Neural network-based electrostatic protection DTSCR device performance prediction method

Abstract

The application discloses a kind of electrostatic protection DTSCR device performance prediction method based on neural network, solve the low efficiency, time-consuming and labor-consuming problems of traditional device simulation and experimental test method in prior art;The method comprises: establishing electrostatic protection DTSCR device model, and obtaining initial data set by simulating electrostatic protection DTSCR device model;Through TLP simulation to DTSCR device model and initial data set, the electrical characteristic curve is obtained, and then the characteristic parameter is obtained, and the data set is established according to the characteristic parameter;Multiple electrostatic protection DTSCR device performance prediction neural networks with the same framework and different parameters are constructed, and training is carried out, to obtain the network after training is completed;The characteristic parameters of the design window of electrostatic protection DTSCR device are predicted using the network after training is completed;The method realizes high efficiency, short cycle, and can quickly and conveniently establish the correlation between electrostatic protection DTSCR device and performance index.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of microelectronic devices and artificial intelligence, and particularly relates to a neural network-based electrostatic protection DTSCR device performance prediction method. BACKGROUND

[0002] The generation and discharge behavior of static charges is a very common electrical phenomenon, however, when an electrostatic discharge event occurs on an integrated circuit chip, due to the extremely small physical size of semiconductor devices, the chip is likely to be damaged under an ESD pulse. Therefore, ESD behavior will seriously deteriorate the reliability of IC products. Therefore, it is necessary to effectively protect the integrated circuit product from static electricity.

[0003] SCR devices are widely used for electrostatic protection due to their large discharge current capacity per unit area, small area and other advantages, and as most of the chip operating environment has changed to a low-voltage environment, the ESD electrostatic protection design window has also been reduced to a low-voltage environment. The high trigger voltage in the traditional SCR device limits its application in advanced low-voltage ESD protection engineering. Among the many variants of optimizing the trigger voltage of the SCR device, the diode-triggered silicon-controlled rectifier (DTSCR) can achieve a very low and adjustable trigger voltage, and is widely used in ESD protection of various advanced low-voltage integrated circuits. There are many devices improved according to the structure of the DTSCR device, so the method for predicting the performance of the DTSCR device has certain representativeness and research significance.

[0004] Currently, most DTSCR devices are designed through TCAD tools, first a device model is established through a numerical simulation tool, and then the electrical characteristics of the device are simulated. When simulating, physical models such as drift-diffusion transport model, mobility model, and non-complete ionization model need to be considered. In addition, the grid in the critical path also needs to be designed to be a little dense, while the grid in other parts needs to be relatively sparse. After a series of parameter modifications and structure modifications, a relatively satisfactory device structure corresponding to the electrical characteristics can be obtained. After the above work is completed, the structure parameters and process parameters corresponding to the best device are extracted, thereby reducing the high cost risk brought by subsequent wafer fabrication.

[0005] When a DTSCR device model is established based on TCAD, in order to obtain a more accurate and ideal structure, the calculation grid needs to be set densely, but this will bring the disadvantage of long simulation time and large calculation cost. When the required sample size is very large, the time will be multiplied, greatly increasing the time cost of scientific research. In addition, the simulation of the electrical characteristics of the DTSCR device based on the TCAD simulation software requires the technical personnel to master certain semiconductor physical knowledge, and the learning cost will also increase.

[0006] In summary, the existing simulation method of the DTSCR device is time-consuming, low in efficiency, and greatly affected by human factors. The neural network provides another choice for fast device performance prediction, which can automatically extract features from data, eliminate high entry costs and human factors, reduce calculation cost, and shorten the development cycle while ensuring accuracy. SUMMARY

[0007] The present application provides a neural network-based electrostatic protection DTSCR device performance prediction method, which solves the problems of low efficiency, time-consuming and labor-intensive of the traditional device simulation and experimental test method in the prior art, and realizes high efficiency, short cycle, and rapid and convenient establishment of the correlation between the electrostatic protection DTSCR device and the performance index.

[0008] The present application provides a neural network-based electrostatic protection DTSCR device performance prediction method, which solves the problems of low efficiency, time-consuming and labor-intensive of the traditional device simulation and experimental test method in the prior art, and realizes high efficiency, short cycle, and rapid and convenient establishment of the correlation between the electrostatic protection DTSCR device and the performance index.

[0009] An electrostatic protection DTSCR device model is established based on Sentaurus TCAD, and an initial data set is obtained by simulating the electrostatic protection DTSCR device model;

[0010] Through TLP simulation of the electrostatic protection DTSCR device model and the initial data set, an electrical characteristic curve is obtained, and parameter extraction is performed on the electrical characteristic curve to obtain characteristic parameters, and a data set is established according to the characteristic parameters;

[0011] A plurality of electrostatic protection DTSCR device performance prediction neural networks with the same framework and different parameters are constructed, each of the electrostatic protection DTSCR device performance prediction neural networks is trained according to the data set and a loss function, and a plurality of initially trained electrostatic protection DTSCR device performance prediction neural networks are obtained;

[0012] The plurality of initially trained electrostatic protection DTSCR device performance prediction neural networks are screened to obtain a trained electrostatic protection DTSCR device performance prediction neural network;

[0013] The characteristic parameters of the design window of the electrostatic protection DTSCR device are predicted by using the trained electrostatic protection DTSCR device performance prediction neural network.

[0014] In a possible implementation, the TLP simulation is performed on the electrostatic protection DTSCR device model to obtain an electrical characteristic curve, and a parameter extraction is performed on the electrical characteristic curve to obtain characteristic parameters, and a data set is established according to the characteristic parameters, including:

[0015] The structural parameters and process parameters of the electrostatic protection DTSCR device model are changed to obtain an electrical characteristic curve of the DTSCR device;

[0016] The characteristic parameters of the electrical characteristic curve are extracted, and the characteristic parameters are integrated into a data set;

[0017] The data set is preprocessed, and the preprocessed data set is divided into a training set, a test set, and a cross-validation set.

[0018] In a possible implementation, the data set is preprocessed, and the preprocessed data set is divided into a training set, a test set, and a cross-validation set, including:

[0019] Each data in the data set is traversed, and each data is executed as follows:

[0020] It is judged whether the data set meets the I / V characteristic curve standard, if not, the data is removed, if yes, the data is normalized to obtain preprocessed data;

[0021] Each processed data is integrated into a processed data set, and the processed data set is divided into a training set, a test set, and a cross-validation set according to a proportion.

[0022] In a possible implementation, each data in the data set includes four structural parameters and four characteristic parameters.

[0023] The four structural parameters include the length of the shallow trench isolation between the P-well / N+ active area and the N-well / P+ active area of the SCR part of the electrostatic protection DTSCR device model, the length of the shallow trench isolation between the N+ active area and the P+ active area in the same well of the SCR part, the length between the N-well of the SCR part and the N-well of the diode DIODE1 part, and the well spacing of the two diodes.

[0024] The four characteristic parameters include a trigger voltage Vt, a trigger current It, a holding voltage Vh, and a holding current Ih.

[0025] In a possible implementation, the framework of the electrostatic protection DTSCR device performance prediction neural network comprises an input layer, an input expansion layer, a convolution layer, a pooling layer, a full connection layer and an output layer connected in sequence.

[0026] The input layer is configured to receive input data.

[0027] The input expansion layer is configured to perform dimension expansion on the input data to obtain expanded dimension input data.

[0028] The convolution layer is configured to perform data feature extraction on the expanded dimension input data to obtain initial feature information.

[0029] The pooling layer is configured to remove redundant information from the initial feature information to obtain feature information.

[0030] The full connection layer is configured to classify the feature information to obtain a classification result.

[0031] The output layer is configured to output the classification result.

[0032] In a possible implementation, the convolution layer is a one-dimensional convolution layer, and the convolution kernel size is 3x1.

[0033] The pooling layer is a one-dimensional pooling layer, and the convolution kernel size is 2x1.

[0034] In a possible implementation, the loss function is represented as:

[0035]

[0036] wherein n represents the number of samples; y i represents an actual value of the i-th sample; represents a predicted value of the i-th sample.

[0037] In a possible implementation, the framework of the electrostatic protection DTSCR device performance prediction neural network with different parameters comprises an input layer, an input expansion layer, a convolution layer, a pooling layer, a full connection layer and an output layer connected in sequence, comprising:

[0038] The input layer and the output layer of the electrostatic protection DTSCR device performance prediction neural network are fixed parameters; the parameters of the input expansion layer, the convolution layer, the pooling layer and the full connection layer of the electrostatic protection DTSCR device performance prediction neural network are non-fixed parameters; the electrostatic protection DTSCR device performance prediction neural network with different parameters is obtained by randomly setting the parameters of the input expansion layer, the convolution layer, the pooling layer and the full connection layer.

[0039] In a possible implementation, the screening of the plurality of primary training completed electrostatic protection DTSCR device performance prediction neural networks to obtain a trained electrostatic protection DTSCR device performance prediction neural network comprises:

[0040] The loss value of each electrostatic protection DTSCR device performance prediction neural network at each training round is calculated, and the mean square error value of each electrostatic protection DTSCR device performance prediction neural network is calculated according to the plurality of loss values;

[0041] The electrostatic protection DTSCR device performance prediction neural network corresponding to the minimum mean square error value is determined as the trained electrostatic protection DTSCR device performance prediction neural network.

[0042] In a possible implementation, the training of each electrostatic protection DTSCR device performance prediction neural network according to the data set and the loss function comprises:

[0043] According to the data set, the electrostatic protection DTSCR device performance prediction neural network is trained by using an SGD optimizer method and the loss function.

[0044] The one or more technical solutions provided in the application have at least the following technical effects or advantages:

[0045] The application utilizes the characteristics of automatic feature extraction of neural network deep learning according to data, applies it to the prediction of electrostatic protection DTSCR device performance, quickly predicts the performance of the device, can adjust the input structure according to the prediction result, and achieves the effect of optimizing the performance of the device; on the other hand, the prediction method can provide a reference for engineers to design, and shorten the design cycle; in the experiment, the network structure used has very high prediction accuracy, and the mean square error of the electrical parameters is as low as 0.004, and the average prediction accuracy reaches 96.9%;

[0046] The first two steps of the application are work that needs to have microelectronic professional background knowledge, and once the prediction network is designed, it can be provided for non-professional technical personnel, so that they can quickly understand the DTSCR device design window size by using neural network, and accurately and quickly predict the performance indicators of the electrostatic protection DTSCR device;

[0047] The method provided in the application is to predict the design window of the electrostatic protection device, and the design window is required to be investigated by all ESD devices. The application is suitable for other electrostatic protection devices, provides a different research direction for the development of electrostatic protection ESD devices, and improves the research efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 A step flow chart of a neural network-based electrostatic protection DTSCR device performance prediction method is provided for the embodiments of the present application.

[0049] Figure 2 An electrostatic protection DTSCR device structure diagram is provided for the embodiments of the present application.

[0050] Figure 3 An SDE structure modeling schematic diagram of the electrostatic protection DTSCR device is provided for the embodiments of the present application.

[0051] Figure 4 An I / V characteristic curve of the electrostatic protection DTSCR device under TLP transmission line pulse test is provided for the embodiments of the present application.

[0052] Figure 5 A trigger point on the I / V characteristic curve of the electrostatic protection DTSCR device is found based on the Kneed method, which is provided for the embodiments of the present application.

[0053] Figure 6 A holding point on the I / V characteristic curve of the electrostatic protection DTSCR device is found based on the Kneed method, which is provided for the embodiments of the present application.

[0054] Figure 7 A structure schematic diagram of the neural network for predicting the performance of the electrostatic protection DTSCR device is provided for the embodiments of the present application.

[0055] Figure 8 A comparison between the predicted value and the true value of the trigger current It after the test set passes through the neural network is provided for the embodiments of the present application.

[0056] Figure 9 A comparison between the predicted value and the true value of the trigger voltage Vt after the test set passes through the neural network is provided for the embodiments of the present application.

[0057] Figure 10 A comparison between the predicted value and the true value of the holding current Ih after the test set passes through the neural network is provided for the embodiments of the present application.

[0058] Figure 11 A comparison between the predicted value and the true value of the holding voltage Vh after the test set passes through the neural network is provided for the embodiments of the present application. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the present application.

[0060] The application provides a neural network-based electrostatic protection DTSCR device performance prediction method, which refers to Figure 1 The method comprises the following steps S101-S105.

[0061] S101, an electrostatic protection DTSCR device model is established based on Sentaurus TCAD, and an initial data set is obtained by simulating the electrostatic protection DTSCR device model;

[0062] Exemplarily, referring to Figure 2 , Figure 3 and Figure 4 , the structure modeling of the DTSCR device is completed by using the SDE tool in the Sentaurus TCAD tool, the doping type and doping mode are added according to the design drawing, and finally the calculation grid is designed. The grid needs to be constantly modified, especially the SCR path grid in the DTSCR device needs to be encrypted; the length of the shallow trench isolation between the P well / N+ active area and the N well / P+ active area of the SCR part of the device is defined as LPath, the length of the shallow trench isolation between the N+ active area and the P+ active area in the well connected to the positive electrode of the SCR part is defined as LA_STI, the length of the shallow trench isolation between the N+ active area and the P+ active area in the well connected to the positive electrode of the SCR part is defined as LC_STI, the length of the active area is defined as Lactive, the length of the well edge area containing the shallow trench isolation is defined as Lspace, the length of the well distance from the device edge on the left and right sides of the device is defined as Louter, the length between the N well of the SCR part and the N well of the DIODE1 part is defined as LSCR_DIODE, the distance between the wells of the two diodes is defined as LD_D, the length of the shallow trench isolation in the diode well is defined as LDIODE_STI, the depth of the active area is defined as Pdiff / Ndiff, the depth of the well is defined as Hwell, the depth of the shallow trench isolation is defined as HSTI, and the depth of the substrate is defined as Hsub. After the structure is established, the process doping area and the grid division are completed by experience design, the well doping concentration of the SCR part is defined as Pwelldoping / Nwelldoping, the well doping concentration of the diode part is defined as DPwelldoping / DNwelldoping, and the doping concentration of the active area is defined as Pdiff_Doping / Ndiff_Doping. After the doping model is established, as shown in Figure 2 .

[0063] After the DTSCR device model is constructed, the device is simulated and analyzed by using the Sdevice tool in Sentaurus TCAD. Here, quasi-static test is used to replace TLP test. The characteristic curve obtained by this method is similar to that of TLP test, but the speed is faster. Then, the I / V characteristic curve is changed by adjusting the parameters and grid size, so that the curve value reaches a reasonable value. Here, the physical models used include mobility model, bandgap narrowing model, and recombination model. After the I / V characteristic curve is changed by adjusting the parameters and grid size, so that the curve value reaches a reasonable value, i.e., as shown in FIG. 1. Figure 3

[0064] In S102, the electrical characteristic curve is obtained by TLP simulation on the electrostatic protection DTSCR device model and the initial data set, and the characteristic parameters are extracted from the electrical characteristic curve to establish the data set according to the characteristic parameters.

[0065] Here, each piece of data in the data set includes 4 structure parameters and 4 characteristic parameters. The structure parameters include the length of the shallow trench isolation between the P-well / N+ active region and the N-well / P+ active region of the SCR part of the electrostatic protection DTSCR device model, the length of the shallow trench isolation between the N+ active region and the P+ active region in the same well of the SCR part, the length between the N-well of the SCR part and the N-well of the diode DIODE1 part, and the well spacing of the two diodes. The characteristic parameters include the trigger voltage Vt, the trigger current It, the holding voltage Vh, and the holding current Ih.

[0066] Specifically, in step S102, the electrical characteristic curve is obtained by TLP simulation on the electrostatic protection DTSCR device model, and the characteristic parameters are extracted from the electrical characteristic curve to establish the data set according to the characteristic parameters, including the following steps S1021 to S1023.

[0067] S1021, the structure parameters and process parameters of the electrostatic protection DTSCR device model are changed to obtain the electrical characteristic curve of the DTSCR device;

[0068] S1022, the characteristic parameters of the electrical characteristic curve are extracted and integrated into a data set;

[0069] S1023, the data set is preprocessed, and the preprocessed data set is divided into a training set, a test set, and a cross-validation set.

[0070] Here, the data set is preprocessed, and the preprocessed data set is divided into a training set, a test set, and a cross-validation set, including:

[0071] (1) Traverse each piece of data in the data set, and perform the following operations on each piece of data: ​

[0072] (2) judging whether the data set meets the I / V characteristic curve standard, if not, eliminating the data; if yes, normalizing the data to obtain pre-processed data;

[0073] (3) integrating each processed data into a processed data set, and dividing the processed data set according to the proportion to obtain a training set, a test set and a cross-validation set.

[0074] For example, a data set is established by simulating the TLP test of a DTSCR device. The structural parameters and process parameters of the design are changed on the basis of the DTSCR device model to obtain the electrical characteristic curve of the DTSCR device, the characteristic parameters on the curve are extracted, and the required parameters are integrated into a data set. After preprocessing the data set, the data set is divided into a training set, a test set and a cross-validation set.

[0075] On the established DTSCR device model, the structural parameters and process parameters are appropriately changed to obtain different device structures and electrical characteristic results. After comparing the effects of various parameter changes on the electrical characteristics, the following four parameters are determined as the input neurons of the neural network: the length LPath of the shallow trench isolation between the P well / N+ active region and the N well / P+ active region of the SCR part of the device, the length LA / C_STI of the shallow trench isolation between the N+ active region and the P+ active region in the same well of the SCR part, the length LSCR_DIODE between the N well of the SCR part and the N well of the DIODE1 part, and the well spacing LD_D of the two diodes.

[0076] After determining the input neurons, the four values are set. Lpath takes values of 0.6, 0.7, 0.8, 0.9 and 1.0; LA / C_STI takes values of 0.8, 0.9, 1.0, 1.1 and 1.2; LSCR_DIODE takes values of 1.2, 1.3, 1.4, 1.5 and 1.6; and LD_D takes values of 1.2, 1.3, 1.4, 1.5 and 1.6. The above data is simulated, and finally the Svisual tool in Sentaurus TCAD is used to collect the I / V characteristic curves corresponding to each device structure.

[0077] Reference Figure 5 and Figure 6, the Kneed method of Python statement is used to extract the electrical parameters, and the extracted curve needs to be reversed in the vertical coordinate and horizontal coordinate value, because the Kneed method is to find the inflection point on the convex set and concave set function, and after the curve data is reversed, the trigger point and the holding point can be extracted more conveniently. The electrical parameters are Vt trigger voltage, It trigger current, Vh holding voltage and Ih holding current. After the electrical parameters are extracted, the data set can be established, and one sample of the data set includes four structure parameters and four electrical parameters, the structure parameters are the four input neurons mentioned above, and the four electrical parameters are Vt trigger voltage, It trigger current, Vh holding voltage and Ih holding current. Eight data are one sample, and many samples constitute a data set.

[0078] There are many unsuitable data in the data set, so data preprocessing is needed. First, the data that do not meet the I / V characteristic curve standard of ESD devices is removed, which can be removed by filtering data values in the Excel table. Then, the data is normalized by using Python statements to calculate the minimum and maximum values of the feature column to realize normalization.

[0079] After data preprocessing, the data in the 564 data sets are divided into training set, test set and cross-validation set according to the ratio of 8:1:1.

[0080] S103, constructing a plurality of static protection DTSCR device performance prediction neural networks with the same framework and different parameters, training each static protection DTSCR device performance prediction neural network according to the data set and the loss function, and obtaining a plurality of initially trained static protection DTSCR device performance prediction neural networks;

[0081] Specifically, in step S103, the plurality of static protection DTSCR device performance prediction neural networks with the same framework and different parameters are constructed, including the following steps S1031 to S1033.

[0082] S1031, the framework of the static protection DTSCR device performance prediction neural network includes an input layer, an input expansion layer, a convolution layer, a pooling layer, a full connection layer and an output layer connected in sequence;

[0083] S1032, the input layer and the output layer of the static protection DTSCR device performance prediction neural network are fixed parameters;

[0084] S1033, the parameters of the input expansion layer, the convolution layer, the pooling layer and the fully connected layer of the electrostatic protection DTSCR device performance prediction neural network are non-fixed parameters, the parameters of the input expansion layer, the convolution layer, the pooling layer and the fully connected layer are randomly set, a plurality of parameter values of the electrostatic protection DTSCR device performance prediction neural network are obtained, and then a plurality of electrostatic protection DTSCR device performance prediction neural networks with the same framework and different parameters are obtained.

[0085] Specifically, in step S103, the framework of the electrostatic protection DTSCR device performance prediction neural network includes an input layer, an input expansion layer, a convolution layer, a pooling layer, a fully connected layer and an output layer connected in sequence.

[0086] Among them, the input layer and the output layer of the electrostatic protection DTSCR device performance prediction neural network are fixed parameters; the parameters of the input expansion layer, the convolution layer, the pooling layer and the fully connected layer of the electrostatic protection DTSCR device performance prediction neural network are non-fixed parameters; the electrostatic protection DTSCR device performance prediction neural networks with different parameters are obtained by randomly setting the parameters of the input expansion layer, the convolution layer, the pooling layer and the fully connected layer.

[0087] Here, the convolution layer is a one-dimensional convolution layer, and the convolution kernel size is 3x1; the pooling layer is a one-dimensional pooling layer, and the convolution kernel size is 2x1.

[0088] Specifically, in step S103, each electrostatic protection DTSCR device performance prediction neural network is trained according to the data set and the loss function, including: according to the data set, using the SGD optimizer method and the loss function to train the electrostatic protection DTSCR device performance prediction neural network.

[0089] Specifically, in step S103, the loss function is expressed as:

[0090]

[0091] Wherein, n represents the number of samples; y i represents the actual value of the i-th sample; represents the predicted value of the i-th sample.

[0092] For example, refer to Figure 6The electrostatic protection DTSCR device performance prediction neural network is constructed based on a convolutional neural network, and the architecture comprises an input layer, an input expansion layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer. The input layer comprises four neurons, namely LPath, LA / C_STI, LSCR_DIODE, and LD_D. The input layer is connected to the input expansion layer, which is composed of n fully connected layers. After each linear layer, a normalization layer is added. Then, the feature extraction process is performed, which comprises m times of convolution and q times of pooling. The convolution kernel size is 3x1, and the pooling kernel size is 2x1. The convolution is filled to maintain the size before and after the convolution. In addition, a normalization layer is added after each convolution. After the above process, the neural network is flattened to one channel, and the feature classification process is performed, which comprises n fully connected layers. After each linear layer, a normalization layer is added. All processes use the ReLU activation function. Finally, the classifier leaves four feature values, namely the trigger current, the trigger voltage, the holding current, and the holding voltage. In the entire electrostatic protection DTSCR device performance prediction neural network, the values of the network layers n, m, and q need to be determined by comparing the loss function, i.e., the mean square error, after subsequent training.

[0093] The modified iterative electrostatic protection DTSCR device performance prediction neural network structure is used to input the training set and the cross-validation set obtained from the simulation DTSCR device. The training result is evaluated by observing and comparing the loss function. After the network training, the network structure modification is stopped, and the network with the minimum loss function, i.e., the mean square error, is fixed as the final prediction model.

[0094] In a specific embodiment provided by the present application, the network structure parameters of the electrostatic protection DTSCR device prediction neural network are randomly initialized, and a first group of neural network structures is designed. The input layer and the output layer are fixed, and the middle layers need to be designed, such as 2 layers of input expansion layer, 3 layers of convolutional layer, 3 layers of pooling layer, and 3 layers of fully connected layer. During the modification iteration process, the number of layers needs to be modified according to the setting of the control group.

[0095] The training set and cross-validation set obtained by the simulation DTSCR device are input into the neural network prediction model, and the training and intermediate verification optimization are performed; the early stopping method is used for training to improve the training efficiency; the cross-validation set is appropriately arranged in the training set test for verification, and the loss function at each training time is obtained; the SGD optimizer is used in the scheme, and the optimizer optimizes the network according to the loss function, until the neural network model converges, and the trained electrostatic protection DTSCR device neural network prediction model is obtained, if the training is restarted in step 4, the learning rate and iteration number of the optimizer and other data need to be modified, and the network is modified by the internal weight bias, and the network structure is not modified.

[0096] S104, screening the plurality of initially trained electrostatic protection DTSCR device performance prediction neural networks to obtain the trained electrostatic protection DTSCR device performance prediction neural network.

[0097] Specifically, in step S104, the plurality of initially trained electrostatic protection DTSCR device performance prediction neural networks are screened to obtain the trained electrostatic protection DTSCR device performance prediction neural network, including the following steps S1041 and S1042.

[0098] S1041, calculating the loss value of each electrostatic protection DTSCR device performance prediction neural network at each training round, and calculating the mean square error value of each electrostatic protection DTSCR device performance prediction neural network according to a plurality of loss values;

[0099] S1042, determining that the electrostatic protection DTSCR device performance prediction neural network corresponding to the minimum mean square error value is the trained electrostatic protection DTSCR device performance prediction neural network.

[0100] Exemplary, the final design of the electrostatic protection DTSCR device performance prediction neural network is as follows Figure 7 , the input layer has 4 neurons, the expansion layer has 3 layers, the convolution and pooling layer has 2 layers, and the last classifier full connection layer has 3 layers.

[0101] The test set is input into the trained prediction model, if the loss function and the prediction accuracy meet the predetermined requirements, the final trained model can be used as a tool for predicting the design window feature points of the DTSCR device, otherwise the retraining is restarted.

[0102] The test result of the neural network is that the average prediction accuracy of the four characteristic values reaches 96.9%, and the mean square error mean is 0.004, and the result is relatively accurate.

[0103] Reference Figure 8- Figure 11, are comparison charts of the true value and the neural network prediction value in the test set test process, respectively, the trigger current, the trigger voltage, the holding current and the holding voltage. It can be observed that the prediction result is basically consistent with the original data, which proves the feasibility of the performance prediction method of the electrostatic protection DTSCR device based on the neural network.

[0104] S105, the feature parameters of the design window of the electrostatic protection DTSCR device are predicted by using the trained electrostatic protection DTSCR device performance prediction neural network.

[0105] Specifically, the trained electrostatic protection DTSCR device performance prediction neural network satisfies the accuracy formula. The specific accuracy expression formula is:

[0106]

[0107] Each embodiment in the specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other. Each embodiment mainly describes the difference from other embodiments. The whole or part of the present application can be used in a plurality of general or special computer system environments or configurations. For example: personal computer, server computer, handheld device or portable device, tablet device, mobile communication terminal, multi-processor system, microprocessor-based system, programmable electronic device, network PC, small computer, large computer, distributed computing environment including any of the above systems or devices, etc.

[0108] The above embodiments are only used to illustrate the technical solutions of the present application, and are not limited to the present application. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can be modified, or some or all of the technical features can be replaced by equivalent replacements; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the present application.

Claims

1. A neural network-based electrostatic protection DTSCR device performance prediction method, characterized by, The application relates to a method for predicting the performance of an electrostatic protection DTSCR device. The method comprises the following steps: establishing an electrostatic protection DTSCR device model based on Sentaurus TCAD, and obtaining an initial data set by simulating the electrostatic protection DTSCR device model; obtaining an electrical characteristic curve by TLP simulation on the electrostatic protection DTSCR device model and the initial data set, and extracting parameters of the electrical characteristic curve to obtain characteristic parameters, and establishing a data set according to the characteristic parameters; each piece of data in the data set comprises four structure parameters and four characteristic parameters; the four structure parameters comprise the length of shallow trench isolation between the P-well / N+ active region and the N-well / P+ active region of the SCR part of the electrostatic protection DTSCR device model, the length of shallow trench isolation between the N+ active region and the P+ active region in the same well of the SCR part, the length between the N-well of the SCR part and the N-well of the diode DIODE1 part, and the well spacing of the two diodes; the four characteristic parameters comprise a trigger voltage Vt, a trigger current It, a holding voltage Vh and a holding current Ih; a plurality of electrostatic protection DTSCR device performance prediction neural networks with the same framework and different parameters are constructed, each of the electrostatic protection DTSCR device performance prediction neural networks is trained according to the data set and a loss function, and a plurality of initially trained electrostatic protection DTSCR device performance prediction neural networks are obtained; the framework of the plurality of electrostatic protection DTSCR device performance prediction neural networks with the same framework and different parameters comprises an input layer, an input expansion layer, a convolution layer, a pooling layer, a full connection layer and an output layer connected in sequence; wherein the input layer and the output layer of the electrostatic protection DTSCR device performance prediction neural network are fixed parameters; the parameters of the input expansion layer, the convolution layer, the pooling layer and the full connection layer of the electrostatic protection DTSCR device performance prediction neural network are non-fixed parameters; the electrostatic protection DTSCR device performance prediction neural networks with different parameters are obtained by randomly setting the parameters of the input expansion layer, the convolution layer, the pooling layer and the full connection layer; the plurality of initially trained electrostatic protection DTSCR device performance prediction neural networks are screened to obtain trained electrostatic protection DTSCR device performance prediction neural networks; 2. The neural network-based electrostatic protection DTSCR device performance prediction method of claim 1, wherein, the trained electrostatic protection DTSCR device performance prediction neural networks are used to predict the characteristic parameters of a design window of an electrostatic protection DTSCR device. The TLP simulation on the electrostatic protection DTSCR device model obtains an electrical characteristic curve, the characteristic parameters of the electrical characteristic curve are extracted, a data set is established according to the characteristic parameters, and the method comprises the following steps: the structure parameters and process parameters of the electrostatic protection DTSCR device model are changed to obtain the electrical characteristic curve of the DTSCR device; the characteristic parameters of the electrical characteristic curve are extracted, and the characteristic parameters are integrated into a data set; the data set is preprocessed, and the preprocessed data set is divided into a training set, a test set and a cross-validation set.

3. The neural network-based electrostatic protection DTSCR device performance prediction method of claim 2, wherein, The pre-processing of the data set, the pre-processed data set is divided into a training set, a test set and a cross-validation set, comprising: Traverse each data in the data set, and perform the following operations on each data: Determine whether the data set meets the I / V characteristic curve standard, if not, eliminate the data; if yes, normalize the data to obtain pre-processed data; Integrate each processed data into a processed data set, and divide the processed data set according to the proportion to obtain the training set, the test set and the cross-validation set.

4. The neural network-based electrostatic protection DTSCR device performance prediction method of claim 1, wherein, The framework of the electrostatic protection DTSCR device performance prediction neural network comprises an input layer, an input expansion layer, a convolution layer, a pooling layer, a full connection layer and an output layer connected in turn; The input layer is used for receiving input data; The input expansion layer is used for dimension expansion of the input data to obtain expanded dimension input data; The convolution layer is used for data feature extraction of the expanded dimension input data to obtain initial feature information; The pooling layer is used for removing redundant information of the initial feature information to obtain feature information; The full connection layer is used for classifying the feature information to obtain a classification result; The output layer is used for outputting the classification result.

5. The neural network-based electrostatic protection DTSCR device performance prediction method of claim 4, wherein, The convolution layer is a one-dimensional convolution layer, and the convolution kernel size is 3x1; The pooling layer is a one-dimensional pooling layer, and the convolution kernel size is 2x1.

6. The neural network-based electrostatic protection DTSCR device performance prediction method of claim 1, wherein, The loss function is represented as: ； wherein, represents the number of samples; represents the actual value of the th sample; represents the predicted value of the th sample.

7. The neural network-based electrostatic protection DTSCR device performance prediction method of claim 1, wherein, The multiple electrostatic protection DTSCR device performance prediction neural networks are screened to obtain a trained electrostatic protection DTSCR device performance prediction neural network, comprising: Calculate the loss value of each electrostatic protection DTSCR device performance prediction neural network at each training round, and calculate the mean square error value of each electrostatic protection DTSCR device performance prediction neural network according to multiple loss values; The electrostatic protection DTSCR device performance prediction neural network corresponding to the minimum mean square error value is determined as the trained electrostatic protection DTSCR device performance prediction neural network.

8. The neural network-based electrostatic protection DTSCR device performance prediction method of claim 1, wherein, The training of each electrostatic protection DTSCR device performance prediction neural network according to the data set and the loss function, comprising: According to the data set, the SGD optimizer method and the loss function are used to train the electrostatic protection DTSCR device performance prediction neural network.

Images