Property Prediction System

Abstract

A system for predicting characteristics of a semiconductor device and a method for predicting characteristics of a semiconductor device are provided. [Solution] The semiconductor element has a memory unit, an input unit, a processing unit, and an arithmetic unit, and the processing unit has a function of creating a training dataset from first data stored in the memory unit, a function of creating prediction data from second data supplied from the input unit, a function of converting qualitative data (material name or composition formula) into quantitative data (element properties and composition), and a function of extracting or removing from the first data and the second data, the first data includes process lists for the first to mth (m is an integer of 2 or more) semiconductor elements and the properties of the first to mth semiconductor elements, and the second data includes a process list for the m+1th semiconductor element, and the arithmetic unit has a function of performing supervised learning and inference, thereby performing learning based on the training dataset and inferring the properties of the semiconductor element from the prediction data.

Description

[Technical Field]

[0001] One aspect of the present invention relates to a system for predicting characteristics of a semiconductor device. , relates to a method for predicting the characteristics of a semiconductor device.

[0002] In this specification and the like, a semiconductor element is a device that can function by utilizing semiconductor characteristics. This refers to an element, such as a transistor, a diode, a light-emitting element, or a light-receiving element. Another example of the semiconductor element is a conductive element such as a capacitor, a resistor, or an inductor. A passive element is formed by a conductive film or an insulating film. is a semiconductor device that includes a circuit having a semiconductor element or a passive element. [Background technology]

[0003] In recent years, artificial intelligence (AI) has In the fields of electronics, robotics, and energy, where high power is used, such as power ICs, Development of new semiconductor elements to solve problems such as increasing computational complexity or power consumption The integrated circuits and semiconductor elements used in integrated circuits that the market demands are complex. While the process is becoming more complex, there is a demand for the early launch of integrated circuits with new functions. In the development of semiconductor devices, process design, device design, or circuit design, The knowledge, know-how, and experience of the engineers are required.

[0004] Recently, in the field of semiconductor devices, methods for optimizing manufacturing processes, devices, In Patent Document 1, a method for estimating the cross-sectional shape of a semiconductor device is proposed. Image feature values ​​are calculated from the SEM image of the pattern, and the correlation between the image feature values ​​and the device characteristics is calculated. A method for estimating device characteristics of a pattern to be evaluated from the correlation is disclosed. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Laid-Open No. 2007-129059 Summary of the Invention [Problem to be solved by the invention]

[0006] In the manufacturing process of semiconductor devices, there are many steps to complete the semiconductor device, and the types of steps The types and processing conditions are also diverse. Semiconductor devices are manufactured through a lengthy process, and their electrical characteristics are The characteristics of semiconductor devices, such as the performance and reliability test results, are measured using measuring equipment. Verifying the causal relationship between the manufacturing process of elements and the characteristics of semiconductor elements one by one through experiments This improves the characteristics of the semiconductor element.

[0007] However, it is necessary to comprehensively adjust the manufacturing process of semiconductor elements and to identify the causes of the characteristics of semiconductor elements. It takes time and money to investigate the correlation between results and outcomes. This is difficult, and therefore requires a great deal of effort to optimize the manufacturing process through experiments.

[0008] Therefore, one aspect of the present invention aims to provide a system for predicting characteristics of a semiconductor element. Another aspect of the present invention is to provide a method for predicting characteristics of a semiconductor element. Another embodiment of the present invention is a method for generating learning data for predicting characteristics of a semiconductor element. One of the objectives is to provide a set of

[0009] The description of these problems does not preclude the existence of other problems. It is not necessary for one embodiment to solve all of these problems. The subject matter will be self-evident from the description, drawings, claims, etc. It is possible to extract other issues from the drawings, claims, etc. [Means for solving the problem]

[0010] One aspect of the present invention is to perform supervised learning based on a training dataset, Based on the results, the characteristics of semiconductor elements are inferred from the prediction data. The semiconductor device characteristic prediction system includes a storage unit, an input unit, a processing unit, and a calculation unit, and a processing unit that calculates a learning dataset from the first data stored in the storage unit. and a function to create prediction data from the second data supplied from the input unit. and the ability to convert qualitative data into quantitative data and the ability to convert the first data and the second data and a function of extracting or removing the first data from the first semiconductor element through the second semiconductor element. A process list for the mth (m is an integer of 2 or more) semiconductor device, and the first to second semiconductor devices The second data includes the characteristics of the mth semiconductor device, and the second data includes the process list for the m+1th semiconductor device. Qualitative data is the name or formula of the material, and quantitative data is the properties of the elements, and composition, and the calculation unit has the function of performing supervised learning and inference.

[0011] In the semiconductor device characteristic prediction system, the element characteristics are determined based on the atomic number, group, period, and charge. molecular configuration, atomic mass, atomic radius (covalent radius, van der Waals radius, ionic radius, or is the metallic bond radius), atomic volume, electronegativity, ionization energy, electron affinity, dipole polarization The melting point of the element, the boiling point of the element, the lattice constant of the element, the density of the element, and the thermal conductivity of the element It is preferable to have one or more of them.

[0012] In the semiconductor device characteristic prediction system, the characteristics of the semiconductor device are predicted by a reliability test. (+GBT stress test, +DBT stress test, -GBT stress test, +DGBT stress test ΔV obtained from stress test, +BGBT stress test, or -BGBT stress test It is preferable that the change in sh over time is the change in sh. In this case, the characteristics of the semiconductor element are preferably Id-Vg characteristics or Id-Vd characteristics. stomach.

[0013] In the semiconductor device characteristic prediction system, the processing unit converts the qualitative data into La It is preferable to have a function for converting data into numbers using bel encoding. [Effects of the Invention]

[0014] According to one aspect of the present invention, a system for predicting characteristics of a semiconductor device can be provided. According to one aspect of the present invention, a method for predicting the characteristics of a semiconductor device can be provided. In addition, according to one aspect of the present invention, a training data set for predicting the characteristics of a semiconductor device is provided. It can be provided.

[0015] The effects of one embodiment of the present invention are not limited to the effects listed above. This does not preclude the existence of other effects. The effects not mentioned in this section are obvious to those skilled in the art. It can be derived from the descriptions in the specifications, drawings, etc., and can be extracted appropriately from these descriptions. It should be noted that one aspect of the present invention has at least one of the effects listed above and / or other effects. Therefore, one aspect of the present invention is, in some cases, There are cases where the effects listed above are not achieved. [Brief explanation of the drawings]

[0016] [Figure 1] 1A and 1B are diagrams illustrating an example of a system for predicting characteristics of a semiconductor device. [Figure 2] FIG. 2 is a flowchart showing an example of a method for predicting characteristics of a semiconductor device. [Figure 3] 3A and 3B are diagrams illustrating the configuration of a neural network. [Figure 4] 4A and 4B are diagrams illustrating the training data set. [Figure 5] Fig. 5A is a diagram illustrating the results obtained in a reliability test of a semiconductor element, and Fig. 5B is a diagram illustrating the Id-Vg characteristics of the semiconductor element. [Figure 6] 6A to 6C are diagrams for explaining a method for creating input data. [Figure 7] 7A and 7B are diagrams for explaining a method for creating input data. [Figure 8] FIG. 8 is a diagram illustrating a computer device. DETAILED DESCRIPTION OF THE INVENTION

[0017] The embodiments will be described in detail with reference to the drawings. However, the present invention is not limited to the following description. The present invention is not limited to the above, and various changes and modifications may be made in form and detail without departing from the spirit and scope of the present invention. It will be readily apparent to those skilled in the art that the present invention can be modified in various ways. The present invention is not to be construed as being limited to the description in the form of

[0018] In the configuration of the invention described below, the same parts or parts having similar functions The same reference numerals are used in common between different drawings, and repeated explanations will be omitted. When referring to similar functions, the hatch pattern may be the same and no particular reference numeral may be given.

[0019] In addition, the position, size, range, etc. of each component shown in the drawings are not necessarily the same as those in the actual embodiment for ease of understanding. Therefore, the disclosed invention may not necessarily represent the actual position, size, range, etc. The position, size, range, etc. are not necessarily limited to those disclosed in the drawings.

[0020] In addition, the ordinal numbers "first," "second," and "third" used in this specification and elsewhere refer to components. It should be noted that the number is added to avoid confusion and is not intended to limit the number.

[0021] (Embodiment 1) In this embodiment, a characteristic prediction system for a semiconductor element and a semiconductor element according to one embodiment of the present invention will be described. A method for predicting the characteristics of a child will be described with reference to FIGS. 1A to 8. FIG.

[0022] The semiconductor device characteristic prediction system according to one aspect of the present invention predicts the characteristics of a semiconductor device from information about the semiconductor device. The system is capable of predicting the characteristics of the semiconductor element. The method for predicting the characteristics of a semiconductor element uses machine learning. is.

[0023] <Semiconductor device characteristic prediction system> FIG. 1A is a diagram showing the configuration of a characteristic prediction system 100. That is, FIG. 1A shows the configuration of the characteristic prediction system 100 according to the present invention. It can also be said that this is an example of the configuration of a semiconductor element characteristic prediction system, which is one aspect of the present invention.

[0024] The characteristic prediction system 100 is an information processing system such as a personal computer used by a user. Alternatively, the processing unit of the characteristic prediction system 100 may be provided in a server. The system may also be configured to be accessed and used from a client PC via a network.

[0025] As shown in FIG. 1A, the characteristic prediction system 100 includes an input unit 101, a processing unit 102, and a calculation unit. The apparatus includes an input unit 101, a processing unit 102, an output unit 104, and a storage unit 105. The calculation unit 103, the output unit 104, and the storage unit 105 are connected to each other via a transmission line. It may also be used.

[0026] The storage unit 105 stores data of information relating to each of the plurality of semiconductor elements. The information on semiconductor devices includes, for example, a process list for semiconductor devices, characteristics of semiconductor devices, This includes information about the shape of the semiconductor element. It may also be referred to simply as the process list for semiconductor devices. The characteristics of semiconductor elements may be expressed simply as the properties of the semiconductor element. The information data may be simply referred to as information relating to the shape of the semiconductor element.

[0027] In the process list for semiconductor devices, multiple processes are set in the order of the manufacturing process of semiconductor devices, and each process Processing conditions are specified for the process.

[0028] The characteristics of semiconductor elements are measured using a measuring device. The data on the characteristics of the semiconductor element may include, for example, the results of the semiconductor element characteristics test, the reliability test, etc. These include measurement data of electrical characteristics and data obtained by carrying out reliability tests.

[0029] The information about the shape of a semiconductor element includes the position, size, and range of the components of the semiconductor element. The data on the shape of the semiconductor element may include, for example, the positions of the components of the semiconductor element. Numerical data representing the size, range, etc., and image data of semiconductor elements and their surroundings. Specifically, measurement data such as channel length and channel width, and scanning electron microscope (SEM) EM: Scanning Electron Microscope) observation image, transmission Transmission Electron Microscopy (TEM) These include images observed using a microscope (cope).

[0030] The storage unit 105 stores at least a process list and characteristics of each of the plurality of semiconductor elements. The process list for semiconductor devices stored in the storage unit 105 includes the following: It is preferable that ID D is assigned to the process list of the semiconductor device. is expressed as a process list ID. The characteristics of the semiconductor device stored in the storage unit 105 are as follows: It is associated with the process list ID. In other words, it is used to read and write the characteristics of semiconductor elements. etc. may be done based on the process list ID.

[0031] The storage unit 105 may store information relating to the shape of each of the plurality of semiconductor elements. At this time, the information about the shape of the semiconductor element stored in the storage unit 105 is stored in the process list. It is preferable that the information about the shape of the semiconductor element is associated with the ID. Importing, exporting, etc. may be done based on the process list ID.

[0032] The process list for a plurality of semiconductor elements and the characteristics of a plurality of semiconductor elements are input to an input unit 101, a recording unit 102, a The information is stored in the storage unit 105 via a storage medium, communication, etc. The information about the above is also stored in the storage unit 105 via the input unit 101, a storage medium, communication, etc. It would be good to do so.

[0033] The process list for multiple semiconductor devices and the characteristics of multiple semiconductor devices are stored as text data. In particular, the characteristics of the plurality of semiconductor elements are preferably stored in the storage unit 105 as numerical values. It is preferable that the data be stored in the storage unit 105 as binary data or binary data. In the field of bivariate data, bivariate data refers to a collection of data related to two variables. The data is obtained by extracting data on two variables from multivariate data of three or more variables. It may also be a collection of.

[0034] The process list for the plurality of semiconductor elements and the characteristics of the plurality of semiconductor elements are image data. In this case, the image data may be stored in the storage unit 105 as image data, but the image data may be stored in the text data. It is preferable that the text data is converted into data and then stored in the storage unit 105. The size is smaller than the data size of the image data, so the image data is converted to text data. By storing the converted data in the storage unit 105, the load on the storage unit 105 can be reduced. can.

[0035] The property prediction system 100 may have optical character recognition (OCR) capabilities. This allows you to recognize characters contained in image data and create text data. For example, the processing unit 102 may have this function. , and may further include a character recognition unit having this function.

[0036] The storage unit 105 may have a function of storing a trained model (also called an inference model). stomach.

[0037] The input unit 101 has a function for the user to input data IN2. The input unit 101 is a keyboard, a mouse, Input devices include sensors, touch sensors, scanners, and cameras. , may be stored in the storage unit 105.

[0038] In addition, when the data IN2 is image data, the characteristic prediction system 100 uses the above-mentioned OCR function. This function allows the recognition of characters contained in image data and the creation of text data. For example, if the processing unit 102 has the OCR function, the data IN2 can be Alternatively, the OCR function may be implemented in the processing unit 1 of the characteristic prediction system 100. If it is provided with a data other than 02, the text data converted from the image data is treated as data IN2. You may do so.

[0039] The processing unit 102 extracts a learning data set from the data IN1 supplied from the storage unit 105. It has the function to generate DS. The training dataset DS is training data for supervised learning. The processing unit 102 also calculates the following from the data IN2 supplied from the input unit 101: It has a function to generate the prediction data DI. The prediction data DI is used to predict the characteristics of the semiconductor element. This is data for measuring

[0040] The data IN1 is a data group used when creating the learning dataset DS. The data group includes data relating to each of the plurality of semiconductor elements stored in the storage unit 105. Contains some or all of the information.

[0041] Here, some or all of the plurality of semiconductor elements are designated as semiconductor element 30_1 to semiconductor element 30_3. 0_m (m is an integer of 2 or more). The process lists of the body element 30_m are designated as process lists 10_1 to 10_m, respectively. In addition, the semiconductor elements 30_1 to 30_m are actually measured using a measuring device. The characteristics are respectively designated as characteristics 20_1 to 20_m. 20_m are produced according to process lists 10_1 to 10_m, respectively. These are characteristics that are actually measured using a measuring device for a semiconductor element.

[0042] Hereinafter, the process list 10_1 to the process list 10_m are collectively referred to as a plurality of process lists. In addition, the characteristics 20_1 to 20_m may be collectively referred to as multiple characteristics. The semiconductor elements 30_1 to 30_m may be collectively referred to as a semiconductor element 20. For example, it may be referred to as a plurality of semiconductor elements 30.

[0043] The data IN1 includes, for example, data of process list 10_1 to process list 10_m, etc. In addition, data of characteristics 20_1 to 20_m is included. Associated with each process list ID of process list 10_1 to process list 10_m The process list may include data on the shape of the semiconductor element being used. The data of the process list 10_1 to the process list 10_m is simply The data of the characteristics 20_1 to 20_m may be written simply as , and may be expressed as characteristic 20_1 to characteristic 20_m.

[0044] The data IN2 is information about a semiconductor device that is designated by a user for predicting the characteristics of the semiconductor device. The data IN2 includes, for example, a process list designated for predicting the characteristics of a semiconductor device. The process list designated for predicting the characteristics of semiconductor devices is referred to as Process List 11. In addition, data IN2 contains the semi-finished product associated with the process list ID of the process list 11. Information regarding the shape of the conductive element may also be included.

[0045] The processing unit 102 also processes qualitative data (qualitative data, categorical data, categorical data, In other words, the processing unit 102 has a function of converting qualitative data into numerical values. Has the ability to convert data into quantitative data (also known as quantitative data, quantity data, etc.) For example, the processing unit 102 may include Label Encoding, One-hot Encoding, It is preferable that coding, target encoding, etc. be implemented.

[0046] Qualitative data is included in data IN1 and data IN2. For example, there are equipment data, material data, etc. The data and the quantification of qualitative data about the materials are discussed below.

[0047] The calculation unit 103 has a function of performing machine learning. For example, the calculation unit 103 uses learning data The calculation unit 103 has a function of performing supervised learning based on the dataset DS. Based on the learning results of the supervised learning, the characteristics of the semiconductor element are calculated from the prediction data DI. By performing supervised learning as machine learning, This can improve the accuracy of inferring the characteristics of the child. You can generate a trained model using

[0048] The supervised learning mentioned above uses neural networks (especially deep learning). As deep learning, for example, convolutional neural networks are used. Convolutional Neural Network (CNN), recurrent neural network Neural Network (RNN: Recurrent Neural Network) , Autoencoder (AE), Variational Autoencoder (VAE) :Variational Autoencoder, Random Forest om Forest), Support Vector Machine chine, Gradient Boosting, Adversarial Learning Generative Adversarial Networks (GANs) It is preferable to use a .rks.

[0049] The output of the calculation unit 103 is used as the characteristic of the semiconductor element. The output is the characteristic of the semiconductor device. The actual measured value is used as the output of the neural network. After the machine learning model is trained, the process list for any semiconductor device is trained using neural network. By inputting data into the network, the characteristics of semiconductor elements can be predicted.

[0050] In addition, in a neural network, a product-sum operation is performed. When the calculation is performed by hardware, it is preferable that the calculation unit 103 has a product-sum calculation circuit. The product-sum operation circuit may be a digital circuit or an analog circuit. The multiplication and accumulation operation may be performed on software using a program.

[0051] The calculation unit 103 may have a function of performing semi-supervised learning as machine learning. The data includes the characteristics of semiconductor elements as training data (also called training signals or correct labels). However, to prepare the training data, it is necessary to actually fabricate a semiconductor device and Compared to supervised learning, semi-supervised learning requires the training dataset to be measured. The amount of training data included in the set can be small, so the time spent on creating training data is reduced. This allows for inference to be performed while shortening the time required.

[0052] The output unit 104 has a function of supplying information. The information is a predicted result of the characteristics of the semiconductor device or information relating to the predicted result. For example, the information is supplied as visual information such as character strings, numerical values, graphs, etc. The characteristic prediction system 100 includes an output device such as a display. 104 may not be provided.

[0053] The above constitutes a semiconductor device characteristic prediction system.

[0054] The characteristic prediction system 100 is not limited to the above configuration. For example, as shown in FIG. As described above, the characteristic prediction system 100 includes an input unit 101, a processing unit 102, a calculation unit 103, an output unit In addition to the memory unit 104 and the storage unit 105, a storage unit 106 may be provided.

[0055] The storage unit 106 has a function of storing the trained model generated by the calculation unit 103. Since the gender prediction system 100 has the storage unit 106, based on the trained model, Therefore, the trained model can be generated in advance and the characteristics of the semiconductor device can be predicted. By doing so, there is no need to perform supervised learning when predicting the characteristics of semiconductor devices. Therefore, the time required to predict the characteristics of a semiconductor device can be reduced. Cut.

[0056] The storage unit 106 is connected to the calculation unit 103 via a transmission line. 6 is connected to an input unit 101, a processing unit 102, an output unit 104, and a storage unit 106 via a transmission path. 5 may be connected to each of the

[0057] The storage unit 106 may be provided in the memory unit 105. It may also serve as the storage section 106 .

[0058] The above is a description of the configuration of the characteristic prediction system 100. By using the semiconductor element characteristic prediction system, it is possible to predict the characteristics of the semiconductor element from the information on the semiconductor element. For example, the characteristics of a semiconductor device can be predicted from the process list of the semiconductor device. For example, from the process list for semiconductor devices, It is possible to extract processes that have a large contribution to the characteristics of the child.

[0059] <Method for predicting semiconductor device characteristics> FIG. 2 is a flowchart illustrating the flow of processing executed by the characteristic prediction system 100. That is, FIG. 2 shows an example of a method for predicting the characteristics of a semiconductor device, which is an embodiment of the present invention. It can also be said to be a flowchart showing the

[0060] The method for predicting the characteristics of a semiconductor device includes steps S001 to S007. Steps S001 to S003 are steps related to supervised learning. Steps S004 to S007 are steps related to inference in supervised learning.

[0061] Step S001 is a step of inputting first data into the processing unit 102. The data corresponds to the data IN1 described above. That is, the first data includes the semiconductor element Specifically, the first device The data includes process lists 10_1 to 10_m and characteristics 20_1 to 20_m. The first data includes the process list 10_1 to the process list 20_m. 10_m Information on the shape of semiconductor elements associated with each process list ID Information may be included.

[0062] Step S002 is a step of creating a learning data set from the first data. Step S002 is performed by the processing unit 102 shown in FIGS. 1A and 1B. The training data set corresponds to the training data set DS described above.

[0063] Step S002 is a step of quantifying the qualitative data included in the first data. The qualitative data includes, for example, qualitative data about the device, qualitative data about the material, etc. The data obtained by digitization is included in the training dataset. It can be enjoyed.

[0064] Step S003 performs supervised learning based on the learning data set. Step S003 is performed by the calculation unit 103 shown in FIGS. 1A and 1B. The supervised learning algorithm (also called the learning method) is a neural network ( In particular, it is preferable to use deep learning. This may generate a trained model for predicting the characteristics of a semiconductor device.

[0065] Step S004 is a step of inputting the second data into the processing unit 102. The data corresponds to the above-mentioned data IN2. That is, the second data contains It contains information about semiconductor devices that are specified for predicting the characteristics of the semiconductor devices. The second data includes a process list 11.

[0066] When the semiconductor device is manufactured according to the process list 11, the second data includes , information about the shape of the semiconductor element, characteristics of the semiconductor element, etc. .

[0067] It is preferable that step S004 is performed after steps S003 have been performed. It is preferable that the step S001 is performed simultaneously with the step S001, or the step S001 to the step S002 are performed simultaneously. This may be performed during the execution of S003.

[0068] Step S005 creates prediction data for the characteristics of the semiconductor element from the second data. Step S005 is performed by the processing unit 102 shown in FIGS. 1A and 1B. In other words, the prediction data of the characteristics of the semiconductor element is compared with the above-mentioned prediction data DI. Respond.

[0069] Step S005 is a step of quantifying the qualitative data included in the second data. The qualitative data includes, for example, qualitative data about the device, qualitative data about the material, etc. The data obtained by quantifying it is useful for predicting the characteristics of the semiconductor device. Included in the data.

[0070] It is preferable that step S005 is performed after steps S003 and S006 have been performed. It is preferable that the step S001 is performed simultaneously with the step S001, or the step S001 to the step S002 are performed simultaneously. This may be performed during the execution of S003.

[0071] Step S006 is based on the results of the supervised learning performed in step S003. and a step of inferring the characteristics of the semiconductor element from the prediction data of the characteristics of the semiconductor element. In other words, in step S006, the characteristics of the semiconductor element are calculated using the trained model. Step S006 is a process of inferring the characteristics of the semiconductor device from the data for predicting the characteristics. This is performed by the calculation unit 103 shown in FIGS. 1A and 1B.

[0072] Step S007 is a step of outputting the third data. The third data is output by the output unit 104 shown in FIG. 1A and FIG. 1B. This includes information about the results or consequences of the above inferences.

[0073] As a result, the characteristics of the semiconductor device can be predicted. Instead, the result of the inference or information about the result of the inference is stored in a memory unit shown in FIG. 1A or the like. Alternatively, step S007 may not be performed. good.

[0074] The steps related to supervised learning (steps S001 to S003) are as follows: This may be performed every time information about the semiconductor device is stored in the storage unit 105, or At predetermined times and on a regular basis (for example, once a day or once a week) etc.) may be implemented.

[0075] The method for predicting the characteristics of a semiconductor device is not limited to the above method. The method for predicting the characteristics is as follows: after step S003, the trained The trained model may be stored in the storage unit 1 shown in FIG. 06. By generating a trained model in advance, the characteristics of semiconductor elements can be predicted. When doing so, steps S001 to S003 can be omitted. This reduces the time required to predict the characteristics of the semiconductor device.

[0076] <<Neural Networks>> Here, we will explain neural networks that can be used for supervised learning. .

[0077] As shown in Figure 3A, the neural network NN consists of an input layer IL, an output layer OL, and It can be composed of an input layer IL, an output layer OL, and a hidden layer HL. Each L has one or more neurons (units). The hidden layer HL has It may be one layer or two or more layers. A neural network can also be called a deep neural network (DNN). Learning using deep neural networks is called deep learning. You can also do this.

[0078] Input data is input to each neuron in the input layer IL. The output signal of the neurons in the previous or next layer is input to each neuron in the output layer OL. The output signal of the neuron in the previous layer is input to each neuron. It can be connected to all neurons (fully connected) or only to some neurons (not fully connected). That's fine.

[0079] Figure 3B shows an example of a neuron operation. Here, we consider a neuron N and a neuron The figure shows two neurons in the front layer that output signals to N. Neuron N has one of the neurons in the front layer. The output x1 of one neuron and the output x2 of the other neuron in the previous layer are input. In neuron N, the multiplication result of output x1 and weight w1 (x1w1) and the multiplication result of output x2 and weight w1 (x1w1) are After the sum of the multiplication results (x2w2) of x1w1 and x2w2 is calculated, The bias b is added to get the value a=x1w1+x2w2+b. Then the value a is the activation The neuron N outputs an output signal y=ah after being transformed by the activation function h. For example, the number h can be a sigmoid function, a tanh function, a softmax function, or a ReLU function. A number, a threshold function, etc. can be used.

[0080] In this way, the operation of a neuron involves adding up the product of the output of the previous layer neuron and the weight. This multiplication and addition operation (x1w1+x2w2 above) is called multiplication and addition. This may be done on software using a program or by hardware. When the multiply-accumulate operation is performed by hardware, a multiply-accumulate circuit can be used. This product-sum operation circuit may be a digital circuit or an analog circuit. When an analog circuit is used for the sum-of-products operation circuit, the circuit scale of the sum-of-products operation circuit can be reduced. Alternatively, the number of times memory is accessed is reduced, improving processing speed and reducing power consumption. It is possible.

[0081] The multiply-and-accumulate circuit is a transistor that contains silicon (such as single crystal silicon) in the channel formation region. The channel forming region may be formed by a silicon transistor (hereinafter also referred to as a silicon transistor). The transistors are composed of transistors containing an oxide semiconductor (hereinafter also referred to as OS transistors). In particular, since the off-state current of an OS transistor is extremely small, it is possible to use it as an analog It is suitable as a transistor for configuring a log memory. The sum-of-products circuit may be configured using both transistors.

[0082] When the product-sum operation is performed by hardware, the product-sum operation circuit is It is preferable that the calculation unit 103 has the above-mentioned function.

[0083] The above is the explanation of the neural network. For this purpose, it is preferable to use deep learning. That is, it is preferable to use a hidden layer (HL) with two or more layers. It is preferable to use a neural network that

[0084] The above is a description of an example of a method for predicting the characteristics of a semiconductor device.

[0085] <Details of the method for predicting semiconductor device characteristics> The method for predicting the characteristics of a semiconductor device will be described in detail below with reference to FIGS. 4A to 7B. I will explain.

[0086] <<Structure of semiconductor elements>> First, the structure of a semiconductor element will be explained. The transistor will now be described.

[0087] Transistors are classified into various categories based on the relative positions and shapes of their components. The transistor structure is based on the positional relationship between the substrate, gate, and channel forming region. The gate is connected between the channel forming region and the substrate. The transistor structure with the channel between the two is called a bottom gate structure. The transistor structure in which the gate formation region is provided between the gate and the substrate is called a top-gate structure. It is called.

[0088] Furthermore, the transistor structure includes a semiconductor layer that forms a source and a drain and a channel. Depending on the connection point, they are classified into bottom-contact type and top-contact type. The source and drain are connected to the semiconductor layer that forms the channel on the substrate side. The top structure is called a bottom-contact structure. The transistor structure in which the semiconductor layer to be formed is connected on the opposite side of the substrate is called top contact. This is called a T-type structure.

[0089] That is, the transistor structure is a BGBC (bottom gate bottom contact) type structure, BGTC (bottom gate top contact) structure, TGTC (top gate top contact) They are classified into two types: tactile (Tact) type structure and TGBC (Top Gate Bottom Contact) type structure.

[0090] In addition to the four structures mentioned above, the transistor structure also includes gates above and below the semiconductor layer. Dual gate structure in which the source and drain are separated from the gate pattern. TGSA (Top-Gate Self-Align) structure formed by rufaline There are some.

[0091] The semiconductor elements 30_1 to 30_m have the same structure or For example, the semiconductor elements 30_1 to 30_m are preferably similar. In the case where the semiconductor elements 30_1 to 30_m are transistors, the structure of the semiconductor elements 30_1 to 30_m is a BGBC type structure. BGTC type structure, TGTC type structure, TGBC type structure, dual gate type structure, or The TGSA structure is preferable. By making the structure of the semiconductor element the same, This can improve the accuracy of predicting the characteristics of children.

[0092] The semiconductor elements 30_1 to 30_m have different structures. When the semiconductor elements 30_1 to 30_m are transistors, for example, For example, a part of the structure of the semiconductor elements 30_1 to 30_m is a TGTC type structure. Others may be TGSA type structures. By combining multiple structures, versatility is improved. This makes it possible to predict the characteristics of semiconductor elements with high accuracy.

[0093] The above is a description of the structure of the semiconductor element.

[0094] <<Semiconductor element characteristics>> Next, the characteristics of the semiconductor element will be described.

[0095] In this specification and elsewhere, the characteristics of a semiconductor element refer to the electrical characteristics of the semiconductor element. The characteristics of the element include, for example, the drain current (Id)-gate voltage (Vg) characteristics, Current (Id) - drain voltage (Vd) characteristics, capacitance (C) - gate voltage (V) characteristics, etc. do.

[0096] The characteristics of the semiconductor element may be the results obtained in a reliability test. The results obtained include, for example, the change in on-current (Ion) over time (the stress time dependence of Ion) The change in ΔVsh over time (also called the stress time dependency of ΔVsh) be.

[0097] ΔVsh is the amount of change in the shift voltage (Vsh). Here, the shift voltage (Vsh) is , in the transistor drain current (Id) - gate voltage (Vg) curve, The tangent at the point where the slope of is maximum is defined as Vg, where the line Id = 1pA intersects with the line Vg. .

[0098] Reliability testing includes +GBT (Gate Bias Temperature) stress Test, +DBT (Drain Bias Temperature) stress test, -G BT stress test, +DGBT(Drain Gate Bias Temperature re) stress test, +BGBT (Back Gate Bias Temperature re) stress test, -BGBT stress test, etc.

[0099] Reliability tests may require long-term measurements, so it may take some time to obtain the results. Furthermore, the measurement device is occupied during the measurement period. Using a characteristic prediction system for semiconductor elements, it is possible to predict the results of reliability tests. Therefore, by deciding whether or not to conduct reliability testing based on the predicted results, It is possible to omit some of the reliability tests or to determine the priority of the reliability tests. This allows for effective use of the measuring device.

[0100] In this specification and elsewhere, the characteristics of semiconductor elements include those calculated from the measurement results of the electrical characteristics of the semiconductor elements. The characteristic values ​​to be output are also included. The characteristic values ​​include, for example, threshold voltage (Vth), Vs h, subthreshold swing value (S value), Ion, field effect mobility (μFE), etc. Here, the subthreshold swing value (S value) is the value obtained by This refers to the amount of change in gate voltage in the subthreshold region that changes the gate current by one order of magnitude. Then, the characteristic values ​​calculated from the measurement results of the electrical characteristics of the semiconductor element are called the characteristic values ​​of the semiconductor element, Or, it may simply be called a characteristic value.

[0101] The characteristics of semiconductor elements also include temperature characteristics. The temperature characteristics include the temperature characteristics of the threshold voltage and the temperature dependence of the capacitance characteristics. Since it is necessary to measure the temperature characteristics at different temperatures, it takes time to evaluate the temperature characteristics. Using a semiconductor element characteristic prediction system, we evaluate the fabrication and temperature characteristics of semiconductor elements. The temperature characteristics can be predicted without performing measurements to determine the temperature.

[0102] When the characteristics of a semiconductor element are characteristic values, the characteristics of the semiconductor element are recorded as numerical data. The characteristics of the semiconductor element are stored in the memory unit 105. In addition, when the characteristics of the semiconductor element are electrical characteristics or temperature characteristics, In this case, the characteristics of the semiconductor element are stored in the storage unit 105 as two-variable data. The characteristics of the semiconductor elements stored in the storage unit 105 are quantified.

[0103] For example, if the characteristics of a semiconductor device are the time-dependent change in ΔVsh obtained in a reliability test, A collection of data relating to the time and ΔVsh is stored in the storage unit 105. When the characteristics of a semiconductor device are Id-Vg characteristics, a collection of data on Vg and Id is stored in the storage unit 105.

[0104] The above is a description of the characteristics of the semiconductor element.

[0105] <<Learning Dataset>> Here, we will explain the training dataset for supervised learning.

[0106] 4A and 4B are diagrams showing the structure of the training data set 50. The data set 50 corresponds to the training data set DS generated by the processing unit 102. The dataset 50 includes training data 51_1 to training data 51_m. The data 51_i (i is an integer between 1 and m) is input data 52_i and training data 53_i. The learning data 51_i includes the data 53_i related to the semiconductor element 30_i. Contains information.

[0107] The learning data set 50 is a data set input to the processing unit 102 shown in FIGS. 1A and 1B. Therefore, the training data set 50 is generated from the data IN1. It is generated by extracting, processing, converting, selecting, removing, etc. the data contained in it.

[0108] In this embodiment, the training data is information about the semiconductor element. That is, the object of prediction in this embodiment is the characteristics of a semiconductor element.

[0109] In this embodiment, the input data includes information about the semiconductor device. It is preferable that the input data is created from a process list of the device. It is preferable that the information includes a part of the process list for the semiconductor device. The characteristics of a semiconductor device depend on the type of semiconductor material used in the layer where the channel is formed, the gate The type of conductive material used in the layer functioning as an electrode, the type of material used in the layer functioning as a gate insulating film The type of insulating material, the thickness of each of these layers, and the film formation conditions for each of these layers must be considered. The type of material used for the layer, the thickness of the layer, the conditions for forming the layer, etc. are affected by the semiconductor device. Therefore, the input data is created from the process list of the semiconductor device. It is preferable that this be done.

[0110] The data contained in the training dataset for supervised learning is preferably quantitative data. In other words, it is preferable that the data is digitized. Compared to when a dataset contains non-numerical data (qualitative data), the training dataset contains The data contained in the data is quantified, which prevents the machine learning model from becoming too complex. can.

[0111] In the learning data set 50 shown in FIG. 4A, input data 52_1 to input data 52_m are created from the process lists 10_1 to 10_m, respectively. , the teacher data 53_1 to the teacher data 53_m are characteristic 20_1 to characteristic 20_ It is created from m.

[0112] As shown in FIG. 4B, the input data 52_1 to 52_m are respectively Process lists 10_1 to 10_m and semiconductor elements 30_1 to 30_m The input data 52_1 to 52_m may be used to generate the input data 52_1 to 52_m. 52_m, information about the shapes of the semiconductor elements 30_1 to 30_m, respectively. By adding this, it is possible to improve the accuracy of predicting the characteristics of semiconductor elements.

[0113] The number of steps in each of the step lists 10_1 to 10_m is the same. This makes it easier to create a training dataset or prediction data. When creating a learning dataset or prediction data, For example, extracting a part of the process list or selecting another part of the process list. Therefore, the number of steps in each of the process lists 10_1 to 10_m is may be different.

[0114] As described above, the characteristics 20_1 to 20_m are digitized data. Without any special conversion, they can be included in the training data 53_1 to 53_m, respectively. This can be done.

[0115] In addition, when the characteristics 20_1 to 20_m are two-variable data, One or more characteristic points are extracted from the training data 53_1 to training data 53_2, respectively. 3_m. Also, the values ​​of one of the two variables may be set at equal intervals. A plurality of points are extracted from the two-variable data and are respectively set as teacher data 53_1 to teacher data 53_2. It may be included in 3_m.

[0116] FIG. 5A is a diagram illustrating the results obtained in a reliability test of a semiconductor device. The horizontal axis is the elapsed time from the start of measurement (also called stress time) [h], and the vertical axis is ΔVsh [mV]. For example, from time A1 to time A10, and from time A1 to time A10 It is advisable to extract the value of ΔVsh in this case and use it as training data.

[0117] Note that even if the value of ΔVsh is characteristic for some or all of the time periods A1 to A10, Alternatively, the time A1 to the time A10 may be at equal intervals. A part of the time A10 is the first interval, and the other part of the time A1 to the time A10 is the second interval. There may be a second interval that is different from the first interval.

[0118] In addition, the number of sets of extracted times and ΔVsh values ​​at those times is limited to 10. The number of the suffixes may be 1 to 9 or 11 or more.

[0119] FIG. 5B is a diagram illustrating the Id-Vg characteristics of a semiconductor element. The drain current value at a gate voltage of 0V is one of the characteristic points. For example, The gate voltage is designated as voltage B4. Voltages B1 to B Also, voltages B5 to B10 are designated as voltages higher than voltage B4. For example, the voltages B1 to B10 and the drains at the voltages B1 to B10 It is advisable to extract the value of the in-current and use it as training data. One of the voltages B1 to B10 may be set to 0V.

[0120] The number of pairs of extracted voltages and drain current values ​​at those voltages is limited to 10. The number is not limited to, and may be 1 to 9 or 11 or more.

[0121] <<How to create input data>> Here, regarding a method for creating the input data 52_1 to 52_m shown in FIG. 4A, explain.

[0122] First, a method for creating input data 52_1 will be described. An example in which the process list 2_1 is created from the process list 10_1 will be described with reference to FIGS. 6A to 6C. do.

[0123] In the process list, a plurality of processes are set in the order of the manufacturing process of the semiconductor device. The manufacturing process includes, for example, film formation, cleaning, resist application, exposure, development, processing, and heating. There are processes such as processing, inspection, and substrate transfer.

[0124] In addition, processing conditions are specified for each of the multiple processes set in the process list. For example, the processing conditions for the film formation process include the equipment, material, film thickness, temperature, pressure, power, flow rate, etc. The processing conditions in the film formation process may affect the characteristics of the semiconductor device. In addition, even in processes other than film formation, the semiconductor This may affect the characteristics of the element.

[0125] The treatment conditions are a mixture of qualitative and quantitative data, and take values ​​on various scales. To express the similarity of features in each process, for example, qualitative data on materials It is advisable to convert the above into quantitative data such as physical properties for each material, and use the set as a feature.

[0126] Here, the process list 10_1 has n processes (n is an integer of 2 or more). For example, in the process list 10_1 shown in FIG. 6A, the first process is a substrate transfer process. The jth process (j is an integer of 2 or more and (n-4) or less) is a film forming process, and (j+ The (j+3)th process is a deposition process, the (j+2)th process is a processing process, and the (j+3)th process is a film deposition process. The n-th process is a heat treatment process, and the n-th process is a substrate transfer process. The numbers shown in Figures 6A and 6B are process numbers.

[0127] The processing conditions specified in the j-th process (film formation process) are defined as condition 1 to condition p (p is 2 or more). In addition, the processing conditions specified in the (j+1)th process (processing process) are , conditions 1 to q (q is an integer equal to or greater than 1). The processing conditions specified in the film forming process are defined as condition 1 to condition r (r is an integer of 2 or more). In addition, the processing conditions specified in the (j+3)th step (heat treatment step) are set as conditions 1 to Let the condition be s (s is an integer greater than or equal to 1).

[0128] First, some processes are extracted from the n processes included in the process list 10_1. The processes that are excluded are, for example, processes that are estimated to have a large effect on the characteristics of semiconductor elements. For example, it is a process in which conditions change frequently. By extracting them, we can reduce the number of parameters in machine learning. The number of neurons in the input layer in supervised learning using neural networks This allows us to optimize the number of hidden layers and the number of neurons in the hidden layers. This reduces the amount of calculation or the time required for learning or inference. It may be possible to prevent it.

[0129] For example, the process conditions specified in the j-th process (film forming process) are listed in the process lists 10_1 to 10_2. If the process list 10_m is often changed, the jth process (film formation process) is It is recommended to extract it from the process list 10_1. Also, for example, the (j+3)th process (heat treatment If it is estimated that the (j+3)th process ( It is advisable to extract the heat treatment process from the process list 10_1.

[0130] Or, from the n processes included in the process list 10_1, select some processes different from the above. The process of removing may be carried out, for example, when the contribution of the semiconductor characteristics is estimated to be small. Also, for example, it is a process in which the processing conditions are not changed. By removing some of the steps involved, the number of parameters in machine learning can be reduced. In other words, in supervised learning using neural networks, the input layer This reduces the number of hidden layers and the number of neurons in the hidden layers. The number of loops is optimized, which can reduce the amount of calculation or the calculation time for learning or inference. It may also be possible to prevent overlearning.

[0131] For example, the substrate transfer process (the first process and the nth process) depends on the characteristics of the semiconductor element. Therefore, it is assumed that there is no influence on the substrate transfer process (the first process and the nth process). It is advisable to remove the (j+1)th process from the process list 10_1. (Processing process) and the processing conditions specified in the (j+2)th process (Film formation process) The processing conditions are the same among the process lists 10_1 to 10_m. In this case, the (j+1)th process (processing process) and the (j+2)th process (film formation process) It is recommended to remove this from process list 10_1.

[0132] As a result, some processes are extracted from the process list 10_1. In Figure 6B, the j-th step (j The example shown in Figure 6B is an example where the first (+3) step is extracted. The (j+1)th step, the (j+2)th step, the nth step, etc. are removed. This is also the case.

[0133] Next, a process list 10_1 from which some processes have been extracted, or a process list different from the extracted processes, The data of the processing conditions included in the process list 10_1 from which some processes have been removed is quantified. do.

[0134] As described above, the processing conditions for the film formation process include, for example, the device, material, film thickness, temperature, pressure, The film thickness, temperature, pressure, power, flow rate, etc. are set as values. These processing conditions are numerically expressed data. Therefore, these processing conditions are particularly It can be included in the input data 52_1 without conversion.

[0135] It is preferable that the set values ​​of each processing condition are all in the same unit. By doing so, the amount of data included in the learning dataset 50 can be reduced. This reduces the time spent on data transmission and reception, learning, inference, etc.

[0136] Equipment data may be included in the process list as qualitative data. Qualitative data related to the equipment include, for example, the name of the equipment (including abbreviations and names), These include techniques such as:

[0137] Additionally, data on materials may be included in the process list as qualitative data. Qualitative data on materials include, for example, the name of the material (including abbreviations and names), composition formula, etc. That is it.

[0138] As mentioned above, the data contained in the training dataset used in supervised learning is numerical. Therefore, the qualitative data included in the process list is quantified. It is preferable.

[0139] [Quantification of qualitative data related to the device] Here, we will explain how to quantify qualitative data related to the equipment. An example will be described in which the device name is input as qualitative data relating to the device in item 1.

[0140] The film formation equipment includes a chemical vapor deposition (CVD) Equipment capable of forming films using the deposition method (sometimes called CVD equipment), Equipment capable of forming films using the sputtering method (sometimes called a sputtering equipment) ), using the atomic layer deposition (ALD) method. There are also devices that can form films using ALD (sometimes called ALD devices).

[0141] The CVD method is a plasma CVD (PECVD) method that uses plasma. Enhanced CVD (TCVD) method, and thermal CVD (TCVD) method. They can be classified into the VD method, the photo CVD method, and the photo CVD method. In the CVD method, different CVD equipment may be used depending on the method. The same applies to sputtering equipment, ALD equipment, etc. do.

[0142] The data about the equipment (here, the equipment name) entered as the processing conditions is qualitative data. Therefore, Label Encoding is used to quantify the qualitative data about the device. For example, it is advisable to manage device names by ID. It is advisable to assign an ID that is different from the process list ID. The generated ID is referred to as the device ID.

[0143] FIG. 7A shows an example of a correspondence table between device names and device IDs. For example, if the device name is CVD1, If the device name is CVD2, the device ID is set to 2. If the device name is SP1, the device ID is 3. By converting the device name to the device ID, The device name can be treated as numeric data.

[0144] The correspondence table may be stored in the storage unit 105. In addition, whenever an available device is added, The new device name and the device ID associated with the new device name are input to the input unit 101. It is advisable to add the information to the correspondence table via a storage medium, communication, or the like.

[0145] Label Encoding is used to quantify qualitative data about the device. However, the method for quantifying qualitative data about the device is not limited to this. One-hot encoding (1 of K Encoding) may also be used.

[0146] For example, if the number of devices that can be used in the film formation process is t (t is an integer of 1 or more), In this case, it is advisable to express the equipment name as a t-dimensional one-hot vector. If the number of devices that can be used is not large, it can be expressed by a low-dimensional vector. This can reduce the amount of calculation or the calculation time for learning or inference.

[0147] In addition to the above, there are other methods for quantifying qualitative data about equipment, such as T Target Encoding, etc. may also be used.

[0148] The above is an explanation of how qualitative data relating to the device is quantified.

[0149] [Quantification of qualitative data on materials] Here, we will explain how to quantify qualitative data related to materials. In item 2, the name of the material (including abbreviations and names) is entered as qualitative data about the material. The material is an inorganic material.

[0150] The crystal structure and film quality of materials used in semiconductor devices change depending on the processing conditions. Furthermore, it also varies depending on the material used for the film to be formed, the roughness of the surface to be formed, and the like. Therefore, using a database, etc., it is necessary to convert the qualitative data, such as the names of materials, into quantitative data. When converting the properties of the material (crystal structure, density, dielectric constant, etc.) into the characteristics of the semiconductor element, Therefore, in this embodiment, qualitative data on the material is used. (Here, the name of the material) is converted into constituent elements and composition.

[0151] First, convert the name of the material into a composition formula. For example, if the condition 2 of the film formation process is "oxidation", If "silicon" is entered, it is recommended to convert it to "SiO2". The conversion from the compound to the composition formula may be performed using a concept dictionary or database, or by using a material dictionary prepared in advance. A correspondence table between the names and composition formulas may be used.

[0152] Next, the composition formula is converted into the constituent elements and composition. For example, if the material contains elements M1, M2, M3, M4, M5, M6, M7, M8, M9, M10, M11, M12, M13, M14, M15, M16, M17, M18, M19, M20, M21, M22, M23, M24, M25, M26, M27, M28, M3 2. It is composed of elements M3 and M4, and M1:M2:M3:M4=w:x:y If the composition is w:x:y:z, the formula can be "M1, M2, M3, M4, w:x:y:z" or " Convert to "M1, w, M2, x, M3, y, M4, z".

[0153] It is preferable that the composition be normalized. For example, the composition should be normalized so that w+x+y+z=1 is satisfied. This allows for the determination of the composition of materials with the same combination of constituent elements but different compositions. It is possible to distinguish between materials that are different from each other.

[0154] If the material is composed of one element, M2, M3, M4, x, y, and z are all zero. Similarly, if a material is composed of two elements, it can be written as M3, M4, y, z In addition, if the material is composed of three elements, M4 and z should be written as zero. It should be written as zero.

[0155] Specifically, if the composition formula is "SiO2", it is expressed as "Si, O, 0, 0, 0.333:0 0.667:0:0" or "Si, 0.333, O, 0.667, 0, 0, 0, 0" Exchange.

[0156] In the above, the number of elements and composition are recorded so that it can be applied to materials with four or less constituent elements. However, the present invention is not limited to this. For example, it can also be applied to materials with five or more constituent elements. Alternatively, the number and composition of elements may be described so that the element can be used in, for example, a semiconductor device. If the material has three or fewer constituent elements, the composition formula is "M1, M2, M3, w:x:y". Or you can convert it to "M1, w, M2, x, M3, y". This allows you to This reduces the number of hidden layers and the number of hidden layer parameters. The number of parameters is optimized, which can reduce the amount of calculation or time required for training or inference. .

[0157] Next, convert the elements into their properties, such as atomic number, group, period, and electron Configuration, atomic mass, atomic radius, atomic volume, electronegativity, ionization energy, electron affinity, bipolar Polarizability, melting point of a simple substance, boiling point of a simple substance, lattice constant of a simple substance, density of a simple substance, thermal conductivity of a simple substance, etc. In addition, atomic radii include covalent radii, van der Waals radii, ionic radii, and and metallic bond radius.

[0158] In particular, the atomic number or electron configuration and electronegativity are selected as the properties of the element to be transformed. When a material is composed of a single element, the characteristics of the material are the atomic number and Also, when a material is composed of two or more elements, the electronegativity is This is likely to occur in bonding patterns between different elements. For example, between elements with similar electronegativity, On the other hand, between elements with large differences in electronegativity, , ionic bonding becomes dominant.

[0159] FIG. 7B illustrates a correspondence table between elements and their properties. These include atomic number, electron configuration, electronegativity, and melting point (K) of the element. The conversion from the data to the element properties can be done using a database or by using a database of elements and elements created in advance. A correspondence table with element characteristics may be used.

[0160] Specifically, when selecting atomic number and electronegativity as the element properties, "Si " is converted to "14, 1.90", and "O" is converted to "8, 3.44".

[0161] As a result, "silicon oxide" entered as condition 2 of the film formation process is set to "14, 1.9 0, 8, 3.44, 0, 0, 0, 0, 0.333:0.667:0:0" or "14, Convert to "1.90, 0.333, 8, 3.44, 0.667, 0, 0, 0, 0, 0, 0" Therefore, qualitative data about materials can be quantified.

[0162] From the above, it is possible to predict the characteristics of semiconductor devices from first principles. Even when using materials that do not have a high resistance, the characteristics of the semiconductor element can be maintained without sacrificing precision. Furthermore, materials that are not listed in databases can be used to predict the properties of semiconductor devices. Even when used in semiconductor devices, the characteristics of semiconductor devices can be predicted without sacrificing accuracy. do.

[0163] Note that the conversion from the material name to the constituent elements and composition may be done directly without going through the composition formula. stomach.

[0164] This concludes the explanation of how to quantify qualitative data related to materials.

[0165] As a result, the input data 52_1, which is made up of digitized data as shown in FIG. 6C, is Specifically, the input data 52_1 can be generated based on the processing conditions shown in FIG. The input data 52_1 may include a process number.

[0166] In addition, the process of selecting (extracting or removing) from the process list and the process of quantifying qualitative data are The order of the steps of converting qualitative data into quantitative data is not limited to the above. After converting the data, the process list can be selected (extracted or removed) to The input data 52_1 may be generated from the input data 10_1.

[0167] As a result, the learning data 51_1 including the digitized data can be generated. The learning data 51_2 to 51_m are the same as the learning data 51_1. That is, by the above method, the learning data 51_2 to the learning data 5 1_m can be generated.

[0168] In FIG. 4A, the input data 52_1 to 52_m are respectively included in the process list 10. Although the example shows a case where the process list is created from _1 to _m, it is not limited to this. For example, as shown in FIG. 4B, the input data 52_1 to 52_m are Process list 10_1 to process list 10_m and semiconductor element 30_1 to semiconductor element 30_ and information about the shape of m.

[0169] The input data 52_1 to 52_m are the process lists 10_1 to 10_m, respectively. a process list 10_m, first characteristics of the semiconductor elements 30_1 to 30_m, The teacher data 53_1 to 53_m are generated from the semiconductor device 30 _1 to 30_m may be created from the second characteristics of the semiconductor elements 30_1 to 30_m.

[0170] In the above, the first characteristic and the second characteristic are made different. For example, the first characteristic is The first characteristic may be a characteristic value of the semiconductor element, and the second characteristic may be the result of a reliability test of the semiconductor element. There are many factors that affect the reliability of semiconductor devices, and each factor is intricately intertwined, so it is difficult to judge the reliability based on experience. It is difficult to predict. Therefore, the reliability of semiconductor devices is a suitable subject for estimation. The characteristic values ​​of semiconductor elements indirectly include information on the manufacturing process of the semiconductor elements. Therefore, by adding the characteristic values ​​of semiconductor elements to the input data, the information is provided to supervised learning. This makes it possible to improve the accuracy of prediction of the characteristics of semiconductor devices.

[0171] The learning data set 50 contains only data on semiconductor devices with the same or similar structure. In other words, the learning data set 50 may be created for each structure of a semiconductor device. This can improve the accuracy of prediction of the characteristics of the semiconductor device. In addition, the learning data set 50 may be composed of data on semiconductor elements regardless of their structure. This makes it possible to predict the characteristics of semiconductor devices in a highly versatile manner.

[0172] This concludes the explanation of the training dataset. By training a machine learning model using this data, it is possible to predict the characteristics of semiconductor devices.

[0173] <<Data for predicting the characteristics of semiconductor elements>> Here, the prediction data for the characteristics of the semiconductor device will be described.

[0174] Prediction data for the characteristics of the semiconductor device is input to a processing unit 102 shown in FIGS. 1A and 1B. Therefore, the data for predicting the characteristics of the semiconductor element is generated from the data IN2. It is generated by extracting, processing, converting, selecting, removing, etc. the data contained in DATA IN2. It is done.

[0175] The data IN2 includes at least information about the semiconductor device. 2 may include the characteristics of semiconductor elements.

[0176] The data for predicting the characteristics of the semiconductor element is the same as the input data for the learning data described above. For example, the input data 52_1 to 52_m may be respectively When the process list 10_1 to the process list 10_m are used, the characteristics of the semiconductor element are The prediction data may be created from the process list 11. For example, the input data 52 _1 to input data 52_m are respectively set as process lists 10_1 to 10_m. , the characteristics of the semiconductor elements 30_1 to 30_m, and when they are created, the semiconductor element The data for predicting the child characteristics includes the process list 11 and the process list ID of the process list 11. It is preferable that the characteristics of the semiconductor element to which the signal is connected be used.

[0177] The above is an explanation of the data for predicting the characteristics of semiconductor elements.

[0178] According to one embodiment of the present invention, a semiconductor element can be formed without using the physical properties of a material contained in the semiconductor element. In addition, by using past experimental data, it is possible to predict the characteristics of semiconductor elements. Structural optimization can be accelerated by virtual screening, which is a method for analyzing data. Even if the results are not interpolative, they may be based on nonlinear or higher-order representations of machine learning models. In addition, the representation obtained by the machine learning model can be considered as interpolative. By cutting out and examining one part of the data, it is possible to discover patterns that would not have been noticed before.

[0179] <Computer equipment> In this section, a computer having a semiconductor device characteristic prediction system according to one aspect of the present invention will be described. The data device will be described with reference to FIG.

[0180] FIG. 8 is a diagram illustrating a computer device having a semiconductor device characteristic prediction system. The computer device 1000 includes an arithmetic unit 1001, a memory 1002, an input / output interface, and a The computer includes an interface 1003, a communication device 1004, and a storage device 1005. The computer device 1000 is connected to a display device 1006a, a display device 1006b, and a display device 1006c via an input / output interface 1003. and keyboard 1006b.

[0181] The computer device 1000 is an information processing device such as a personal computer used by a user. In this case, the arithmetic unit 1001 may be a processing unit that performs the processing shown in FIGS. 1A and 1B. 1A and 1B. The display device 1006a includes a memory unit 105 and / or a storage unit 106 shown in FIG. 1A and 1B. The keyboard 1006b corresponds to the output unit 104 shown in FIG. 1A and 1B.

[0182] The trained model may be stored in memory 1002 or in storage 1005. This may be done.

[0183] The computer device 1000 also transmits data via a network. Database 1011, Remote Computer 1012, and Remote Computer 101 3. The computer device 1000 may be connected to the It is electrically connected to the network interface 1007. The interface 1007 communicates with the database 10 via a network. 11, the remote computer 1012, and the remote computer 1013 electrically Connected.

[0184] Here, the above network includes a local area network (LAN) and an internet The above network may be either wired or wireless. In addition, wireless communication can be used in the above network. When using short-range wireless technology such as Wi-Fi (registered trademark) or Bluetooth (registered trademark), In addition to communication methods, there are also communication methods that comply with the third generation mobile communication system (3G), LTE (3. 9G) and 4th generation mobile communication system (4G) or communication methods compliant with the 5th generation mobile communication system (5G). Various communication means can be used.

[0185] As described above, the processing unit of the semiconductor device characteristic prediction system is provided in the server, and the client It may also be configured so that it can be accessed and used from a computer via a network. The computer device 1000 is regarded as the client PC, and the remote computer 1012 and The local and / or remote computer 1013 may be considered the server.

[0186] At this time, the remote computer 1012 and / or the remote computer 101 3, the processing unit 102 and the calculation unit 103 shown in FIGS. 1A and 1B are provided. , the remote computer 1012 and / or the remote computer 1013 have The calculation device includes a processing unit 102 and a calculation unit 103. The database 1011 includes: It includes the memory unit 105 and / or the storage unit 106 shown in FIGS. 1A and 1B.

[0187] As described above, one embodiment of the present invention can provide a system for predicting characteristics of a semiconductor element. Furthermore, one embodiment of the present invention can provide a method for predicting characteristics of a semiconductor element. Another aspect of the present invention provides a training data set for predicting the characteristics of a semiconductor device. It is possible.

[0188] This embodiment can be carried out by combining parts thereof as appropriate. [Explanation of symbols]

[0189] IN1: Data, IN2: Data, 10: Multiple process lists, 10_m: Process list, 1 0_1: Process list, 11: Process list, 20: Multiple characteristics, 20_m: Characteristics, 20_1 : characteristics, 30: plural semiconductor elements, 30_i: semiconductor elements, 30_m: semiconductor elements, 30 _1: Semiconductor element, 50: Training data set, 51_i: Training data, 51_m: Training Training data, 51_1: Training data, 51_2: Training data, 52_i: Input data , 52_m: input data, 52_1: input data, 53_i: training data, 53_m: training data Teacher data, 53_1: Teacher data, 100: Characteristic prediction system, 101: Input section, 102 : processing unit, 103: calculation unit, 104: output unit, 105: memory unit, 106: storage unit, 100 0: Computer device, 1001: Arithmetic unit, 1002: Memory, 1003: Input / output interface interface, 1004: communication device, 1005: storage, 1006a: display device , 1006b: Keyboard, 1007: Network interface, 1011: Data database, 1012: remote computer, 1013: remote computer

Claims

[Claim 1] A semiconductor device characteristic prediction system that performs supervised learning based on a learning data set, and infers characteristics of the semiconductor device from prediction data based on a result of the learning, the semiconductor device characteristic prediction system includes a storage unit, an input unit, a processing unit, and an arithmetic unit; The processing unit a function of creating the learning dataset from the first data stored in the storage unit; a function of generating the prediction data from the second data supplied from the input unit; The ability to convert qualitative data into quantitative data and A function of extracting or removing the first data and the second data; and the first data includes a process list for a first semiconductor element to an mth semiconductor element (m is an integer of 2 or more) and characteristics of the first semiconductor element to the mth semiconductor element; the second data includes a process list for an (m+1)th semiconductor device; The qualitative data is the name or composition formula of the material; The quantitative data is elemental properties and composition; The calculation unit has a function of performing learning and inference in the supervised learning. A system for predicting semiconductor device characteristics.

Images