A High-Efficiency Semiconductor Device Simulation Method and System Based on Physical Parameter Dimensionality Reduction

Abstract

This invention belongs to the field of semiconductor device simulation technology, and discloses a high-efficiency semiconductor device simulation method and system based on physical parameter dimensionality reduction. The method includes: acquiring TCAD simulation data and simulating the current-voltage characteristics of a MOSFET; data normalization processing to normalize process parameters and optimize input data; extracting key physical parameters from the simulation data; constructing a process parameter-key physical parameter ANN model by training an artificial neural network (ANN) using normalized parameters and bias voltage to output key physical parameters; and constructing a key physical parameter-current ANN model to predict the drain-source current of the MOSFET. This invention provides a nonlinear dimensionality reduction method based on physical parameters, significantly reducing training time and accelerating the efficiency of process parameter optimization.

Description

Technical Field

[0001] This invention belongs to the field of semiconductor device simulation technology, and in particular relates to a high-efficiency semiconductor device simulation method and system based on physical parameter dimensionality reduction. Background Technology

[0002] In the development of new semiconductor processes and the maintenance and upgrading of older processes, semiconductor process simulation and device simulation tools (Technology Computer Aided Design, TCAD) play a crucial role in optimizing process parameters. Unlike traditional semiconductor device simulation based on equivalent circuit models, TCAD primarily relies on the numerical solution of various semiconductor device physical equations, ensuring the physical meaning of the simulation model and its calculation results. However, in actual simulation calculations, the simultaneous numerical solution of complex semiconductor transcendental equations is slow, making iterative parameter optimization based on TCAD extremely time-consuming. Therefore, exploring high-efficiency semiconductor device simulation methods with high physical information fidelity has become a research hotspot in the field of semiconductor device simulation technology both domestically and internationally.

[0003] In recent years, emerging machine learning-based semiconductor device simulation technologies have demonstrated advantages in both simulation efficiency and accuracy, making them an ideal solution to replace TCAD-based device simulation methods. Various machine learning algorithms have been proposed to attempt to replace TCAD simulation. In 2019, H. Carrillo... Researchers have applied machine learning techniques to the simulation of junctionless nanowire field-effect transistors (FETs). Through TCAD simulation, they generated multiple sets of current curves under different discrete doping conditions and established a mapping relationship between key electrical parameters and threshold voltage using a multilayer neural network. This method achieved accurate prediction of threshold voltage characteristic changes under different doping conditions. In 2021, C. Akbar et al. proposed a machine learning (ML)-assisted simulation method for three-dimensional multichannel gate all-silicon nanosheet MOSFET work function fluctuations. Based on TCAD simulation data under 330 different process parameter combinations, a model established using a random forest regressor algorithm can be used to accurately predict Ik. D –V G The curves were accurately predicted, exhibiting the same accuracy as device simulations. In the same year, K. Mehta et al. demonstrated the effectiveness of machine learning algorithms in predicting FinFET current-voltage and capacitance-voltage curves. Using limited training data (25-50 sets) obtained from TCAD simulations, they predicted current-voltage and capacitance-voltage curves using an autoencoder algorithm, achieving high accuracy. In 2022, R. Butola et al. proposed a machine learning-based method for modeling internal parameter fluctuations in all-silicon nanosheet MOSFETs. Results showed that the proposed model predicted outputs with 99% accuracy in R... 2 Fractions and less than 1% error.

[0004] The above methods all generate a mapping dataset of device parameters and electrical output characteristics by TCAD simulation, which is used for Artificial Neural Network (ANN) training, and then a TCAD proxy model is established. This method improves the simulation efficiency of semiconductor devices by combining the predictability of machine learning algorithms and device physical simulation. However, when applied to the simulation calculation scene of semiconductor devices with multiple input process parameters, this kind of method still faces problems such as too many input variables of a single ANN model, long training time of ANN model and convergence difficulty caused by large amount of training data. Therefore, how to break through the ANN-based device simulation method for joint calculation of multiple process parameters is a key problem to improve the simulation efficiency of semiconductor devices based on machine learning.

[0005] Through the above analysis, the problems and defects of the prior art are:

[0006] When the current machine learning algorithm is applied to the simulation of semiconductor devices with multiple input process parameters, the number of input variables of a single ANN model is too large, and the amount of training data is large, which leads to a long training time of the artificial neural network model. SUMMARY

[0007] In view of the problems existing in the prior art, the present application provides a high-efficiency semiconductor device simulation method and system based on nonlinear dimension reduction.

[0008] The present application is implemented in the following way: a high-efficiency semiconductor device simulation method based on nonlinear dimension reduction, comprising:

[0009] TCAD simulation data acquisition;

[0010] Data normalization processing;

[0011] Key physical parameter extraction;

[0012] Process parameter-key physical parameter ANN model construction;

[0013] Key physical parameter-current ANN model construction.

[0014] Further, step 1 is specifically:

[0015] In the TCAD software, the current-voltage characteristics of a specific structure of a planar metal-oxide-semiconductor field-effect transistor (MOSFET) are simulated under different combinations of process parameters to obtain the drain-source current and drain-source voltage relationship curve Ids-Vds and the drain-source current and gate-source voltage relationship curve Ids-Vgs, wherein the process parameters involved include P-type well doping concentration Pdep, gate oxide layer thickness tox, surface defect concentration Cit, gate length Lg, electron mobility u0, and lightly doped drain doping concentration Ldd.

[0016] Further, the step 2 is specifically:

[0017] The method for normalizing the input parameters, i.e., the process parameters, is specifically as follows:

[0018] x i =(X i -min{X i}) / (max{X i}-min{X i}) (1)

[0019] wherein Xi is the data set corresponding to the ith input parameter, xi is the data set corresponding to the normalized ith input parameter, and min{Xi} and max{Xi} represent the minimum value and the maximum value in the data set corresponding to the ith input parameter, respectively.

[0020] Further, the step 3 is specifically:

[0021] The on-state current (Ion), the off-state current (Ioff), and the threshold voltage (Vth) are selected as the key parameters, and the specific extraction method is as follows:

[0022] ① Vth extraction

[0023] The derivative of Ids with respect to Vgs is calculated based on the Ids-Vgs data obtained in step one, to obtain the transconductance Gm curve of the device, and then the Vgs corresponding to the maximum point is calculated and recorded as Vg_gmmax; then a tangent line is drawn at the point Vg=Vg_gmmax on the Ids-Vgs curve, and the intersection point with the Vgs axis is obtained, and the Vgs at the intersection point is subtracted by 0.05 to obtain Vth.

[0024] ② Ion extraction

[0025] The drain-source current Ids at Vgs=0V and Vds=1.2V is selected as Ion based on the Ids-Vds data obtained in step one.

[0026] ③ Ioff extraction

[0027] In step one, the Ids-Vds data obtained by TCAD simulation: select the drain current Ids when Vgs=1.2V, Vds=1.2V as Ioff;

[0028] Wherein, Vgs is the input gate voltage, Ids is the drain current, and Vdd is the maximum drain voltage.

[0029] Further, the step 4 is specifically:

[0030] The normalized device process parameters and bias voltages Vgs, Vds obtained in step 2 are taken as the input of the ANN model, and the key physical parameters Vth / Ion / Ioff obtained in step 3 are taken as the output of the ANN, to form a process parameter-key physical parameter sub-ANN model, that is, formula (2):

[0031] Para i =net i (x i ) (i=1,2,3) (2)

[0032] In the formula, Parai is the ith key physical parameter, wherein net1 corresponds to the threshold voltage Vth, net2 corresponds to the on-state current Ion, and net3 corresponds to the off-state current Ioff, and xi is the data set corresponding to the normalized ith input parameter obtained in step one.

[0033] Further, the step 5 is specifically:

[0034] According to the process parameter-key physical parameter ANN model obtained in step 4, the key physical parameters output by the model are taken as the input of the ANN model constructed in this step, and the drain current Ids of the MOSFET is taken as the output of the ANN, to form a key physical parameter-current ANN model, that is, formula (3):

[0035] f ANN (Para,V gs ,V ds )=I ds (3)

[0036] Wherein, Para is the key physical parameter, Vds is the drain voltage, and f is the ANN mapping relationship function.

[0037] Another object of the application is to provide a high-efficiency semiconductor device simulation system based on nonlinear dimension reduction, comprising:

[0038] The TCAD simulation module is configured to simulate the current-voltage characteristics of a planar metal oxide semiconductor field effect transistor (MOSFET) of a specific structure, and generate a drain-source current-voltage relationship curve;

[0039] Data processing module: with normalization processing function, standardizing each process parameter;

[0040] Key parameter extraction module: used for extracting on-state current (Ion), off-state current (Ioff) and threshold voltage (Vth) from simulation data;

[0041] ANN modeling module: constructing process parameter-key physical parameter sub-ANN model according to normalized process parameters and bias voltage;

[0042] Current prediction module: constructing key physical parameter-current ANN model based on the output of process parameter-key physical parameter ANN model, and predicting the drain-source current of MOSFET.

[0043] Another object of the present application is to provide a computer device comprising a memory and a processor, the memory storing a computer program, the computer program being executed by the processor to make the processor execute the steps of the high-efficiency semiconductor device simulation method based on nonlinear dimension reduction.

[0044] Another object of the present application is to provide a computer-readable storage medium storing a computer program, the computer program being executed by a processor to make the processor execute the steps of the high-efficiency semiconductor device simulation method based on nonlinear dimension reduction.

[0045] Another object of the present application is to provide an information data processing terminal for implementing the high-efficiency semiconductor device simulation method based on nonlinear dimension reduction.

[0046] Another object of the present application is to provide a system for semiconductor device simulation, comprising:

[0047] TCAD simulation module, used for obtaining current-voltage characteristic data of a planar metal-oxide-semiconductor field-effect transistor (MOSFET) of a specific structure;

[0048] Data processing module, used for normalizing the TCAD simulation data and extracting key physical parameters such as on-state current (Ion), off-state current (Ioff) and threshold voltage (Vth);

[0049] Two ANN (artificial neural network) models, the first one is used to establish the relationship between process parameters and key physical parameters, and the second one is used to map the key physical parameters to the drain-source current of MOSFET;

[0050] Control unit, used for coordinating the work of each module, ensuring the accurate flow and processing of data.

[0051] The TCAD simulation module simulates the current-voltage characteristics of the MOSFET under different process parameter combinations, including the relationship between the drain-source current and the drain-source voltage and the relationship between the drain-source current and the gate-source voltage.

[0052] The data processing module normalizes the data obtained by the TCAD simulation, and then extracts the on-state current, the off-state current and the threshold voltage as the key physical parameters.

[0053] The ANN model construction module includes two sub-modules: the first sub-module takes the normalized process parameters and the bias voltage as input and the key physical parameters as output; and the second sub-module takes the key physical parameters as input and the drain-source current of the MOSFET as output, realizing efficient simulation from the process parameters to the current characteristics.

[0054] In combination with the above technical solutions and the technical problems solved, the technical solutions to be protected by the present application have the following advantages and positive effects:

[0055] First, the prior art uses a TCAD simulation to generate a mapping data set of device parameters and electrical output characteristics, which is used for artificial neural network (ANN) training, and then a TCAD proxy model is established. However, when this method is applied to the simulation calculation scene of semiconductor devices with multiple input process parameters, it still faces problems such as too many input variables of a single ANN model, large amount of training data, long ANN model training time, and convergence difficulty. The present patent proposes a nonlinear dimensionality reduction method based on physical parameters, which maps process parameters to key physical parameters by introducing a process parameter-key physical parameter ANN model, and then establishes a key physical parameter-current ANN model. Finally, a specific structure of a planar metal-oxide-semiconductor field effect transistor is used to perform current-voltage simulation calculation under different process parameter combinations in the TCAD software. The verification results show that, compared with the traditional method, the technical solution proposed by the present patent significantly reduces the model training time while maintaining high precision.

[0056] Second, considering the technical solution as a whole or from the perspective of the product, the technical solution to be protected by the present application has the following technical effects and advantages:

[0057] Improve simulation accuracy: by combining TCAD simulation and advanced ANN technology, the present application significantly improves the accuracy of semiconductor device simulation.

[0058] Improve simulation speed: by using a nonlinear dimensionality reduction method based on physical parameters, the model training time is greatly reduced, and the process parameter optimization efficiency is accelerated.

[0059] Efficiency of data processing is improved: the application of normalization processing and ANN model greatly improves the processing efficiency of a large amount of simulation data, making the simulation more suitable for rapid design iteration.

[0060] Simplify the process of extracting key parameters: the innovative parameter extraction method simplifies the process of extracting key physical parameters from complex data, improving the ease of operation.

[0061] Wide application potential: the method of the present application is not only suitable for the simulation of MOSFET, but also can be extended to the simulation analysis of other semiconductor devices.

[0062] Third, the technical solution of the present application solves the technical problem that people have been eager to solve but have failed to succeed: the prior art uses TCAD simulation to generate a mapping data set of device parameters and electrical output characteristics, which is used for artificial neural network training, and then establishes a TCAD proxy model. However, when applied to the simulation calculation scene of semiconductor devices with multiple input process parameters, this kind of method still faces problems such as too many input variables of a single ANN model, large amount of training data, long ANN model training time, and convergence difficulty. The present patent provides a nonlinear dimensionality reduction method based on physical parameters, which greatly reduces the training time and accelerates the process parameter optimization efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0063] Figure 1 is a flowchart of the high-efficiency semiconductor device simulation method based on nonlinear dimensionality reduction provided by the embodiments of the present application.

[0064] Figure 2 is a Vth calculation schematic diagram provided by the embodiments of the present application.

[0065] Figure 3 is a sub-ANN model schematic diagram of input process parameters-key physical parameters provided by the embodiments of the present application.

[0066] Figure 4 is a key physical parameter-current ANN model schematic diagram provided by the embodiments of the present application.

[0067] Figure 5 is a comparison between the model prediction results and the TCAD simulation current output characteristics and transfer characteristics (asterisk is TCAD simulation data, and circle is ANN model prediction result). DETAILED DESCRIPTION

[0068] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below with reference to the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0069] Below are four specific embodiments of "High-efficiency semiconductor device simulation method based on nonlinear dimensionality reduction" and their implementation schemes:

[0070] Embodiment 1: Semiconductor simulation for optimizing transistor design

[0071] 1) TCAD simulation:

[0072] Simulate the current-voltage characteristics of MOSFETs with different process parameter combinations in TCAD software.

[0073] Obtain data for drain-source current and voltage (Ids-Vds) and drain-source current and gate-source voltage (Ids-Vgs) relationships.

[0074] 2) Data processing and key physical parameter extraction:

[0075] Normalize the simulation data.

[0076] Extract key physical parameters, including on-state current (Ion), off-state current (Ioff), and threshold voltage (Vth).

[0077] 3) ANN model construction and training:

[0078] Construct two ANN models: one to correlate process parameters with key physical parameters, and another to map key physical parameters to drain-source current.

[0079] Train the ANN models with a large amount of simulation data.

[0080] 4) Application of simulation results:

[0081] Use the trained ANN models to quickly simulate new process parameter combinations and predict their current-voltage characteristics.

[0082] Apply these results to optimize MOSFET design for improved performance or reduced power consumption.

[0083] Embodiment 2: Device simulation for new semiconductor materials

[0084] 1) TCAD simulation and data acquisition:

[0085] Simulate MOSFETs based on new semiconductor materials (such as GaN, SiC, etc.) in TCAD software.

[0086] Obtain current-voltage characteristic data under different process parameters.

[0087] 2) Data processing and parameter extraction:

[0088] Normalize the simulation data.

[0089] Extract key physical parameters such as Ion, Ioff, and Vth from the data.

[0090] 3) Construction and application of ANN model:

[0091] Construct ANN model based on the extracted key physical parameters.

[0092] Use the model to quickly predict the performance of new material devices.

[0093] 4) Analysis of simulation results:

[0094] Analyze the simulation results to evaluate the performance of new materials under different process parameters.

[0095] Adjust the device design based on the simulation results to achieve higher performance.

[0096] Example 3: Design of a new metal-oxide-semiconductor field-effect transistor (MOSFET) device

[0097] Semiconductor process simulation and device simulation tool (TCAD) data acquisition: Use TCAD software to simulate the current-voltage characteristics of the new MOSFET structure under different process parameters, and obtain the drain-source current and voltage relationship data.

[0098] Data normalization: Normalize the process parameter data obtained from simulation to eliminate the order of magnitude difference between different parameters.

[0099] Key physical parameter extraction: Extract key parameters such as on-state current (Ion), off-state current (Ioff), and threshold voltage (Vth) from the normalized simulation data.

[0100] Artificial neural network (ANN) model construction: Construct an ANN model of process parameters-key physical parameters to predict key physical parameters under different process conditions.

[0101] Current prediction model: Use the key physical parameter-current ANN model to further predict the drain-source current under different process conditions.

[0102] Design optimization: Use the model prediction results to guide the design of the new MOSFET, optimize its performance and reliability.

[0103] Example 4: Rapid performance evaluation of semiconductor devices

[0104] TCAD simulation: Perform TCAD simulation on a variety of semiconductor devices with different structures to obtain current-voltage characteristic data under various process parameters.

[0105] Data Standardization: Normalize all simulation data to ensure consistency of input data and effectiveness of comparison.

[0106] Automatic Parameter Extraction: Automatically extract key physical parameters such as Ion, Ioff, and Vth from complex simulation data to quickly obtain core information.

[0107] ANN Model Application: Use pre-trained ANN models to predict the key physical characteristics of devices based on normalized process parameters.

[0108] Performance Prediction and Analysis: Based on the predicted key physical parameters, evaluate the performance of different design schemes to accelerate the device design selection process.

[0109] Result Application: Use the predicted results of the model to perform performance comparison analysis and quickly select the optimal device design scheme.

[0110] The present invention mainly improves the following problems and defects of the prior art and realizes significant technical progress:

[0111] Insufficient simulation accuracy: Traditional semiconductor device simulation methods often cannot achieve high accuracy when dealing with complex current-voltage relationships, especially when considering the influence of multiple process parameters.

[0112] Low data processing efficiency: Existing technologies have low efficiency in processing and analyzing large amounts of simulation data, limiting their application in rapid iterative design.

[0113] Difficult extraction of key physical parameters: In traditional methods, it is often difficult and time-consuming to extract key physical parameters (such as Ion, Ioff, Vth) from complex data.

[0114] In view of the problems existing in the prior art, the technical scheme adopted by the present invention is:

[0115] TCAD simulation and data processing: Perform detailed current-voltage characteristic simulation through TCAD software and consider multiple process parameters to obtain high-quality simulation data.

[0116] Data normalization: Normalize the process parameter data obtained by simulation to solve the problem of large order of magnitude difference in input parameters and optimize subsequent neural network training.

[0117] Efficient key physical parameter extraction method: Use an innovative method to accurately and quickly extract key parameters such as Ion, Ioff, and Vth from simulation data.

[0118] Construction and application of ANN model: two ANN models are constructed and trained, one for converting normalized process parameters into key physical parameters, and the other for predicting drain-source current based on key physical parameters, greatly improving the efficiency and accuracy of simulation.

[0119] As Figure 1 indicated, the application provides a high-efficiency semiconductor device simulation method based on nonlinear dimension reduction, comprising the following steps:

[0120] S101, TCAD simulation data acquisition;

[0121] S102, data normalization processing;

[0122] S103, key physical parameter extraction;

[0123] S104, process parameter-key physical parameter ANN model construction;

[0124] S105, key physical parameter-current ANN model construction.

[0125] Step 1: TCAD simulation data acquisition

[0126] In the TCAD software, for a specific structure of a planar metal oxide semiconductor field effect transistor (MOSFET), the current-voltage characteristic simulation calculation is carried out under different process parameter combinations, and the drain-source current and drain-source voltage relationship curve Ids-Vds and the drain-source current and gate-source voltage relationship curve Ids-Vgs are obtained, wherein the process parameters involved include P-type well doping concentration Pdep, gate oxide layer thickness tox, surface defect concentration Cit, gate length Lg, electron mobility u0, and lightly doped drain doping concentration Ldd. In the simulation calculation, the scanning range of gate-source bias voltage Vgs is 0-1.2V with a step of 0.1V, and the scanning range of drain-source bias voltage Vds is 0-1.2V with a step of 0.05V.

[0127] Step 2: data normalization processing

[0128] Different process parameters (such as planar MOSFET doping concentration and gate oxide layer thickness) differ greatly in order of magnitude, and if they are directly used as input for artificial neural network model training, the neural network model obtained by training will ignore the influence of parameters with small order of magnitude. To this end, the application adopts a method of normalizing input parameters, i.e. process parameters, and the specific form is as follows:

[0129] x i =(X i -min{X i}) / (max{X i}-min{X i}) (1)

[0130] where Xi is the data set corresponding to the ith input parameter, xi is the normalized data set corresponding to the ith input parameter, and min{Xi} and max{Xi} represent the minimum and maximum values in the data set corresponding to the ith input parameter, respectively.

[0131] In addition, the plane MOSFET drain-source current obtained by TCAD simulation is processed using function conversion method.

[0132] Step 3: Key physical parameter extraction

[0133] On-state current (Ion), off-state current (Ioff), and threshold voltage (Vth) are important parameters for describing the electrical characteristics of a plane MOSFET. Therefore, these parameters are selected as the key parameters of this patent, and the specific extraction method is as follows: (Vgs is the input gate voltage, Ids is the drain-source current, and Vdd is the maximum drain voltage)

[0134] ④Vth extraction

[0135] In step one, the Ids-Vgs data obtained by TCAD simulation is used to calculate the derivative of Ids with respect to Vgs, and the transconductance Gm curve of the device is obtained. Then, the Vgs corresponding to the maximum point is calculated and denoted as Vg_gmmax. Then, a tangent line is drawn at the point Vg = Vg_gmmax on the Ids-Vgs curve, and the intersection point with the Vgs axis is obtained. The Vgs at this intersection point is then subtracted by 0.05 to obtain Vth. The schematic diagram of this step is shown in Figure 2 .

[0136] ⑤Ion extraction

[0137] In step one, the Ids-Vds data obtained by TCAD simulation is used to select the drain-source current Ids when Vgs = 0V and Vds = 1.2V as Ion.

[0138] ⑥Ioff extraction

[0139] In step one, the Ids-Vds data obtained by TCAD simulation is used to select the drain-source current Ids when Vgs = 1.2V and Vds = 1.2V as Ioff.

[0140] Step 4: Process parameter-key physical parameter ANN model establishment

[0141] The normalized device process parameters and bias voltages Vgs, Vds obtained in step 2 are used as the input of the ANN model, and the key physical parameters Vth / Ion / Ioff obtained in step 3 are used as the output of the ANN, forming a process parameter-key physical parameter sub-ANN model, i.e., formula (2):

[0142] Parai = net i (x i ) (i = 1, 2, 3) (2)

[0143] where Parai is the ith key physical parameter, net1 corresponds to threshold voltage Vth, net2 corresponds to on-state current Ion, net3 corresponds to off-state current Ioff, xi is the normalized data set of the ith input parameter obtained in step 1.

[0144] As shown in FIG. 3, the input of the sub-ANN model includes bias (Vgs, Vds) and the normalized input process parameter data set obtained in step 2, and the output is the key physical parameter Vth / Ion / Ioff. The activation function of the neuron is the hyperbolic tangent function tanh, and the ANN is trained by using the back propagation algorithm available in the standard ANN software package PyTorch. The loss function is the root mean square error of Vth / Ion / Ioff. Figure 3 Step 5: Key physical parameter-current ANN model construction

[0145] According to the process parameter-key physical parameter ANN model obtained in step 4, the output (key physical parameter) thereof is taken as the input of the ANN model constructed in this step, and the drain-source current Ids of the MOSFET is taken as the output of the ANN, thereby forming a key physical parameter-current ANN model, i.e., formula (3):

[0146] f ANN (Para, V gs , V ds ) = I ds (3)

[0147] where Para is the key physical parameter, Vds is the drain-source voltage, and f is the ANN mapping relationship function. The ANN network topology corresponding to the above formula is shown in FIG. 4.

[0148] Figure 4

[0149] The activation function of the neuron in the ANN is the hyperbolic tangent function tanh, and the ANN is trained by using the back propagation algorithm available in the standard ANN software package PyTorch. The loss function is the root mean square error of Ids. Figure 4

[0150] ​​In order to verify the effectiveness and value of the technical scheme of the present application, the current-voltage characteristics of a specific structure of a planar metal oxide semiconductor field effect transistor (MOSFET) are simulated and calculated in a TCAD software under different process parameter combinations, and the drain-source current and drain-source voltage relationship curve Ids-Vds and the drain-source current and gate-source voltage relationship curve Ids-Vgs are obtained, wherein the process parameters involved include P-type well doping concentration Pdep, gate oxide layer thickness tox, surface defect concentration Cit, gate length Lg, electron mobility u0, and lightly doped drain doping concentration Ldd. In the simulation calculation, the scanning range of the gate-source bias voltage Vgs is 0-1.2V with a step of 0.1V, and the scanning range of the drain-source bias voltage Vds is 0-1.2V with a step of 0.05V. The Ids-Vds and Ids-Vgs data obtained by simulation are selected by the hyper-Latin sampling method, 80% of the data are selected as the training set, and 20% of the data are selected as the test set. Then, the ANN model is trained and constructed by using the technical scheme proposed in the present application according to the training set data, and the test set is used for verification.

[0151] Evidence of the effects of the embodiments. The embodiments of the present application have achieved some positive effects in the process of research and development or use, and indeed have great advantages compared with the prior art. The following content is described in combination with the data and graphs of the test process.

[0152] The model prediction results obtained by using the method of the present application are compared with the current output characteristics (Ids-Vds) and transfer characteristics (Ids-Vgs) curves obtained by TCAD simulation, and the results are shown in Figure 5

[0153] The mean square error function is used where N s is the data amount, Y data is the simulation value, and Y pred is the ANN model prediction value. The finally calculated Ids-Vgs curve error is less than 1%, and the Ids-Vds curve error is less than 1%. It can be seen that the method provided by the present application can accurately represent the TCAD simulation results.

[0154] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any modification, equivalent replacement and improvement made by any person skilled in the art within the technical range disclosed in the present application, as long as it is within the spirit and principles of the present application, should be covered within the protection scope of the present application.​

Claims

1. A high-efficiency semiconductor device simulation method based on nonlinear dimension reduction, characterized by, Comprise: Step 1, TCAD simulation data acquisition; Step 2, data normalization processing; Step 3, key physical parameter extraction; Step 4, process parameter-key physical parameter ANN model construction; Step 5, key physical parameter-current ANN model construction; Step 1 is specifically: In the TCAD software, the plane metal oxide semiconductor field effect transistor of a specific structure is simulated and calculated under different process parameter combinations, and the drain-source current and drain-source voltage relationship curve Ids-Vds and the drain-source current and gate-source voltage relationship curve Ids-Vgs are obtained, wherein the process parameters involved include P-type well doping concentration Pdep, gate oxide layer thickness tox, surface defect concentration Cit, gate length Lg, electron mobility u0, and light doping drain doping concentration Ldd; Said step 3 is specifically: Selecting the on-state current, off-state current, and threshold voltage as the key parameters, the specific extraction method is as follows: ①Vth extraction In the Ids-Vgs data obtained in step one TCAD simulation, the partial derivative of Ids to Vgs is calculated to obtain the transconductance Gm curve of the device, and then the Vgs corresponding to the maximum point is calculated, which is denoted as Vg_gmmax; Then draw a tangent line at the Vg=Vg_gmmax point on the Ids-Vgs curve, and the intersection point with the Vgs axis is obtained, and then the Vgs at the intersection point is subtracted by 0.05 to obtain Vth; ②Ion extraction In the Ids-Vds data obtained in step one TCAD simulation, the drain-source current Ids when Vgs=0V and Vds=1.2V is selected as Ion; ③Ioff extraction In the Ids-Vds data obtained in step one TCAD simulation: select the drain-source current Ids when Vgs=1.2V and Vds=1.2V as Ioff; Wherein, Vgs is the input gate voltage, Ids is the drain current, and Vdd is the maximum drain voltage.

2. The high-efficiency semiconductor device simulation method based on nonlinear dimension reduction of claim 1, wherein, Said step 2 is specifically: The method for normalizing the input parameters, i.e., each process parameter, is as follows: x i = (X i -min{X i}) / (max{X i}-min{X i}) (1) Wherein Xi is the data set corresponding to the i-th input parameter, xi is the data set corresponding to the normalized i-th input parameter, and min{Xi} and max{Xi} represent the minimum and maximum values in the data set corresponding to the i-th input parameter, respectively.

3. The high-efficiency semiconductor device simulation method based on nonlinear dimension reduction of claim 1, wherein, Said step 4 is specifically: The normalized device process parameters and bias voltages Vgs, Vds obtained in step 2 are used as the input of the ANN model, and the key physical parameters Vth / Ion / Ioff obtained in step 3 are used as the output of the ANN, forming a process parameter-key physical parameter sub-ANN model, i.e., formula (2): Para i = net i (x i ) (i = 1,2,3) (2) Wherein Parai is the i-th key physical parameter, wherein net1 corresponds to the threshold voltage Vth, net2 corresponds to the on-state current Ion, and net3 corresponds to the off-state current Ioff, and xi is the data set corresponding to the normalized i-th input parameter obtained in step one.

4. The high-efficiency semiconductor device simulation method based on nonlinear dimension reduction of claim 1, wherein, Said step 5 is specifically: According to the process parameter-key physical parameter ANN model obtained in step 4, the output key physical parameters are taken as the input of the ANN model constructed in this step, and the drain-source current Ids of the MOSFET is taken as the output of the ANN, to form a key physical parameter-current ANN model, that is, formula (3): f ANN (Para, V gs ,V ds ) = I ds (3) Wherein Para is the key physical parameter, Vds is the drain-source voltage, and f is the ANN mapping relationship function.

5. A high efficiency semiconductor device simulation system based on nonlinear dimensionality reduction of the method of claim 1, wherein, It includes: A TCAD simulation module configured to simulate the current-voltage characteristics of a planar metal-oxide-semiconductor field-effect transistor of a specific structure to generate a drain-source current-voltage relationship curve; A data processing module with normalization processing function for standardizing each process parameter; A key parameter extraction module for extracting on-state current Ion, off-state current Ioff and threshold voltage from simulation data; An ANN modeling module for constructing a process parameter-key physical parameter sub-ANN model according to the normalized process parameters and bias voltage; A current prediction module for constructing a key physical parameter-current ANN model based on the output of the process parameter-key physical parameter ANN model to predict the drain-source current of the MOSFET.

6. A nonlinear dimensionality reduction-based high-efficiency semiconductor device simulation system implementing the nonlinear dimensionality reduction-based high-efficiency semiconductor device simulation method according to any one of claims 1 to 4, characterized by It includes: A TCAD simulation module for obtaining current-voltage characteristic data of a planar metal-oxide-semiconductor field-effect transistor of a specific structure; A data processing module for normalizing TCAD simulation data and extracting key physical parameters; Two ANN models, the first one is used to establish the relationship between process parameters and key physical parameters, and the second one is used to map key physical parameters to the drain-source current of the MOSFET; A control unit for coordinating the work of each module to ensure accurate data flow and processing.

7. The high-efficiency semiconductor device emulation system based on nonlinear dimensionality reduction of claim 6, wherein, The TCAD simulation module simulates the current-voltage characteristics of the MOSFET under different process parameter combinations, including the relationship between drain-source current and drain-source voltage, and drain-source current and gate-source voltage; The data processing module normalizes the data obtained by TCAD simulation, and then extracts the on-state current, off-state current and threshold voltage as the key physical parameters.

8. The high efficiency semiconductor device simulation system based on nonlinear dimension reduction of claim 6, wherein, The ANN model construction module includes two sub-modules: the first sub-module takes the normalized process parameters and bias voltage as input and the key physical parameters as output; the second sub-module takes the key physical parameters as input and the drain-source current of the MOSFET as output, realizing efficient simulation from process parameters to current characteristics.

Images