preferably has a function of performing supervised learning based on the learning data set DS. In addition, the calculation unit 103 preferably has a function of inferring the characteristics of the light-emitting device from the data DI used for predicting the characteristics based on the learning result of the supervised learning. By performing supervised learning, it is possible to improve the accuracy of inferring the characteristics of the light-emitting device. Note that a trained model may be generated by performing the supervised learning.

[0034] For the supervised learning, it is preferable to use a neural network (particularly, deep learning). As the deep learning, it is preferable to use, for example, a convolutional neural network (CNN), a recurrent neural network (RNN), an autoencoder (AE), a variational autoencoder (VAE), a random forest, a support vector machine, gradient boosting, generative adversarial networks (GAN), or the like.

[0035] In addition, a product-sum operation is performed in a neural network. When the product-sum operation is performed by hardware, it is preferable that the calculation unit 103 has a product-sum operation circuit. As the product-sum operation circuit, a digital circuit or an analog circuit may be used. In addition, the product-sum operation may be performed on software using a program.

[0036] The calculation unit 103 may have a function of performing semi-supervised learning as machine learning. Data on the characteristics of the light-emitting device is provided as training data (also called training signal, correct label, etc.) to the training data, but in order to prepare the training data, it is necessary to actually manufacture the light-emitting device and measure the characteristics of the light-emitting device. Compared to supervised learning, semi-supervised learning requires a smaller amount of training data included in the training data set, so that inference can be performed while reducing the time spent on creating training data.

[0037] The output unit 104 has a function of providing data OUT. The data OUT includes the result of the above inference.

[0038] The above constitutes a characteristics prediction system for predicting the characteristics of a light emitting device.

[0039] The characteristic prediction system 100 is not limited to the above configuration. For example, as shown in FIG. 1B, the characteristic prediction system 100 may have a storage unit 105 in addition to the input unit 101, the processing unit 102, the calculation unit 103, and the output unit 104.

[0040] The storage unit 105 has a function of storing the trained model generated by the calculation unit 103. The characteristic prediction system 100 has the storage unit 105, so that the characteristic of the light-emitting device can be predicted based on the trained model. Therefore, by generating the trained model in advance, it is not necessary to perform supervised learning when predicting the characteristic of the light-emitting device. Therefore, the time required to predict the characteristic of the light-emitting device can be shortened.

[0041] The storage unit 105 is connected to the calculation unit 103 via a transmission line. Note that the storage unit 105 may be connected to each of the input unit 101, the processing unit 102, the calculation unit 103, and the output unit 104 via a transmission line.

[0042] The above is a description of an example of the configuration of the characteristic prediction system.

[0043] <Methods for predicting properties> Next, an example of a method for predicting the characteristics of a light-emitting device will be described with reference to FIGS. 2, 3, 4A, and 4B.

[0044] 2 is a flow diagram showing an example of a method for predicting characteristics of a light-emitting device. The method for predicting characteristics of a light-emitting device includes steps S001 to S007. Steps S001 to S003 are steps for performing supervised learning, and steps S004 to S007 are steps for inferring characteristics of the light-emitting device.

[0045] Step S001 is a step of inputting first data. Step S001 is performed by the input unit 101 shown in Figures 1A and 1B. The first data corresponds to data IN1 shown in Figures 1A and 1B. That is, the first data includes data such as information about the light-emitting device and characteristics of the light-emitting device.

[0046] Step S002 is a process of creating a training dataset from the first data. Step S002 is performed by the processing unit 102 shown in Figures 1A and 1B. The training dataset corresponds to the training dataset DS shown in Figures 1A and 1B.

[0047] Furthermore, in the case where the first data includes a molecular structure of an organic compound, step S002 includes a step of digitizing the molecular structure of the organic compound. The digitized molecular structure of the organic compound is included in the learning dataset.

[0048] Furthermore, when the first data includes a structure of an inorganic compound, step S002 may include a step of digitizing the structure of the inorganic compound. Furthermore, the digitized structure of the inorganic compound may be included in the learning dataset.

[0049] Step S003 is a step of performing supervised learning based on the learning data set. Step S003 is performed by the calculation unit 103 shown in FIG. 1A and FIG. 1B. It is preferable to use a neural network (particularly, deep learning) for the supervised learning. Note that the supervised learning may generate a trained model for predicting the characteristics of the light-emitting device.

[0050] Step S004 is a process of inputting second data. Step S004 is performed by the input unit 101 shown in FIG. 1A and FIG. 1B. The second data corresponds to the data IN2 shown in FIG. 1A and FIG. 1B. That is, the second data includes at least information related to the light-emitting device. Note that the second data may include data such as characteristics of the light-emitting device.

[0051] It should be noted that step S004 may be performed simultaneously with step S001, while steps S001 to S003 are being performed, or after steps S003 have been performed.

[0052] Step S005 is a process of creating data used for predicting characteristics from the second data. Step S005 is performed by the processing unit 102 shown in Figures 1A and 1B. That is, the data corresponds to the data DI used for predicting characteristics shown in Figures 1A and 1B.

[0053] Furthermore, in the case where the second data includes a molecular structure of an organic compound, step S005 includes a process of digitizing the molecular structure of the organic compound. The digitized molecular structure of the organic compound is included in the data.

[0054] In addition, when the second data includes a structure of an inorganic compound, step S005 may include a step of digitizing the structure of the inorganic compound. Furthermore, the digitized structure of the inorganic compound may be included in the data.

[0055] Step S006 is a process of inferring the characteristics of the light emitting device from the data based on the learning result of the supervised learning. Step S006 is performed by the calculation unit 103 shown in FIGS. 1A and 1B.

[0056] Step S007 is a step of outputting the third data. Step S007 is performed by the output unit 104 shown in Figures 1A and 1B. The third data corresponds to the data OUT shown in Figures 1A and 1B. That is, the third data includes the result of the above inference.

[0057] From the above, the characteristics of the light-emitting device can be predicted.

[0058] The method for predicting the characteristics of a light-emitting device is not limited to the above. For example, as shown in Fig. 3, the method for predicting the characteristics of a light-emitting device may include step S008 after step S003.

[0059] Step S008 is a step of storing the trained model generated in step S003. The trained model is stored in the storage unit 105 shown in FIG. 1B. By generating the trained model in advance, it is possible to omit steps S001 to S003 when predicting the characteristics of the light-emitting device. Therefore, it is possible to shorten the time required to predict the characteristics of the light-emitting device.

[0060] <<Neural Networks>> Here, we will explain neural networks that can be used for supervised learning.

[0061] As shown in FIG. 4A, the neural network NN can be configured with an input layer IL, an output layer OL, and a hidden layer HL. The input layer IL, the output layer OL, and the hidden layer HL each have one or more neurons (units). The hidden layer HL may be one layer or two or more layers. A neural network having two or more hidden layers HL can also be called a deep neural network (DNN). Learning using a deep neural network can also be called deep learning.

[0062] Input data is input to each neuron in the input layer IL. An output signal from a neuron in the previous or next layer is input to each neuron in the hidden layer HL. An output signal from a neuron in the previous layer is input to each neuron in the output layer OL. Each neuron may be connected to all neurons in the previous or next layer (fully connected), or may be connected to only a portion of the neurons.

[0063] FIG. 4B shows an example of a neuron operation. Here, neuron N and two neurons in the previous layer that output signals to neuron N are shown. Neuron N receives the output x1 of the previous layer neuron and the output x2 of the previous layer neuron. Neuron N then calculates the sum x1w1+x2w2 of the multiplication result (x1w1) of the output x1 and the weight w1 and the multiplication result (x2w2) of the output x2 and the weight w2, and then adds a bias b as necessary to obtain a value a=x1w1+x2w2+b. Then, the value a is transformed by the activation function h, and the output signal y=ah is output from neuron N. For example, a sigmoid function, a tanh function, a softmax function, a ReLU function, a threshold function, or the like can be used as the activation function h.

[0064] In this way, the calculations performed by neurons include the calculation of adding up the product of the output of a neuron in the previous layer and the weight, that is, the product-sum calculation (x1w1+x2w2 above). This product-sum calculation may be performed on software using a program, or it may be performed by hardware. When the product-sum calculation is performed by hardware, a product-sum calculation circuit can be used. This product-sum calculation circuit may be a digital circuit or an analog circuit. When an analog circuit is used for the product-sum calculation circuit, it is possible to reduce the circuit scale of the product-sum calculation circuit or to reduce the number of memory accesses, thereby improving processing speed and reducing power consumption.

[0065] The product-sum circuit may be formed using transistors including silicon (such as single crystal silicon) in a channel formation region (hereinafter also referred to as Si transistors) or transistors including an oxide semiconductor in a channel formation region (hereinafter also referred to as OS transistors). In particular, OS transistors have an extremely small off-state current and are therefore suitable as transistors that constitute the analog memory of the product-sum circuit. Note that the product-sum circuit may be formed using both Si transistors and OS transistors.

[0066] When the product-sum calculation is performed by hardware, the product-sum calculation circuit may be included in the calculation unit 103 of the characteristic prediction system 100 .

[0067] The above is a description of the neural network. In one embodiment of the present invention, it is preferable to use deep learning. In other words, it is preferable to use a neural network having two or more hidden layers HL.

[0068] The above is a description of one example of a method for predicting the characteristics of a light-emitting device.

[0069] <Details on how to predict the characteristics of light-emitting devices> The method for predicting the characteristics of a light emitting device will be described in detail below with reference to FIGS. 5 to 11B.

[0070] <<Light-emitting device structure>> First, the structure of the light emitting device will be described.

[0071] A light-emitting device is basically a device in which a layer containing a light-emitting material (also referred to as a light-emitting layer) is sandwiched between a pair of electrodes. By applying a voltage to the device, light can be emitted from the light-emitting material. An organic compound or an inorganic compound can be used as the light-emitting material. The light-emitting layer may be provided using one organic compound or one inorganic compound, may be provided by mixing or stacking a plurality of organic compounds, may be provided by mixing or stacking a plurality of inorganic compounds, or may be provided by mixing or stacking an organic compound and an inorganic compound. In the following, a light-emitting device using an organic compound as a light-emitting material will be described as an example.

[0072] Fig. 5 is a schematic diagram showing the structure of a light-emitting device 10. The light-emitting device 10 has a structure in which multiple layers are stacked. For example, the light-emitting device 10 has a layer 20 in which n layers (n is an integer of 3 or more) are stacked. In other words, the layer 20 is composed of layers 20(1) to 20(n). Note that a substrate is not shown in Fig. 5.

[0073] The light-emitting device has a structure in which a layer containing a light-emitting material is sandwiched between a pair of electrodes. Thus, the layer 20(1) functions as one of the anode and cathode of the light-emitting device 10, and the layer 20(n) functions as the other of the anode and cathode of the light-emitting device. The anode and / or cathode of the light-emitting device 10 are not limited to a single layer, and may be a laminated layer. In this case, one of the anode and cathode of the light-emitting device 10 is composed of layers 20(1) to 20(j) (j is an integer of 1 to (n-2) inclusive), and the other of the anode and cathode of the light-emitting device 10 is composed of layers 20(k) to 20(n) (k is an integer of (j+2) to n inclusive). Hereinafter, for ease of explanation, the anode and cathode of the light-emitting device 10 are both considered to be composed of a single layer.

[0074] When the anode and cathode of the light-emitting device 10 are both single layers, some or all of the layers 20(2) to 20(n-1) are layers containing an organic compound. Furthermore, any one or more of the layers 20(2) to 20(n-1) have a light-emitting material. Hereinafter, all of the layers 20(2) to 20(n-1) are considered to be layers containing an organic compound. Furthermore, the layers (layers 20(2) to 20(n-1)) located between the pair of electrodes may be collectively referred to as an intermediate layer 25.

[0075] When n is 3, intermediate layer 25 is composed of a single layer. That is, intermediate layer 25 is composed of only layer 20(2). The structure of light-emitting device 10 when n is 3 is sometimes called a single structure. When n is 4 or more, intermediate layer 25 is composed of multiple layers. That is, intermediate layer 25 has a laminated structure.

[0076] [Light-emitting device with a laminated intermediate layer] The following describes a light-emitting device 10 in which the intermediate layer has a laminated structure.

[0077] When n is 5, the intermediate layer 25 has a three-layer laminate structure. In this case, the intermediate layer 25 can have a laminate structure of a hole transport layer, a light emitting layer, and an electron transport layer. This can increase the current efficiency and external quantum efficiency of the light emitting device 10. When n is 6 or more, the intermediate layer 25 has a structure in which four or more layers are laminated. Therefore, any one or more of the hole transport layer, the light emitting layer, and the electron transport layer can have a laminate structure.

[0078] When layer 20(1) functions as an anode and layer 20(n) functions as a cathode, intermediate layer 25 has a structure in which a hole transport layer, a light emitting layer, and an electron transport layer are laminated in this order. Note that when layer 20(1) functions as a cathode and layer 20(n) functions as an anode, the order of lamination is reversed.

[0079] In addition, by appropriately adjusting the thickness of each of the anode, hole transport layer, light emitting layer, electron transport layer, and cathode, it is possible to improve the characteristics of the light emitting device, such as the luminous efficiency and external quantum efficiency. In other words, the characteristics of the light emitting device are affected by the thickness of each of the layers 20(1) to 20(n).

[0080] When n is 7, the intermediate layer 25 has a stacked structure of five layers. In this case, the intermediate layer 25 can have a stacked structure of a hole injection layer, a hole transport layer, a light emitting layer, an electron transport layer, and an electron injection layer. This can further increase the current efficiency and external quantum efficiency of the light emitting device 10. When n is 8 or more, the intermediate layer 25 has a structure in which six or more layers are stacked. Therefore, any one or more of the hole injection layer, the hole transport layer, the light emitting layer, the electron transport layer, and the electron injection layer can have a stacked structure.

[0081] When layer 20(1) functions as an anode and layer 20(n) functions as a cathode, intermediate layer 25 has a structure in which a hole injection layer, a hole transport layer, a light emitting layer, an electron transport layer, and an electron injection layer are laminated in this order. Note that when layer 20(1) functions as a cathode and layer 20(n) functions as an anode, the order of lamination is reversed.

[0082] The light-emitting layer may have a light-emitting material or a combination of a plurality of materials, and may be configured to obtain fluorescent or phosphorescent light of a desired emission color. The light-emitting layer may have a stacked structure with different emission colors. In this case, the light-emitting material and other materials used in each stacked light-emitting layer may be different from each other.

[0083] In light-emitting device 10, for example, by using layer 20(1) as a reflective electrode and layer 20(n) as a semi-transmissive and semi-reflective electrode to form a micro-optical resonator (microcavity) structure, the light emitted from the light-emitting layer included in intermediate layer 25 can be resonated between both electrodes, and the light emitted through layer 20(n) can be strengthened.

[0084] When the layer 20(1) of the light-emitting device 10 is a reflective electrode having a laminated structure of a conductive material having reflectivity and a conductive material having translucency (transparent conductive film), optical adjustment can be performed by controlling the film thickness of the transparent conductive film. Specifically, it is preferable to adjust the inter-electrode distance between the layer 20(1) and the layer 20(n) to be approximately m1λ / 2 (where m1 is a natural number) for the wavelength λ of the light obtained from the light-emitting layer.

[0085] In order to amplify the desired light (wavelength: λ) obtained from the light-emitting layer, it is preferable to adjust the optical distance from layer 20(1) to the region (light-emitting region) where the desired light of the light-emitting layer is obtained and the optical distance from layer 20(n) to the region (light-emitting region) where the desired light of the light-emitting layer is obtained to be close to (2m2+1)λ / 4 (where m2 is a natural number). Note that the light-emitting region here refers to the recombination region of holes and electrons in the light-emitting layer.

[0086] By carrying out such optical adjustment, it is possible to narrow the spectrum of a specific monochromatic light obtained from the light-emitting layer, thereby obtaining light emission with good color purity.

[0087] In the above case, the inter-electrode distance between the layer 20(1) and the layer 20(n) can be said to be the total thickness from the reflection area in the layer 20(1) to the reflection area in the layer 20(n), strictly speaking. However, since it is difficult to strictly determine the reflection area in the layer 20(1) or the layer 20(n), it is assumed that the above-mentioned effect can be sufficiently obtained by assuming any position in the layer 20(1) or the layer 20(n) as the reflection area. Furthermore, it can be said that the optical distance between the layer 20(1) or the layer 20(n) and the light-emitting layer from which the desired light is obtained is the optical distance between the reflection area in the layer 20(1) or the layer 20(n) and the light-emitting area in the light-emitting layer from which the desired light is obtained, strictly speaking. However, since it is difficult to strictly determine the reflection area in the layer 20(1) or the layer 20(n), and the light-emitting area in the light-emitting layer from which the desired light is obtained, it is assumed that the above-mentioned effect can be sufficiently obtained by assuming any position in the layer 20(1) or the layer 20(n) as the reflection area, and any position in the light-emitting layer from which the desired light is obtained as the light-emitting area.

[0088] In the above case, since the light-emitting device 10 has a microcavity structure, even if it has the same intermediate layer 25, it is possible to extract light of different wavelengths (monochromatic light). Therefore, when the light-emitting device 10 is used as a display element of a display device, it is not necessary to paint different colors (e.g., RGB) to obtain different luminescent colors. Therefore, it is easy to achieve high definition. It can also be combined with a colored layer. Furthermore, it is possible to increase the luminescent intensity of a specific wavelength in the front direction, which can reduce power consumption.

[0089] The light-emitting device 10 does not have to have a microcavity structure. In this case, the light-emitting layer may be structured to emit white light, and a colored layer may be provided to extract light of a desired color (e.g., RGB). When forming the intermediate layer 25, if different colors are applied to obtain different luminescent colors, light of a desired color may be extracted without providing a colored layer.

[0090] At least one of the layer 20(1) and the layer 20(n) can be an electrode having translucency (transparent electrode, semi-transmissive / semi-reflective electrode, etc.). When the electrode having translucency is a transparent electrode, the visible light transmittance of the transparent electrode is 40% or more. When the electrode having translucency is a semi-transmissive / semi-reflective electrode, the visible light reflectance of the semi-transmissive / semi-reflective electrode is 20% or more and 80% or less, preferably 40% or more and 70% or less. Furthermore, the resistivity of these electrodes is 1×10 -2 Ωcm or less is preferable.

[0091] When the layer 20(1) or the layer 20(n) is a reflective electrode (reflective electrode), the visible light reflectance of the reflective electrode is 40% or more and 100% or less, preferably 70% or more and 100% or less. The resistivity of this electrode is 1×10 -2 Ωcm or less is preferable.

[0092] When n is 5, the intermediate layer 25 has a three-layer laminate structure. In this case, the intermediate layer 25 can have a laminate structure of a first light-emitting layer, a charge generating layer, and a second light-emitting layer. That is, the light-emitting device 10 can have a tandem structure. By making the light-emitting device 10 have a tandem structure, the current efficiency and external quantum efficiency of the light-emitting device 10 can be increased.

[0093] When n is 6 or more, the intermediate layer 25 has a structure in which 4 or more layers are laminated. In this case, any one or more of the layer sandwiched between the layer 20(1) and the charge generating layer (sometimes referred to as the first intermediate layer), the charge generating layer, and the layer sandwiched between the charge generating layer and the layer 20(n) (sometimes referred to as the second intermediate layer) can have a laminated structure. For example, when n is 13, each of the first intermediate layer and the second intermediate layer can have the same configuration as the intermediate layer 25 having the above-mentioned 5-layer laminated structure.

[0094] The charge generation layer has the function of injecting electrons into one of the first intermediate layer and the second intermediate layer and injecting holes into the other when a voltage is applied between layer 20(1) and layer 20(n). Therefore, when a voltage is applied so that the potential of layer 20(1) is higher than the potential of layer 20(n), electrons are injected from the charge generation layer into the first intermediate layer, and holes are injected from the charge generation layer into the second intermediate layer.

[0095] From the viewpoint of light extraction efficiency, the charge generation layer is preferably transparent to visible light. Specifically, the charge generation layer preferably has a visible light transmittance of 40% or more. The conductivity of the charge generation layer may be lower than the conductivity of the layer 20(1) or the conductivity of the layer 20(n).

[0096] When n is 7, the intermediate layer 25 has a stacked structure of five layers. In this case, the intermediate layer 25 can have a stacked structure of a first light-emitting layer, a first charge generating layer, a second light-emitting layer, a second charge generating layer, and a third light-emitting layer. That is, the light-emitting device 10 can have a tandem structure. By making the light-emitting device 10 have a tandem structure, the current efficiency and external quantum efficiency of the light-emitting device 10 can be further increased.

[0097] When n is 8 or more, the intermediate layer 25 has a structure in which 6 or more layers are laminated. Any one or more of the layer sandwiched between the layer 20(1) and the first charge generation layer (sometimes referred to as the first intermediate layer), the first charge generation layer, the layer sandwiched between the first charge generation layer and the second charge generation layer (sometimes referred to as the second intermediate layer), the second charge generation layer, and the layer sandwiched between the second charge generation layer and the layer 20(n) (sometimes referred to as the third intermediate layer) can have a laminate structure. For example, when n is 19, each of the first intermediate layer, the second intermediate layer, and the third intermediate layer can have the same structure as the intermediate layer 25 having the above-mentioned five-layer laminate structure.

[0098] So far, the intermediate layer 25 has been described as having one charge generating layer and two intermediate layers, or two charge generating layers and three intermediate layers, but is not limited thereto. The intermediate layer 25 may have three or more charge generating layers and four or more intermediate layers. By increasing the number of charge generating layers and the number of intermediate layers, the current efficiency and external quantum efficiency of the light-emitting device 10 can be increased. Note that n may be appropriately adjusted according to the number of charge generating layers.

[0099] The above is a description of the light-emitting device 10 in which the intermediate layer has a laminated structure.

[0100] Note that the layer included in the intermediate layer 25 may be formed by co-evaporation of a plurality of organic compounds. The layer formed by co-evaporation contains a plurality of organic compounds. Thus, the intermediate layer 25 may contain a layer containing a plurality of organic compounds. When the intermediate layer 25 contains a layer containing a plurality of organic compounds, the concentration ratio of the plurality of organic compounds affects the characteristics of the light-emitting device.

[0101] The above is a description of the structure of the light emitting device.

[0102] <<Light Emitting Device Characteristics>> The characteristics of the light emitting device are described below. Examples of the characteristics of the light emitting device include the initial characteristics of the light emitting device, the results of a reliability test of the light emitting device (sometimes referred to as the reliability of the light emitting device), and the like.

[0103] Examples of the initial characteristics of a light-emitting device include luminance-current density characteristics, current efficiency-luminance characteristics, luminance-voltage characteristics, current-voltage characteristics, external quantum efficiency-luminance characteristics, chromaticity-luminance characteristics, and emission spectrum.

[0104] As a reliability test for a light-emitting device, for example, the initial luminance is set to a certain value, the light-emitting device is driven under a constant current density condition, and the change in luminance over driving time is measured. In this case, the luminance may be a normalized luminance with the initial luminance set to 100%. The smaller the decrease in luminance over driving time, the more reliable the light-emitting device is.

[0105] Data on characteristics of light-emitting devices, such as initial characteristics and reliability of light-emitting devices, are binary data and are often expressed as numerical values.

[0106] The above is a description of the characteristics of the light emitting device.

[0107] <<Training Data Set>> Here, we will explain the learning dataset used in supervised learning.

[0108] 6A and 6B are diagrams showing the configuration of a training data set 50. The training data set 50 has training data 51_1 to training data 51_m (m is an integer equal to or greater than 2). The training data 51_1 to training data 51_m have input data 52_1 to input data 52_m and teacher data 53_1 to teacher data 53_m, respectively. The training data 51_1 to training data 51_m include information on the light-emitting devices 10_1 to 10_m and data on the characteristics of the light-emitting devices 10_1 to 10_m, respectively.

[0109] The target of prediction in this embodiment is the characteristics of a light-emitting device. As described above, the characteristics of a light-emitting device are affected by the type of organic compound used in the intermediate layer 25, the type of conductive material used in the layers 20(1) and 20(n) that function as electrodes, the film thickness of each of the layers 20(1) to 20(n), and the concentration ratio of a plurality of organic compounds in a layer having the plurality of organic compounds. Therefore, it is preferable to include these data in the learning dataset.

[0110] 1A and 1B, the learning dataset 50 is generated by the processing unit 102. The data IN1 input to the processing unit 102 includes information about the above-mentioned light-emitting device, data on the characteristics of the above-mentioned light-emitting device, etc. Therefore, the learning dataset is generated by extracting and processing or converting the data included in the data IN1.

[0111] It is preferable that the data included in the training dataset used in supervised learning is quantified. Compared to a case where the training dataset includes non-numeric data, quantification of the data included in the training dataset can prevent the machine learning model from becoming complicated.

[0112] 6A, input data 52_1 to input data 52_m include information on light-emitting devices 10_1 to 10_m, respectively. Moreover, teacher data 53_1 to teacher data 53_m include data on the characteristics of light-emitting devices 10_1 to 10_m, respectively.

[0113] As described above, the data on the characteristics of the light-emitting devices 10_1 to 10_m are digitized data, and therefore can be included in the teacher data 53_1 to 53_m, respectively, without any special conversion. Note that characteristic points of the data on the characteristics of the light-emitting devices 10_1 to 10_m may be extracted and included in the teacher data 53_1 to 53_m, respectively. Also, points may be extracted so that the values ​​of the control variables are equally spaced, and included in the teacher data 53_1 to 53_m, respectively.

[0114] Here, information related to the light emitting devices 10_1 to 10_m will be described. First, information related to the light emitting device 10_1 included in the input data 52_1 will be described with reference to FIG.

[0115] Information on the light-emitting device includes the structure of the light-emitting device, the manufacturing conditions (process conditions) of the light-emitting device, etc. Information on the structure of the light-emitting device includes, for example, the film thickness of each layer, the material contained in each layer, the concentration ratio of the material contained in each layer, etc. Information on the manufacturing conditions of the light-emitting device includes, for example, the type (shape, etc.) of the deposition source, the deposition rate, the film formation temperature, the state (degree of vacuum, etc.) of the deposition chamber or the atmosphere to which the light-emitting device is exposed during the film formation process, the purity of the organic compound of the deposition source (type of impurities, etc.), the type of measurement device, the type of film formation device, the film formation method, etc.

[0116] As shown in FIG. 6C, for example, the input data 52_1 includes information on the structure of the light-emitting device 10_1 as information on the light-emitting device 10_1. Specifically, the information includes the film thicknesses (film thicknesses 22(1) to 22(n)) of the layers 20(1) to 20(n), one or more materials (materials 21(1) to 21(n)) of the layers 20(1) to 20(n), and the concentration ratios (concentration ratios 23(1) to 23(n)) of the layers 20(1) to 20(n). Hereinafter, the film thicknesses 22(1) to 22(n) may be collectively referred to as film thicknesses 22. Also, the materials 21(1) to 21(n) may be collectively referred to as material 21. Also, the concentration ratios 23(1) to 23(n) may be collectively referred to as concentration ratios 23.

[0117] The concentration ratio 23 included in the input data is preferably described as a weight ratio within the layer. For example, when material A and material B are mixed in a weight ratio of A:B=p:q (wt%) (p and q are each a real number equal to or greater than 0) to form one layer, the concentration ratio 23 included in the input data may be described as p:q. When one layer is composed of only material A, the concentration ratio 23 included in the input data may be described as 1:0 or 0:0. Also, assuming a case in which three materials are mixed, for example, the concentration ratio 23 included in the input data may be described as p:q:r (r is a real number equal to or greater than 0).

[0118] The concentration ratio 23 included in the input data is not limited to being described as a weight ratio, and may be described as a molar concentration ratio. The concentration ratio 23 included in the input data is not limited to being described as a concentration ratio within a layer, and may be described as a concentration ratio at the time of co-deposition. The concentration ratio 23 included in the input data is not limited to being described as a ratio such as p:q, and may be described as a ratio such as q / p.

[0119] It is preferable that the film thicknesses 22(1) to 22(n) use the same unit of film thickness. By using the same unit of film thickness, the amount of data included in the learning data set 50 can be reduced. This makes it possible to reduce the time spent on data transmission and reception, supervised learning, inference, and the like.

[0120] As described above, it is preferable that the data included in the learning data set used in supervised learning is digitized. Since the film thickness 22 and the concentration ratio 23 are input as numerical values, they can be included in the input data 52_1 without any special conversion.

[0121] On the other hand, in the data IN1, information on the organic compound contained in the light-emitting device is often input as non-numeric data such as a structural formula, a molecular structure written in SMILES (Simplified molecular input line entry specification syntax) notation (sometimes simply referred to as molecular structure), a name according to the compound nomenclature determined by IUPAC, etc. Therefore, it is preferable to convert information on the organic compound input as non-numeric data (for example, molecular structure) into a numerical value.

[0122] As a means of quantifying a molecular structure, for example, there is a method of replacing it with the physical properties of an organic compound represented by the molecular structure. Examples of the physical properties of an organic compound include emission spectrum, absorption spectrum, transmission spectrum, reflection spectrum, S1 level, T1 level, oxidation potential, reduction potential, HOMO level, LUMO level, glass transition point, melting point, crystallization temperature, carrier mobility, etc. Although these physical properties can be treated as numerical values, it is necessary to perform measurements and simulations, and therefore it takes a lot of effort to generate a learning dataset.

[0123] Therefore, in one embodiment of the present invention, the molecular structure specifying the material 21 is converted by a certain method. The certain method may be capable of expressing molecular similarity. Well-known methods for expressing molecular similarity include quantitative structure-activity relationship (QSAR), fingerprints, and graph structures. For example, it is preferable to digitize the molecular structure using a linear notation or a matrix notation. By digitizing the molecular structure specifying the material 21, it can be used as learning data.

[0124] When quantifying the structure of an inorganic compound, it is advisable to use descriptors such as the radial distribution function (RDF) and the orbital field matrix (OFM).

[0125] [Example of quantifying the molecular structure of an organic compound] Here, an example of digitization (mathematical formulation) of the molecular structure of an organic compound will be described.

[0126] When information about organic compounds is input as non-numerical data other than SMILES notation, it is preferable to first convert it to SMILES notation. SMILES notation is preferable for computer-handled data because it represents organic compounds as a continuous string of characters. In addition, SMILES notation and the fingerprinting method described below are both classified as linear notations, and are preferable because they are easily converted between each other.

[0127] The open source cheminformatics toolkit RDKit can be used to quantify molecular structures. RDKit can convert the SMILES representation of the input molecular structure into mathematical formula data (quantification) using the fingerprint method.

[0128] In fingerprinting, a molecular structure is represented by assigning a partial structure (fragment) of the molecular structure to each bit, as shown in Figure 7, and if the corresponding partial structure exists in the molecule, the bit is set to "1", and if not, "0". In other words, by using fingerprinting, it is possible to obtain a mathematical formula that extracts the characteristics of a molecular structure. Furthermore, the formula of a molecular structure represented by fingerprinting is generally several hundred to tens of thousands of bits long, which is an easy-to-handle size. Furthermore, by using fingerprinting to represent a molecular structure as a mathematical formula of 0s and 1s, it is possible to realize extremely high-speed calculation processing.

[0129] In addition, there are many types of fingerprinting methods (those that use different bit generation algorithms, those that take into account atom types, bond types, and aromaticity conditions, and those that dynamically generate bit lengths using hash functions, etc.), each with its own characteristics.

[0130] Examples of types of fingerprinting are shown in Figures 8A to 8D. Representative types of fingerprinting include the circular type shown in Figure 8A (atoms around the starting atom up to a specified radius are considered as a partial structure), the path-based type shown in Figure 8B (atoms from the starting atom up to a specified path length are considered as a partial structure), the substructure keys type shown in Figure 8C (partial structures are defined for each bit), and the atom pair type shown in Figure 8D (atom pairs generated for all atoms in a molecule are considered as a partial structure). The RDKit implements fingerprints of each of these types.

[0131] Figure 9 shows an example of the molecular structure of an organic compound actually expressed as a formula using the fingerprint method. In this way, the molecular structure can be first converted into a SMILES notation and then converted into a fingerprint.

[0132] In addition, when expressing the molecular structure of an organic compound by the fingerprint method, the formula obtained may be the same between different organic compounds having similar structures. As described above, there are several types of fingerprint methods depending on the notation method, but the tendency of compounds to become the same varies depending on the notation method, as shown in FIG. 10, Circular type (Morgan Fingerprint), Path-based type (RDK Fingerprint), Substructure keys type (Avalon Fingerprint), and Atom pair type (Hash atom pair). In FIG. 10, the molecules in each double arrow indicate the same formula (notation). Therefore, as the fingerprint method used for learning, it is preferable to use a fingerprint method in which the notation of each organic compound is different when the molecular structure of each organic compound to be learned is expressed using at least one of them. In FIG. 10, it can be seen that compounds with different atom pair types can be expressed without overlap, but depending on the population of organic compounds to be learned, other notation methods may also be possible without overlap.

[0133] Therefore, when describing the molecular structure of an organic compound by a fingerprint method, it is preferable to use a plurality of different types of fingerprint methods. Any number of types may be used, but about two or three types are preferable because they are easy to handle in terms of the amount of data. When using a plurality of types of fingerprint methods, the molecular structure of an organic compound may be described by connecting a formula represented by one type of fingerprint method to a formula represented by another type of fingerprint method, or may be described as having a plurality of different types of formulas for each organic compound. Figures 11A and 11B show an example of a method for describing the molecular structure of an organic compound using a plurality of different types of fingerprints.

[0134] The fingerprint method describes the presence or absence of a partial structure, and information on the entire molecular structure is lost. However, if a molecular structure is expressed mathematically using multiple fingerprints of different types, different partial structures are generated for each fingerprint type, and information on the presence or absence of these partial structures can be used to supplement information on the entire molecular structure. If a feature that cannot be expressed by a certain fingerprint has a significant effect on the characteristics of a light-emitting device, it can be supplemented by other fingerprints, so the method of describing a molecular structure using multiple fingerprints of different types is effective.

[0135] As shown in FIG. 11A, when representation is performed using two types of fingerprint methods, it is preferable to use the Atom Pair type and the Circular type because this enables accurate prediction of physical properties.

[0136] Furthermore, as shown in FIG. 11B, when using three types of fingerprinting methods for representation, it is preferable to use the Atom Pair type, the Circular type, and the Substructure keys type, because these enable accurate prediction of physical properties.

[0137] Furthermore, when a circular fingerprint method is used, the radius r is preferably 3 or more, and more preferably 5 or more. Note that the radius r is the number of atoms counted in connection from a certain atom that is the starting point, which is set as 0.

[0138] As described above, when selecting a fingerprint method to be used, it is preferable to select at least one method in which the molecular structures of the organic compounds are all different from one another.

[0139] By increasing the bit length (number of bits) of a fingerprint, the possibility of generating a description in which the notation is completely identical between each organic compound can be reduced, but if the bit length is made too large, a trade-off occurs in that the calculation cost and database management cost increase. On the other hand, by expressing multiple fingerprints simultaneously, even if there are multiple molecular structures whose notations are completely identical in a certain fingerprint type, by combining different fingerprint types, there is a possibility that the notations will not be completely identical as a whole. As a result, a state in which multiple organic compounds whose notations are completely identical by fingerprints can be generated with as small a bit length as possible. There is no particular limit to the bit length of the fingerprint to be generated, but considering the calculation cost and the database management cost, if the molecular weight of each molecule is up to about 2000, a fingerprint in which the notation is not completely identical between each organic compound can be generated with a bit length of 4096 or less, preferably 2048 or less, and in some cases even 1024 or less for each fingerprint type.

[0140] In addition, the bit length of the fingerprints generated for each fingerprint type can be adjusted appropriately taking into account the characteristics of the type and the overall molecular structure, and does not need to be unified. For example, the bit length can be expressed as 1024 bits for the Atom Pair type and 2048 bits for the Circular type, and then these can be concatenated.

[0141] The above is an explanation of the quantification of the molecular structure of an organic compound.

[0142] In this manner, the learning data 51_1 including digitized data can be generated. The learning data 51_2 to 51_m including digitized data have the same configuration as the learning data 51_1.

[0143] When information about a light-emitting device consisting of k layers (k is an integer greater than or equal to 3 and less than or equal to (n-1)) is to be included in the learning data, it is preferable to input zeros (zero-pad) for material 21(k+1) through material 21(n), film thickness 22(k+1) through film thickness 22(n), and concentration ratio 23(k+1) through concentration ratio 23(n) in the learning data.

[0144] In the training data set, it is preferable to specify the value of n. By fixing the value of n, it is possible to determine the number of neurons (units) that the input layer IL has.

[0145] n is preferably 5 or more, and more preferably 7 or more. There is no particular limit to the upper limit of n, but if the value of n is too large, the number of neurons in the input layer IL increases, and there is a risk of increasing the time spent on supervised learning and inference. Therefore, for example, n should be set to 30 or less.

[0146] Furthermore, each of the input data 52_1 to 52_m may not need to include information about the layer 20(1) and the layer 20(n), which are the anode and cathode of the light-emitting device 10. In this case, the amount of data included in the learning dataset 50 is reduced. This allows the time spent on data transmission and reception, supervised learning, and inference to be reduced.

[0147] In Fig. 6A, the input data 52_1 to 52_m include information on the light-emitting devices 10_1 to 10_m, respectively, but are not limited to this. For example, as shown in Fig. 6B, the input data 52_1 to 52_m may include information on the light-emitting devices 10_1 to 10_m, respectively, and data on first characteristics of the light-emitting devices 10_1 to 10_m, and the teacher data 53_1 to 53_m may include data on second characteristics of the light-emitting devices 10_1 to 10_m, respectively.

[0148] In the above case, the first characteristic of the light emitting device 10_1 to the light emitting device 10_m is different from the second characteristic of the light emitting device 10_1 to the light emitting device 10_m. For example, the initial characteristic of the light emitting device 10 may be used as the first characteristic of the light emitting device 10, and the reliability of the light emitting device 10 may be used as the second characteristic of the light emitting device 10. The reliability of the light emitting device is suitable as a subject for inference because there are many factors that affect the reliability and each factor is complicatedly intertwined, making it difficult to predict based on experience. In addition, the initial characteristic of the light emitting device 10 indirectly includes information such as the manufacturing conditions and measurement conditions of the light emitting device. Therefore, by adding the first characteristic of the light emitting device 10 to the input data, the information is provided to supervised learning, and the accuracy of prediction of the reliability of the light emitting device can be improved.

[0149] 6C shows a case where the input data 52_1 to 52_m include information on the structures of the light-emitting devices 10_1 to 10_m, respectively, but is not limited thereto. For example, as shown in FIG. 6D, the input data 52_1 to 52_m may include information on the structures of the light-emitting devices 10_1 to 10_m, respectively, and information on the manufacturing conditions of the light-emitting devices. For example, the input data 52_1 may include the material 21(1), the film thickness 22(1), the concentration ratio 23(1), the deposition rate 31(1) of the material 21(1), the film formation temperature 32(1) of the material 21(1), etc.

[0150] The learning dataset 50 may be composed of only data of light-emitting devices having the same or similar emission color. In other words, the learning dataset 50 may be created for each emission color. This can improve the accuracy of prediction of the characteristics of the light-emitting device. The learning dataset 50 may also be composed of data of light-emitting devices regardless of the emission color. This can enable highly versatile prediction of the characteristics of the light-emitting device.

[0151] The above is an explanation of the learning dataset. The characteristics of a light-emitting device can be predicted by training a machine learning model using input data and teacher data. When predicting the initial characteristics of a light-emitting device, it is preferable to include information on the structure of the light-emitting device in the input data and provide the initial characteristics of the light-emitting device as teacher data.

[0152] <<Data used to predict characteristics>> Here, the data used to predict the characteristics will be described.

[0153] The data used to predict the characteristics is generated by the processing unit 102 shown in Figures 1A and 1B. The data IN2 input to the processing unit 102 includes at least information related to the structure of the light-emitting device. The data IN2 may also include data such as the characteristics of the light-emitting device.

[0154] The data used for predicting the characteristics may have the same configuration as the input data of the learning data described above. For example, when the input data of the learning data includes information on the structure of the light-emitting device, the data used for predicting the characteristics may include information on the structure of the light-emitting device. Also, for example, when the input data of the learning data includes information on the structure of the light-emitting device and data on the characteristics of the light-emitting device, the data used for predicting the characteristics may include information on the structure of the light-emitting device and data on the characteristics of the light-emitting device.

[0155] Specific data used to predict the characteristics include film thicknesses 22(1) to 22(n), materials 21(1) to 21(n), and concentration ratios 23(1) to 23(n).

[0156] The above is a description of the data used to predict the characteristics.

[0157] As described above, one aspect of the present invention can provide a method for predicting characteristics of a light-emitting device, and a characteristic prediction system for predicting characteristics of a light-emitting device.

[0158] According to one embodiment of the present invention, the characteristics of a light-emitting device can be predicted without using the physical properties of an organic compound contained in the light-emitting device. In addition, by using past experimental data, the optimization of the structure of a light-emitting device can be accelerated by virtual screening. Even if a person browses the data and finds that it is not interpolative, it may become interpolative due to the nonlinear or high-order expression of the machine learning model. In addition, by cutting out and examining the expression obtained by the machine learning model in fragments, it is possible to learn rules that were not noticed in the past.

[0159] This embodiment can be carried out by combining parts thereof as appropriate. [Explanation of symbols]

[0160] DI: data, DS: learning dataset, HL: hidden layer, IL: input layer, IN1: data, IN2: data, OL: output layer, OUT: data, 10: light-emitting device, 10_1: light-emitting device, 10_m: light-emitting device, 20: layer, 21: material, 22: film thickness, 23: concentration ratio, 25: intermediate layer, 31: deposition rate, 32: deposition temperature, 50: learning dataset, 51_1: learning data, 51_2: learning data, 51_m: learning data, 52_1: input data, 52_m: input data, 53_1: teacher data, 53_m: teacher data, 100: characteristic prediction system, 101: input unit, 102: processing unit, 103: calculation unit, 104: output unit, 105: memory unit

Claims

[Claim 1] A characteristic prediction system for predicting characteristics of a light-emitting device having a layer including an organic compound, comprising: The apparatus includes an input unit, a processing unit, a calculation unit, and an output unit, the input unit has a function of supplying a structure of the light-emitting device or a characteristic of the light-emitting device; The processing unit has a function of generating a learning data set or data used for predicting characteristics, and a function of quantifying a molecular structure of an organic compound; The calculation unit has a function of performing supervised learning based on the learning data set, and a function of inferring characteristics of a light-emitting device from the data based on a learning result of the supervised learning, The output unit has a function of providing a result of the inference. Trait prediction system.

Images