Data enhancement method and device, equipment and medium
By using a generative adversarial network (GAN) to enhance semiconductor device data, the problems of high data acquisition time and cost in the prior art are solved, and the expansion of data samples and the improvement of acquisition efficiency are achieved.
Patent Information
- Application Number
- CN202311499283.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-11-10
- Publication Date
- 2025-05-13
AI Technical Summary
The prior art faces the problems of high data acquisition time and cost when acquiring simulation or actual test data of semiconductor devices, especially when processing nanometer-scale devices, TCAD simulation and actual test have limitations of high computational cost and long simulation time.
Generative adversarial network (GAN) is used as the data augmentation model, and the second device data is generated by training the first device data, including device parameters and leakage current data, thereby expanding the data samples and reducing the time and cost of data acquisition.
Through the data enhancement method, a large amount of second device data can be generated, which reduces the time and cost of data acquisition and improves the efficiency of data acquisition.
Smart Images
Figure CN119990020A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of semiconductor devices, and in particular to a data enhancement method, device, equipment and medium. Background Art
[0002] Over the past decade, complementary metal oxide semiconductor (CMOS) has been following Moore's Law, doubling the number of transistors per unit area every 18-24 months. CMOS devices continue to shrink, and as transistor size continues to shrink into the nanometer scale, traditional device theory and process technology have been challenged by physical limits, and device performance has seriously declined.
[0003] In order to improve the performance of the device, it is necessary to optimize the device based on simulation or actual test data. Usually, engineers rely on semiconductor process and device simulation tools (Technology Computer Aided Design, TCAD) to simulate and model semiconductor devices, but TCAD-based device modeling exposes deficiencies: 1) The physics-based model equations require a lot of time and expertise; 2) TCAD has some limitations when dealing with devices with relatively small feature sizes, and has high computational costs and long simulation times. These deficiencies make it difficult to obtain TCAD simulation data. At the same time, the acquisition of actual test data is also subject to multiple restrictions. First, high-precision test equipment and technology are required to obtain accurate data; second, these test processes usually take a long time and have high economic and labor costs. Therefore, it is not easy to obtain enough data through simulation or actual testing. Therefore, providing a suitable data enhancement method has become a technical problem that needs to be solved urgently. Summary of the invention
[0004] In view of this, the purpose of this application is to provide a data enhancement method, device, equipment and medium to enhance data and generate new data samples, that is, to obtain a large amount of second device data and reduce the time and cost of data collection. The specific scheme is as follows:
[0005] On the one hand, the present application provides a data enhancement method, comprising:
[0006] Acquire first device data from simulation data or actual test data; the first device data includes a first device parameter and a first leakage current corresponding to the first device parameter; the first device parameter includes at least one of a channel length, a gate oxide layer thickness, a gate voltage or a drain voltage;
[0007] The first device data is input into a data enhancement model for training, and when the training is completed, second device data is output; the second device data includes second device parameters and a second leakage current corresponding to the second device parameters.
[0008] Specifically, the data enhancement model includes a generative adversarial network, the generative adversarial network includes a generator and a discriminator, and the training process of the generative adversarial network includes:
[0009] Using a random variable as an input of the generator, and outputting third device data; the third device data includes a third device parameter and a third leakage current corresponding to the third device parameter; the third device data and the first device data have similar distribution;
[0010] Inputting the first device data and the third device data into the discriminator, and outputting a first discrimination result corresponding to the first device parameter and a second discrimination result corresponding to the third device data;
[0011] The first parameters of the generator and the second parameters of the discriminator are alternately iteratively optimized until the first loss function of the generator and the second loss function of the discriminator are minimized to obtain the trained generative adversarial network; the second loss function is determined according to the first discrimination result and the second discrimination result, and the first loss function is determined according to the second discrimination result.
[0012] Specifically, the first device data and the third device data are input into the discriminator, and a first discrimination result corresponding to the first device data and a second discrimination result corresponding to the third device data are output, including:
[0013] Preprocessing the first device data to obtain fourth device data; the fourth device data includes fourth device parameters obtained by standardizing the first device parameters and fourth leakage current obtained by logarithmically processing the first leakage current;
[0014] The third device data and the fourth device data are input into the discriminator, and the first discrimination result corresponding to the fourth device data and the second discrimination result corresponding to the third device data are output.
[0015] Specifically, the first loss function is expressed as:
[0016]
[0017] Among them, the G loss is the first loss function, the z i is the i-th random variable, the G(zi ) is the third device data, and the D(G(z i )) is the second discrimination result, and the n fake represents the number of samples of the third device data;
[0018] The second loss function is expressed as:
[0019]
[0020] Among them, the D loss is the second loss function, the x i is the i-th first device data, the D(x i ) is the first discrimination result, and the n real Indicates the sample number of the first device data.
[0021] Specifically, the data enhancement model includes a boundary search generative adversarial network, a W-distance generative adversarial network, or a conditional generative adversarial network.
[0022] Specifically, the method further includes:
[0023] The deep neural network model is trained using the first device data and the second device data until a third loss function is minimized to obtain a trained deep neural network model; the deep neural network model is used to predict leakage current.
[0024] Specifically, the activation function in the data augmentation model includes at least one of a sigmoid activation function, a tanh activation function or a relu activation function.
[0025] In another aspect, the embodiment of the present application further provides a data enhancement device, including:
[0026] an acquisition unit, configured to acquire first device data from simulation data or actual test data; the first device data comprising a first device parameter and a first leakage current corresponding to the first device parameter; the first device parameter comprising at least one of a channel length, a gate oxide layer thickness, a gate voltage or a drain voltage;
[0027] An output unit is used to input the first device data into a trained data enhancement model and output second device data; the second device data includes second device parameters and a second leakage current corresponding to the second device parameters.
[0028] In another aspect, an embodiment of the present application provides a computer device, the computer device comprising a processor and a memory:
[0029] The memory is used to store program code and transmit the program code to the processor;
[0030] The processor is configured to execute the method described above according to the instructions in the program code.
[0031] On the other hand, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store a computer program, and the computer program is used to execute the method described in the above aspects.
[0032] The embodiment of the present application provides a data enhancement method, apparatus, device and medium, which obtains first device data from simulation data or actual test data; the first device data includes a first device parameter and a first leakage current corresponding to the first device parameter; the first device parameter includes at least one of a channel length, a gate oxide thickness, a gate voltage or a drain voltage; the first device data is input into a data enhancement model for training, and when the training is completed, the second device data is output; the second device data includes a second device parameter and a second leakage current corresponding to the second device parameter. In the embodiment of the present application, the time and cost of obtaining the first device data from simulation data or actual test data is high. By using a data enhancement model, the data can be enhanced to generate new data samples, that is, a large amount of second device data is obtained, reducing the time and cost of data collection. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0034] Figure 1 A schematic diagram of a process flow of a data enhancement method provided in an embodiment of the present application is shown;
[0035] Figure 2 A schematic diagram of the structure of a GAN model provided in an embodiment of the present application is shown;
[0036] Figure 3 A structural block diagram of a data enhancement device provided in an embodiment of the present application;
[0037] Figure 4 A structural diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0038] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are described in detail below with reference to the accompanying drawings.
[0039] In the following description, many specific details are set forth to facilitate a full understanding of the present application, but the present application may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present application. Therefore, the present application is not limited to the specific embodiments disclosed below.
[0040] To facilitate understanding, a data enhancement method, device, equipment and medium provided in an embodiment of the present application are described in detail below with reference to the accompanying drawings.
[0041] refer to Figure 1 As shown, it is a flow chart of a data enhancement method provided in an embodiment of the present application, and the method may include the following steps.
[0042] S101, obtaining first device data from simulation data or actual test data; the first device data includes a first device parameter and a first leakage current corresponding to the first device parameter; the first device parameter includes at least one of a channel length, a gate oxide layer thickness, a gate voltage or a drain voltage.
[0043] In the embodiment of the present application, simulation data can be obtained from TCAD simulation, or actual testing can be performed, and part of the simulation data or actual test data is used as a training set, that is, as first device data, the first device data includes a first device parameter and a corresponding first leakage current, wherein the first device parameter may include a channel length L g , gate oxide thickness T ox , gate voltage V g Or drain voltage V d In addition, the data obtained through simulation or actual testing can be divided into a training set, a test set, and a validation set, for example, the ratio of the training set, the test set, and the validation set is 6:3:1. The training set can be input into the GAN for training.
[0044] Specifically, the leakage current of the device under different design parameters and bias conditions, i.e., the first device parameter and the first leakage current, can be obtained through TCAD simulation or actual testing. S102, inputting the first device data into the data enhancement model for training, and outputting the second device data when the training is completed; the second device data includes the second device parameter and the second leakage current corresponding to the second device parameter.
[0045] In an embodiment of the present application, the data augmentation model may be a generative model such as a generative adversarial network (GAN), or other generative models, such as an autoencoder (AE), a variational autoencoder (VAE), and the like.
[0046] Specifically, the first device data can be input into the data enhancement model for training. The data enhancement model is used to expand and enhance the data. When the training is completed, the second device data can be output. The second device data is new data expanded by the data enhancement model. The second device data includes second device parameters and a second leakage current. The amount of the second device data is large. By using a small amount of the first device data for training to generate new data samples, the data is enhanced, thereby effectively reducing the time and cost of data collection.
[0047] In the embodiment of the present application, the data enhancement model may be a GAN, which includes a generator G and a discriminator D. Figure 2 As shown in the figure, the purpose of the generator is to learn the distribution of real data and generate data G(z) that is as similar to the real data as possible, so that the discriminator cannot judge that the generated data G(z) is fake data. The purpose of the discriminator D is to judge as accurately as possible whether the input data is real data or fake data G(z) generated by the generator G. The input of the discriminator D consists of two parts, namely the real data x obtained from TCAD simulation or actual test and the data G(z) generated by the generator. Its output is usually a probability value, which indicates the probability that D determines that the input is a real distribution. If the input comes from real data, the output is 1, otherwise the output is 0. At the same time, the output of the discriminator will be fed back to the generator G to guide the training of the generator G.
[0048] The discriminator and generator continuously improve their discriminative and generative capabilities through such adversarial training, and eventually reach a Nash equilibrium state. At this time, the generator has learned a probability distribution that is close to the real data, and the discriminator can no longer correctly judge whether the input data comes from real data or from the false data G(z) generated by the generator. That is, the probability value output by the discriminator each time is 0.5, and the model reaches the optimal state.
[0049] Specifically, the training process of the generative adversarial network may specifically include the following steps.
[0050] S201, taking a random variable as an input of a generator, and outputting third device data, wherein the third device data includes a third device parameter and a third leakage current corresponding to the third device parameter, and the third device data and the first device data have similar distribution.
[0051] Specifically, the first device data can be represented by x, the generator can be composed of a fully connected neural network, the input is a random variable z, and the output is a third device data G(z) having a similar distribution to x. The generator can include multiple hidden layers, such as 2, 3, 5, etc. When the number of hidden layers is 2, the generator can be described as:
[0052] G(z)=w3×f0(w2×f0(w1×z+b1)+b2)+b3
[0053] Among them, w1, w2 and w3 are weights, b1, b2 and b3 are bias parameters, and f0 is the leaky relu activation function, which can also be other activation functions. The leaky relu activation function can be expressed as follows, and α is a variable parameter that can be set to 0.2:
[0054]
[0055] S202 , inputting the first device data and the third device data into a discriminator, and outputting a first discrimination result corresponding to the first device data and a second discrimination result corresponding to the third device data.
[0056] In an embodiment of the present application, the discriminator can be composed of a fully connected neural network, the input of the discriminator is real data (i.e., first device data) x and data generated by the generator (i.e., third device data) G(z), and the output is a first discrimination result D(x) corresponding to the first device data x, and a second discrimination result D(G(z)) corresponding to the third device data G(z).
[0057] The first discrimination result D(x) can be understood as the result of the discriminator's judgment on the real data x, and the second discrimination result D(G(z)) can be understood as the result of the discriminator's judgment on the data G(z) generated by the generator. The first discrimination result D(x) and the second discrimination result D(G(z)) can be probability values, both of which represent the probability that the sample is "true", that is, the probability that the label is 1.
[0058] Specifically, the discriminator may include multiple hidden layers. When two hidden layers are included and the input is the real data x, the discriminator can be expressed as:
[0059] D(x)=f2(w3·f1(w2·f1(w1·x+b1)+b2)+b3)
[0060] Among them, f1 can be a leaky relu function or other activation functions, and f2 can be a sigmoid function. Since the first discrimination result is a probability value, a sigmoid layer needs to be added to limit the output between 0 and 1, that is, f2 is a sigmoid function, and f2 is expressed as:
[0061]
[0062] When the input is the data G(z) generated by the generator, the discriminator can be expressed as:
[0063] D(G(z))=f2(w3×f1(w2×f1(w1×G(z)+b1)+b2)+b3)
[0064] In the embodiment of the present application, the first device data can be preprocessed to facilitate subsequent calculations. S202 can be specifically as follows: preprocessing the first device data to obtain fourth device data; the fourth device data includes fourth device parameters obtained by normalizing the first device parameters, and fourth leakage current obtained by logarithmically processing the first leakage current; the third device data and the fourth device data are input into the discriminator, and the first discrimination result corresponding to the fourth device data and the second discrimination result corresponding to the third device data are output.
[0065] Specifically, the first device parameter X can be standardized to obtain the fourth device parameter X norm , can be expressed using the following formula:
[0066]
[0067] Among them, μ represents the mean of the data, and s represents the standard deviation of the data.
[0068] Since the leakage current has a large span distribution, the first leakage current Y can be processed logarithmically to obtain the fourth leakage current Y norm , the fourth leakage current Y norm It can be expressed as:
[0069] Y norm =log 10 (Y)
[0070] S203, alternately iteratively optimize the first parameter of the generator and the second parameter of the discriminator until the first loss function of the generator and the second loss function of the discriminator are minimized to obtain a trained generative adversarial network; the second loss function is determined according to the first discrimination result and the second discrimination result, and the first loss function is determined according to the second discrimination result.
[0071] In the embodiment of the present application, during the training process, the cross entropy loss can be used to calculate the loss function of the discriminator and the generator. The generator has a first loss function G loss , the discriminator has a second loss function D loss .
[0072] Specifically, the first loss function is expressed as:
[0073]
[0074] Among them, G loss is the first loss function, z i is the i-th random variable, G(z i ) is the data of the third device, D(G(z i )) is the second discrimination result, and the n fake represents the number of samples of the third device data;
[0075] The second loss function is expressed as:
[0076]
[0077] Among them, D loss is the second loss function, x i is the data of the first device of the ith order, D(x i ) is the first discrimination result, and the n real Indicates the sample number of the first device data.
[0078] Specifically, the second loss function D of the discriminator is loss It can be composed of two parts, namely the loss function D of the discriminator on the real data x loss-x The loss function D of the discriminator on the data G(z) generated by the generator loss-G(z) .
[0079] When the input is the first device data x, the corresponding loss function D loss-x It can be expressed as:
[0080]
[0081] When the discriminator inputs the third device data G(z), the corresponding loss function D loss-G(z) It can be expressed as:
[0082]
[0083] Therefore, the total loss function of the discriminator is the sum of the two loss functions, which can be expressed as:
[0084]
[0085] Specifically, since the generator needs to make the generated data G(z) as similar as possible to the real data x, that is, to make the probability of the discriminator output close to 1 when the generated data G(z) is used as input, the loss function of the generator can be expressed as:
[0086]
[0087] Specifically, one of the generator or the discriminator can be fixed, and the first parameter of the generator and the second parameter of the discriminator can be alternately iteratively optimized through the Adam optimizer. For example, in a round of iterative optimization, the second parameter of the discriminator can be fixed first, and the first parameter of the generator can be optimized through the Adam optimizer. After the optimization is completed, the first parameter of the generator is fixed again, and the second parameter of the discriminator is optimized through the Adam optimizer, that is, alternating iterative optimization is performed. Then, alternating iterative optimization is also performed in the next round until the first loss function of the generator and the second loss function of the discriminator are minimized, indicating that the generator and the discriminator converge to the optimal, the probability of the discriminator outputting each time is 0.5, the generator generates data close to the real data x, and a trained generative adversarial network is obtained.
[0088] In an embodiment of the present application, the data augmentation model may include a boundary-seeking generative adversarial network (BGAN), a wasserstein generative adversarial network (WGAN) or a conditional generative adversarial network (CGAN), and those skilled in the art may select one according to actual conditions.
[0089] In an embodiment of the present application, the first device data and the second device data can be combined to form a new training data set, and the new training data set can be used to train a deep neural network model (Deep Neural Networks, DNN), so that the deep neural network model can predict the leakage current size, and can predict the electrical characteristics of the device under different parameters at a lower time cost. The deep neural network model can be trained using the first device data and the second device data, or the deep neural network model can be trained using the fourth device data and the second device data after preprocessing until the third loss function is minimized to obtain a trained deep neural network model.
[0090] It is understandable that after each round of DNN training, the data of the validation set can be input into the DNN, and the third loss function of the DNN on the validation set can be calculated. The model parameters that minimize the third loss function of the validation set are retained each time for the next round of training, and an optimal deep neural network model is finally obtained. Furthermore, the data of the test set can be input into the optimal deep neural network model, and the test results can be evaluated using the third loss function.
[0091] Specifically, the input of DNN can be expressed as x'=(Lg ,T ox ,...,V g ,V d ), i.e., device parameters, the output is the predicted leakage current DNN can have multiple hidden layers, for example, 3. The training of the DNN model uses a number of weight coefficient matrices w, bias vectors b and input value vectors x' to perform a series of linear operations and activation operations. Starting from the input layer, through 3 hidden layers, the predicted leakage current value is obtained. Therefore, the DNN model can be expressed as:
[0092]
[0093] Among them, w1, w2, w3 and w4 are weights, b1, b2, b3 and b4 are bias parameters, and f3 is the activation function tanh, specifically:
[0094]
[0095] Specifically, the performance of the DNN model is evaluated by the loss function Root Mean Squared Error (RMSE), which measures the error between the predicted output leakage current value and the actual leakage current value. The calculation formula is as follows:
[0096]
[0097] Among them, y i is the leakage current in the test set, is the leakage current predicted by DNN.
[0098] Specifically, the loss function can be optimized using the Adam optimizer. By continuously optimizing the network parameters w and b, the loss function is minimized and the optimal prediction result is obtained.
[0099] In an embodiment of the present application, the activation function in the data enhancement model or the deep neural network model may include at least one of a sigmoid activation function, a tanh activation function, or a relu activation function.
[0100] The embodiment of the present application provides a data enhancement method, which obtains first device data from simulation data or actual test data; the first device data includes a first device parameter and a first leakage current corresponding to the first device parameter; the first device parameter includes at least one of a channel length, a gate oxide thickness, a gate voltage or a drain voltage; the first device data is input into a data enhancement model for training, and when the training is completed, second device data is output; the second device data includes a second device parameter and a second leakage current corresponding to the second device parameter. In the embodiment of the present application, the time and cost of obtaining the first device data from simulation data or actual test data are high. By using a data enhancement model, the data can be enhanced to generate new data samples, that is, a large amount of second device data is obtained, reducing the time and cost of data collection.
[0101] Based on the above data enhancement method, the present application embodiment also provides a data enhancement device, referring to Figure 3 As shown, it is a structural block diagram of a data enhancement device provided in an embodiment of the present application, and the device may include:
[0102] An acquisition unit 201 is used to acquire first device data from simulation data or actual test data; the first device data includes a first device parameter and a first drain current corresponding to the first device parameter; the first device parameter includes at least one of a channel length, a gate oxide layer thickness, a gate voltage or a drain voltage;
[0103] The output unit 202 is used to input the first device data into the data enhancement model for training, and output second device data when the training is completed; the second device data includes second device parameters and a second leakage current corresponding to the second device parameters.
[0104] Specifically, the data enhancement model includes a generative adversarial network, the generative adversarial network includes a generator and a discriminator, and the training process of the generative adversarial network includes:
[0105] Using a random variable as an input of the generator, and outputting third device data; the third device data includes a third device parameter and a third leakage current corresponding to the third device parameter; the third device data and the first device data have similar distribution;
[0106] Inputting the first device data and the third device data into the discriminator, and outputting a first discrimination result corresponding to the first device data and a second discrimination result corresponding to the third device data;
[0107] The first parameters of the generator and the second parameters of the discriminator are alternately iteratively optimized until the first loss function of the generator and the second loss function of the discriminator are minimized to obtain the trained generative adversarial network; the second loss function is determined according to the first discrimination result and the second discrimination result, and the first loss function is determined according to the second discrimination result.
[0108] Specifically, the first device data and the third device data are input into the discriminator, and a first discrimination result corresponding to the first device parameter and a second discrimination result corresponding to the third device data are output, including:
[0109] Preprocessing the first device data to obtain fourth device data; the fourth device data includes fourth device parameters obtained by standardizing the first device parameters and fourth leakage current obtained by logarithmically processing the first leakage current;
[0110] The third device data and the fourth device data are input into the discriminator, and the first discrimination result corresponding to the fourth device data and the second discrimination result corresponding to the third device data are output.
[0111] Specifically, the first loss function is expressed as:
[0112]
[0113] Among them, the G loss is the first loss function, the z i is the i-th random variable, the G(z i ) is the third device data, and the D(G(z i )) is the second discrimination result, and the n fake represents the number of samples of the third device data;
[0114] The second loss function is expressed as:
[0115]
[0116] Among them, the D loss is the second loss function, the x i is the i-th first device data, the D(x i ) is the first discrimination result, and the n real Indicates the sample number of the first device data.
[0117] Specifically, the data enhancement model includes a boundary search generative adversarial network, a W-distance generative adversarial network, or a conditional generative adversarial network.
[0118] Specifically, the device also includes:
[0119] A training unit is used to train a deep neural network model using the first device data and the second device data, each round of training is verified using a validation set, and the deep neural network model that minimizes the third loss function on the validation set is retained until the third loss function is minimized to obtain a trained deep neural network model; the deep neural network model is used to predict leakage current.
[0120] Specifically, the activation function in the data augmentation model includes at least one of a sigmoid activation function, a tanh activation function or a relu activation function.
[0121] The embodiment of the present application provides a data enhancement device, wherein an acquisition unit is used to acquire first device data from simulation data or actual test data; the first device data includes a first device parameter and a first leakage current corresponding to the first device parameter; the first device parameter includes at least one of a channel length, a gate oxide thickness, a gate voltage, or a drain voltage; an output unit is used to input the first device data into a data enhancement model for training, and when the training is completed, output second device data; the second device data includes a second device parameter and a second leakage current corresponding to the second device parameter. In the embodiment of the present application, the time and cost of acquiring the first device data from simulation data or actual test data are high. By using a data enhancement model, the data can be enhanced to generate new data samples, that is, a large amount of second device data can be obtained, reducing the time and cost of data acquisition.
[0122] In another aspect, the present application provides a computer device, referring to Figure 4 , which is a structural diagram of a computer device provided in an embodiment of the present application, wherein the computer device includes a processor 310 and a memory 320:
[0123] The memory 320 is used to store program codes and transmit the program codes to the processor 310;
[0124] The processor 310 is used to execute the method provided in the above embodiment according to the instructions in the program code.
[0125] The computer device may include a terminal device or a server, and the aforementioned apparatus may be configured in the computer device.
[0126] On the other hand, an embodiment of the present application further provides a storage medium, wherein the storage medium is used to store a computer program, and the computer program is used to execute the method provided by the above embodiment.
[0127] A person skilled in the art can understand that all or part of the steps of implementing the above method embodiment can be completed by program instruction hardware, and the above program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the above storage medium can be at least one of the following media: read-only memory (English: Read-only Memory, abbreviated: ROM), RAM, magnetic disk or optical disk, etc. Various media that can store program codes.
[0128] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.
[0129] The above is only a preferred implementation of the present application. Although the present application has been disclosed as a preferred embodiment, it is not intended to limit the present application. Any technician familiar with the art can use the above disclosed methods and technical contents to make many possible changes and modifications to the technical solution of the present application without departing from the scope of the technical solution of the present application, or modify it into an equivalent embodiment of equivalent changes. Therefore, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present application without departing from the content of the technical solution of the present application still falls within the scope of protection of the technical solution of the present application.
Claims
1. A data enhancement method, characterized in that: include: Acquire first device data from simulation data or actual test data; the first device data includes a first device parameter and a first leakage current corresponding to the first device parameter; the first device parameter includes at least one of a channel length, a gate oxide layer thickness, a gate voltage or a drain voltage; The first device data is input into a data enhancement model for training, and when the training is completed, second device data is output; the second device data includes second device parameters and a second leakage current corresponding to the second device parameters.
2. The method according to claim 1, characterized in that The data enhancement model includes a generative adversarial network, the generative adversarial network includes a generator and a discriminator, and the training process of the generative adversarial network includes: Using a random variable as an input of the generator, and outputting third device data; the third device data includes a third device parameter and a third leakage current corresponding to the third device parameter; the third device data and the first device data have similar distribution; Inputting the first device data and the third device data into the discriminator, and outputting a first discrimination result corresponding to the first device data and a second discrimination result corresponding to the third device data; The first parameters of the generator and the second parameters of the discriminator are alternately iteratively optimized until the first loss function of the generator and the second loss function of the discriminator are minimized to obtain the trained generative adversarial network; the second loss function is determined according to the first discrimination result and the second discrimination result, and the first loss function is determined according to the second discrimination result.
3. The method according to claim 2, characterized in that Inputting the first device data and the third device data into the discriminator, and outputting a first discrimination result corresponding to the first device data and a second discrimination result corresponding to the third device data, including: Preprocessing the first device data to obtain fourth device data; the fourth device data includes fourth device parameters obtained by standardizing the first device parameters and fourth leakage current obtained by logarithmically processing the first leakage current; The third device data and the fourth device data are input into the discriminator, and the first discrimination result corresponding to the fourth device data and the second discrimination result corresponding to the third device data are output.
4. The method according to claim 2, characterized in that: The first loss function is expressed as: Among them, the G loss is the first loss function, the z i is the i-th random variable, the G(z i ) is the third device data, and the D(G(z i )) is the second discrimination result, and the n fake represents the number of samples of the third device data; The second loss function is expressed as: Among them, the D loss is the second loss function, the x i is the i-th first device data, the D(x i ) is the first discrimination result, and the n real Indicates the sample number of the first device data.
5. The method according to claim 2, characterized in that: The data enhancement model includes a boundary search generative adversarial network, a W-distance generative adversarial network, or a conditional generative adversarial network.
6. The method according to any one of claims 1 to 5, characterized in that: The method further comprises: The deep neural network model is trained using the first device data and the second device data until a third loss function is minimized to obtain a trained deep neural network model; the deep neural network model is used to predict leakage current.
7. The method according to any one of claims 1 to 5, characterized in that: The activation function in the data enhancement model includes at least one of a sigmoid activation function, a tanh activation function or a relu activation function.
8. A data enhancement device, characterized in that: include: an acquisition unit, configured to acquire first device data from simulation data or actual test data; the first device data comprising a first device parameter and a first leakage current corresponding to the first device parameter; the first device parameter comprising at least one of a channel length, a gate oxide layer thickness, a gate voltage or a drain voltage; An output unit is used to input the first device data into a data enhancement model for training, and output second device data when the training is completed; the second device data includes second device parameters and a second leakage current corresponding to the second device parameters.
9. A computer device, characterized in that: The computer device comprises a processor and a memory: The memory is used to store program codes and transmit the program codes to the processor; The processor is configured to execute the method according to any one of claims 1 to 7 according to the instructions in the program code.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store a computer program, and the computer program is used to execute the method according to any one of claims 1 to 7.