A CNN-based method for predicting electrical characteristics of FDSOI devices
Through the prediction method based on CNN network, the problem of large amount of calculation and time-consuming calculation of electrical characteristics prediction of FDSOI devices is solved, and efficient and accurate electrical characteristics prediction is achieved, improving prediction efficiency.
Patent Information
- Application Number
- CN202210095126.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-01-26
- Publication Date
- 2025-03-14
- Estimated Expiration
- 2042-01-26
AI Technical Summary
The prediction of electrical characteristics of FDSOI devices is computationally expensive and takes a long time in traditional research.
Using a prediction method based on CNN network, a CNN network prediction model is constructed by obtaining the physical parameters of multiple sets of FDSOI devices and their corresponding electrical characteristics, and a CNN network prediction model is iteratively updated with the regression loss function to achieve efficient prediction of the electrical characteristics of FDSOI devices.
It improves the efficiency and accuracy of the electrical characteristics prediction of FDSOI devices, reduces the calculation amount and time-consuming, and significantly improves the prediction efficiency.
Smart Images
Figure CN114528805B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the fields of microelectronic technology and artificial intelligence technology, and in particular to a method for predicting electrical characteristics of a FDSOI (fully depleted silicon on insulator) device based on a CNN network (convolutional neural network). Background Art
[0002] After the feature size of integrated circuit technology is reduced to 45nm, various non-ideal effects, such as short channel effect and drain-induced barrier lowering effect, will seriously affect the performance of bulk silicon MOS tubes. SOI (Silicon on Insulator) technology was proposed to solve this problem. Silicon on insulator means adding insulating material between silicon transistors, that is, using the device structure of "silicon-insulator-silicon", which greatly reduces the parasitic capacitance between the two, thereby improving device performance. SOI technology is divided into FDSOI technology and PDSOI (Partially Depleted Silicon on Insulator) technology. Compared with PDSOI technology, FDSOI technology has a thinner silicon film thickness. When the device is working, the depletion layer below fills the entire silicon layer, which can better control the short channel effect, limit the source-drain junction depth, and improve the subthreshold characteristics. Therefore, FDSOI technology has become a good choice for small-size processes, and the research on FDSOI devices is of great significance.
[0003] Traditional research on FDSOI technology is mostly based on simulation software, which is computationally intensive and time-consuming. However, as one of the representative algorithms of deep learning, the CNN network can automatically extract features based on data and has representation learning capabilities, and is receiving more and more attention.
[0004] In recent years, CNN networks have made great achievements in research fields such as speech recognition, image recognition, image segmentation, and natural language processing. Its rapid development has provided other ideas for the prediction of device electrical properties. Summary of the invention
[0005] The purpose of the present invention is to provide a method for predicting the electrical characteristics of FDSOI devices based on a CNN network to solve the problem that the prediction of the electrical characteristics of FDSOI devices is computationally intensive and time-consuming in the traditional research process.
[0006] In order to achieve the above object, the present invention adopts the following technical solutions to achieve it.
[0007] A method for predicting electrical characteristics of FDSOI devices based on a CNN network comprises the following steps:
[0008] Step 1, obtaining multiple groups of physical parameters of FDSOI devices and their corresponding electrical characteristics as sample sets;
[0009] Among them, the physical parameters of FDSOI devices include gate length, buried oxide layer thickness, body region doping concentration and gate metal work function; the electrical characteristics of FDSOI devices include threshold voltage, transconductance, subthreshold swing and current switching ratio;
[0010] Step 2, randomly divide the data in the sample set into training set, cross-validation set and test set according to proportion;
[0011] Step 3, construct a CNN network prediction model;
[0012] Step 4: Input the training set data into the CNN network prediction model, use the regression loss function to iteratively update the parameters of the CNN network prediction model, and obtain the trained CNN network prediction model; then input the cross-validation set data into the trained CNN network prediction model, and finally obtain the trained CNN network prediction model; input the test set data into the trained CNN network prediction model to evaluate the prediction accuracy and generalization ability of the CNN network prediction model;
[0013] Step 5: Input the physical parameters of the FDSOI device to be predicted into the trained CNN network prediction model, and the trained CNN network prediction model outputs the prediction results of the electrical characteristics of the FDSOI device.
[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: the use of the CNN network solves the problem of large amount of calculation and long time consumption in the traditional research process of predicting the electrical characteristics of FDSOI devices, and has a higher accuracy rate, thereby improving the efficiency of prediction. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0016] Figure 1 It is a flow chart of the method for predicting electrical characteristics of FDSOI devices based on CNN network of the present invention;
[0017] Figure 2 is a schematic diagram of the structure of the FDSOI device selected in the embodiment of the present invention;
[0018] Figure 3 is a transfer characteristic curve diagram of the FDSOI device selected in the embodiment of the present invention under the parameters of Table 1;
[0019] Figure 4 It is a schematic diagram of the structure of the CNN network prediction model of the present invention;
[0020] Figure 5 A comparison chart of the predicted value and the actual value of the electrical characteristics of the test collector in the embodiment;
[0021] Figure 5 (a) is a comparison diagram of the predicted value and the true value of the switch current ratio of the test set in the embodiment;
[0022] Figure 5 (b) is a comparison chart of the predicted value and the true value of the subthreshold swing of the test set in the embodiment;
[0023] Figure 5 (c) is a comparison diagram of the predicted value and the true value of the threshold voltage of the test set in the embodiment;
[0024] Figure 5 (d) is a comparison chart of the predicted value and the true value of the test set transconductance in the embodiment;
[0025] Figure 6 A classification statistical diagram of the prediction accuracy of the test collector electrical characteristics in the embodiment;
[0026] Figure 6 (a) is a classification statistical diagram of the switch current ratio prediction accuracy of the test set in the embodiment;
[0027] Figure 6 (b) is a classification statistical diagram of the subthreshold swing prediction accuracy of the test set in the embodiment;
[0028] Figure 6 (c) is a classification statistical diagram of the prediction accuracy of the threshold voltage of the test set in the embodiment;
[0029] Figure 6 (d) is a classification statistical diagram of the threshold voltage prediction accuracy of the test set in the embodiment. DETAILED DESCRIPTION
[0030] The embodiments of the present invention will be described in detail below with reference to examples, but those skilled in the art will understand that the following examples are only used to illustrate the present invention and should not be construed as limiting the scope of the present invention.
[0031] refer to Figure 1 , a method for predicting electrical characteristics of FDSOI devices based on CNN network, comprising the following steps:
[0032] Step 1, obtaining 900 sets of physical parameters of FDSOI devices and their corresponding electrical characteristics as a sample set;
[0033] Among them, the physical parameters of FDSOI devices include gate length, buried oxide layer thickness, body region doping concentration and gate metal work function; the electrical characteristics of FDSOI devices include threshold voltage, transconductance, subthreshold swing and current switching ratio;
[0034] That is, each set of data includes gate length, buried oxide layer thickness, body region doping concentration, gate metal work function, threshold voltage, transconductance, subthreshold swing, and current switching ratio;
[0035] Specifically, the threshold voltage is the voltage at which the FDSOI device has a drain current of 1×10 -5 Gate voltage value at A / μm;
[0036] The transconductance is the maximum transconductance value in the transfer characteristic curve of the FDSOI device. The transconductance calculation formula is shown in formula (1):
[0037]
[0038] In formula (1), g m is the transconductance, I d is the leakage current, V g is the gate voltage;
[0039] The subthreshold swing calculation formula is shown in formula (2):
[0040]
[0041] In formula (2), SS is the subthreshold swing, I d is the leakage current in the transfer characteristic curve of the FDSOI device when the gate voltage is the threshold voltage, V g is the gate voltage when the gate voltage is the threshold voltage in the transfer characteristic curve of the FDSOI device;
[0042] The current switching ratio is the ratio of the leakage current of the FDSOI device when the gate voltage is 1V to the leakage current when the gate voltage is 0V.
[0043] refer to Figure 2 , is a schematic diagram of the structure of the FDSOI device used in this embodiment;
[0044] refer to Figure 3 , the FDSOI device used in this embodiment has the following parameters in the linear region (V D =0.1V) and saturation region (V D =1.0V) is good and can be used to predict electrical characteristics.
[0045] Table 1
[0046] Gate length(L) 28nm Back gate concentration (Nbp) <![CDATA[2e18cm -3 ]]> Buried oxide layer thickness (Tbox) 20nm Substrate doping concentration (Nsub) <![CDATA[1e14cm -3 ]]> Gate dielectric thickness (Tox) 1.2nm Body doping concentration (Nbd) <![CDATA[1e18cm -3 ]]> Back gate thickness (Tbp) 25nm Source region doping concentration (Ns) <![CDATA[4.4e20cm -3 ]]> Metal work function (WF) 4.52eV Drain doping concentration (Nd) <![CDATA[1e20cm -3 ]]>
[0047] The physical parameters in the 900 sets of data in the sample set are designed by ourselves, and the corresponding electrical characteristics of each set of data are obtained by TCAD software simulation; the values of the physical parameters in the 900 sets of data in the sample set are as follows:
[0048] The gate lengths are 16nm, 20nm, 24nm, 28nm, and 32nm respectively; the buried oxide layer thicknesses are 10nm, 20nm, 30nm, 40nm, 50nm, and 60nm respectively; and the body doping concentrations are 1×10 18 cm -3 , 2×10 18 cm -3 , 3×10 18 cm -3 , 4×10 18 cm -3 , 5×10 18 cm -3 , the gate metal work function takes values of 4.44eV, 4.48eV, 4.52eV, 4.56eV, 4.60eV, and 4.64eV respectively.
[0049] Step 2: randomly divide the 900 sets of data in the sample set into training set, cross-validation set and test set according to proportion;
[0050] Specifically, 60% of the sample set is the training set, 20% is the cross-validation set, and the remaining 20% is the test set, that is, 540 sets of data are used as training sets, 180 sets of data are used as cross-validation sets, and 180 sets of data are used as test sets;
[0051] The training set is used to train the parameters of the CNN network prediction model, the cross-validation set is used to verify whether the CNN network prediction model is overfitting, and the test set is used to evaluate the accuracy and generalization ability of the CNN network prediction model;
[0052] Step 3, construct a CNN network prediction model;
[0053] Specifically, the CNN network prediction model inputs the physical parameters of the FDSOI device, predicts the electrical characteristics of the FDSOI device and outputs them;
[0054] refer to Figure 4 , the CNN network prediction model includes an input expansion module, a transposed convolution module, a convolution module and a fully connected module connected in sequence;
[0055] The input expansion module consists of 3 layers, each of which consists of a Linear layer (fully connected layer), a BN layer (batch normalization layer), and a ReLU activation function.
[0056] The transposed convolution module consists of a ConvT layer (transposed convolution layer), a BN layer, a ConvT layer, and a BN layer connected in sequence;
[0057] The convolution module consists of 3 layers, each of which consists of a convolutional layer (Conv layer), a BN layer, and a ReLU activation function connected in sequence;
[0058] The fully connected module consists of 3 layers, each of which consists of a linear layer, a batch normalization layer, and a ReLU activation function connected in sequence;
[0059] The input expansion module is used to expand the input data to a higher dimension, the transposed convolution module is used to increase the size of the feature size, the convolution module is used to extract data features, and the fully connected module is used to map the distributed features learned by the network model to the sample label space;
[0060] The BN layer is used to speed up the convergence of the CNN network prediction model and prevent overfitting. A layer of ReLU activation function is added to the input expansion module, convolution module and fully connected module to complete the nonlinear change of data and avoid gradient explosion and gradient disappearance.
[0061] Step 4: Input the training set data into the CNN network prediction model, use the regression loss function to iteratively update the parameters of the CNN network prediction model, and obtain the trained CNN network prediction model; then input the cross-validation set data into the trained CNN network prediction model, and finally obtain the trained CNN network prediction model; input the test set data into the trained CNN network prediction model to evaluate the prediction accuracy and generalization ability of the CNN network prediction model;
[0062] Specifically, the gate length, buried oxide layer thickness, body region doping concentration, and gate metal work function of each set of data are used as input data, and the threshold voltage, transconductance, subthreshold swing, and current switching ratio of each set of data are used as output labels;
[0063] Specifically, the regression loss function is a mean square error (MSE) function, which calculates the sum of squares of the difference between the predicted value and the target value. The calculation formula is shown in formula (3):
[0064]
[0065] In formula (3), n is the number of samples, y i represents the true value of the i-th sample, is the predicted value of the i-th sample.
[0066] The test set data is input into the trained CNN network prediction model, and the trained CNN network prediction model outputs the predicted electrical characteristics. The predicted electrical characteristics are compared with the actual electrical characteristics. The comparison results are as follows: Figure 5 As shown in Figure 6 shown.
[0067] refer to Figure 5 (a) Figure 5 (b) Figure 5(c) Figure 5 (d) The predicted electrical characteristics have a good fitting effect with the actual electrical characteristics, and the mean square error of the test set is 0.001; among them, the fitting effects of transconductance and subthreshold swing are better than those of threshold voltage and current switching ratio.
[0068] refer to Figure 6 (a) Figure 6 (b) Figure 6 (c) Figure 6 (d) The prediction accuracy of the four electrical characteristics is relatively high, and the prediction accuracy is basically above 90%. Among the 180 test set data, more than 90% of the data sets have a prediction accuracy of 95% for the four electrical characteristics. Among them, the subthreshold swing prediction accuracy of all test set data is above 95%.
[0069] Step 5: Input the physical parameters of the FDSOI device to be predicted into the trained CNN network prediction model, and the trained CNN network prediction model outputs the prediction results of the electrical characteristics of the FDSOI device.
[0070] Although the present invention has been described in detail in general terms and in specific embodiments in this specification, it is obvious to those skilled in the art that some modifications or improvements may be made to the present invention. Therefore, these modifications or improvements made without departing from the spirit of the present invention all fall within the scope of protection claimed by the present invention.
Claims
1. A method for predicting electrical characteristics of FDSOI devices based on CNN network, characterized in that: The following steps are involved: Step 1, obtaining multiple groups of physical parameters of FDSOI devices and their corresponding electrical characteristics as sample sets; Among them, the physical parameters of FDSOI devices include gate length, buried oxide layer thickness, body region doping concentration and gate metal work function; the electrical characteristics of FDSOI devices include threshold voltage, transconductance, subthreshold swing and current switching ratio; Step 2, randomly divide the data in the sample set into training set, cross-validation set and test set according to proportion; Step 3, constructing a CNN network prediction model; the CNN network prediction model inputs the physical parameters of the FDSOI device, predicts the electrical characteristics of the FDSOI device and outputs them; The CNN network prediction model comprises an input expansion module, a transposed convolution module, a convolution module and a fully connected module connected in sequence; The input expansion module consists of 3 layers, each of which consists of a linear layer, a batch normalization layer, and a ReLU activation function connected in sequence; The transposed convolution module consists of a ConvT layer, a BN layer, a ConvT layer, and a BN layer connected in sequence; The convolution module consists of 3 layers, each of which consists of a convolution layer, a BN layer, and a ReLU activation function connected in sequence; The fully connected module consists of 3 layers, each of which consists of a linear layer, a batch normalization layer, and a ReLU activation function connected in sequence; Step 4: Input the training set data into the CNN network prediction model, use the regression loss function to iteratively update the parameters of the CNN network prediction model, and obtain the trained CNN network prediction model; then input the cross-validation set data into the trained CNN network prediction model, and finally obtain the trained CNN network prediction model; input the test set data into the trained CNN network prediction model to evaluate the prediction accuracy and generalization ability of the CNN network prediction model; Step 5: Input the physical parameters of the FDSOI device to be predicted into the trained CNN network prediction model, and the trained CNN network prediction model outputs the prediction results of the electrical characteristics of the FDSOI device.
2. The method for predicting electrical characteristics of FDSOI devices based on CNN network according to claim 1, characterized in that: The threshold voltage, transconductance, subthreshold swing and current switching ratio in step 1 are as follows: specifically, the threshold voltage is the voltage at which the FDSOI device is subjected to a drain current of 1×10 -5 Gate voltage value at A / μm; The transconductance is the maximum transconductance value in the transfer characteristic curve of the FDSOI device. The transconductance calculation formula is shown in formula (1): In formula (1), g m is the transconductance, I d is the leakage current, V g is the gate voltage; The subthreshold swing calculation formula is shown in formula (2): In formula (2), SS is the subthreshold swing, I d is the leakage current in the transfer characteristic curve of the FDSOI device when the gate voltage is the threshold voltage, V g is the gate voltage when the gate voltage is the threshold voltage in the transfer characteristic curve of the FDSOI device; The current switching ratio is the ratio of the leakage current of the FDSOI device when the gate voltage is 1V to the leakage current when the gate voltage is 0V.
3. The method for predicting electrical characteristics of FDSOI devices based on CNN network according to claim 1, characterized in that: In step 4, the training set data is input into the CNN network prediction model. Specifically, the gate length, buried oxide layer thickness, body region doping concentration, and gate metal work function of each set of data are used as input data, and the threshold voltage, transconductance, subthreshold swing, and current switching ratio of each set of data are used as output labels.
4. The method for predicting electrical characteristics of FDSOI devices based on CNN network according to claim 1, characterized in that: The regression loss function in step 4 is specifically a mean square error (MSE) function, which calculates the sum of squares of the difference between the predicted value and the target value. The calculation formula is shown in formula (3): In formula (3), n is the number of samples, y i represents the true value of the i-th sample, is the predicted value of the i-th sample.