A method for modeling a MOSFET device based on a multi-physical constraint neural network and related equipment

Abstract

The application discloses a MOSFET device modeling method based on a multi-physical constraint neural network and related equipment, and the method comprises the following steps: acquiring various data of a MOSFET device in a normal working state, pre-processing the obtained data to obtain a data set required for training a multi-physical constraint neural network model; constructing the multi-physical constraint neural network model, taking charge balance and carrier transport equations as physical constraints, and adding the model to a design and training process to realize physical constraint neural network modeling; and iteratively training the multi-physical constraint neural network model by using the data set until evaluation indexes of the model meet preset requirements. The application introduces the charge balance condition in the vertical channel direction and the carrier transport equation as physical constraints into the MOSFET neural network model design and training process, and can accurately fit the behavior characteristics of the MOSFET device in the working state. The application can be widely applied to the field of integrated circuit technology.

Description

Technical Field

[0001] This invention relates to the field of integrated circuit technology, and in particular to a MOSFET device modeling method and related equipment based on a multi-physics constraint neural network. Background Technology

[0002] Building MOSFET device models can not only help users analyze the behavior of MOSFET devices under different operating conditions at extremely low cost and find the optimal design parameters, but also help users study the impact of different materials and process conditions on device performance, conduct digital experiments and tests through electronic simulation, and thus improve device design and manufacturing technology.

[0003] Traditional device modeling primarily employs two approaches: physical process modeling based on the device's semiconductor physical characteristics and data-driven modeling based on device test data. Data-driven models can efficiently and concisely describe the device's circuit behavior, but they completely ignore the device's inherent physical characteristics and are mainly used for post-fabrication operation mode and performance degradation analysis. Physical characteristic models can accurately describe various higher-order characteristics of devices and are excellent tools for device design, manufacturing, and academic research. However, physical models are generally very complex, requiring significant computational resources to solve, and often suffer from simulation non-convergence and time-consuming computation in applications such as EDA. Summary of the Invention

[0004] To at least partially address one of the technical problems existing in the prior art, the present invention aims to provide...

[0005] The first technical solution adopted in this invention is:

[0006] A MOSFET device modeling method based on a multi-physics constraint neural network includes the following steps:

[0007] Acquire various data under normal operating conditions of MOSFET devices, and preprocess the acquired data to obtain the dataset required for training a multi-physics constrained neural network model;

[0008] A multi-physics-constrained neural network model is constructed, incorporating charge balance and transport equations as physical constraints into the model design and training process to achieve physical constraint neural network modeling.

[0009] The multi-physics-constrained neural network model is trained iteratively using a dataset until the model's performance metrics meet the preset requirements.

[0010] Furthermore, the multi-physics constraint neural network model includes three multi-input single-output neural networks, which are used to solve the surface potential at the MOSFET channel, the semiconductor charge, and the drain-source current of the MOSFET, respectively.

[0011] The first neural network is used to solve for the surface potential at the MOSFET channel, and its describing equation is:

[0012] NN(V gs V ds ,I ds ,T emp ,T ox N d )=φ s_pred

[0013] The input parameters of the first neural network include: gate-source voltage V gs Drain-source voltage V ds Drain-source current I ds Temperature T emp MOSFET gate oxide thickness T ox The semiconductor doping concentration N of MOSFET d The output parameter is the calculated value of the channel surface potential φ. s_pred ;

[0014] The second neural network is used to solve for semiconductor charge, and its describing equation is:

[0015] NN(φ s_pred ,T emp N d )=Q semi_pred

[0016] The input parameters of the second neural network include: the calculated value of the channel surface potential φ. s_pred Temperature T emp The semiconductor doping concentration N of MOSFET d The output parameter is the calculated semiconductor charge value Q. semi_pred ;

[0017] The third neural network is used to solve for the drain-source current of the MOSFET, and its describing equation is:

[0018] NN(φ s_pred V ds ,T emp N d ) = I ds_pred

[0019] The input parameters of the third neural network include: the calculated surface potential value φ of the channel. s_pred Drain-source voltage V ds Temperature T emp The semiconductor doping concentration N of MOSFET d The output parameter is the calculated drain-source current I of the MOSFET. ds_pred .

[0020] Furthermore, the charge balance refers to the charge neutrality condition in the vertical direction at the oxide-semiconductor interface when the MOSFET is operating normally, that is:

[0021] Q gate +Q semi =0

[0022] In the formula, Q gate Q is the gate charge. semi Semiconductor charge;

[0023]

[0024] Q semi =f(φ s_pred ,T emp N d )=NN(φ s_pred ,T emp N d )

[0025] In the formula, ε ox V is the dielectric constant of the oxide layer. fb For flat-band voltage, φ s_pred The surface potential calculated for a neural network; f(φ) s_pred ,T emp N d φ is a precise calculation function for semiconductor charge. This function is difficult to solve, therefore a neural network model NN(φ) is used. s_pred ,T emp N d Fit the function.

[0026] Furthermore, the transport equation describes the relationship between the current and the movement of charge carriers within the semiconductor, including both drift and diffusion. The analytical calculation process needs to consider the distribution of charge carriers and the distribution of the electric field.

[0027] Among them, the neural network NN(φ) is used s_pred V ds ,T emp N d ) = I ds_pred Fitting the drain-source current I of the MOSFET ds .

[0028] Furthermore, the loss function during the training process of the multi-physics constrained neural network model is:

[0029] Loss = w data Loss data +w physics1 Loss physics1 +w physics2 Lossphysics2

[0030] Loss data =MSE(φ s_pred -φ s ,0)

[0031] Loss physics1 =MSE(Q gate +Q semi_pted ,0)

[0032] Loss physics2 =MSE(I ds_pred -I ds ,0)

[0033] In the formula, Loss represents the total error of the model. data Loss represents the computational error of the neural network. physics1 Loss is the physical loss due to MOSFET charge balance. physics2 For the physical loss in the MOSFET delivery equation, w data w physics1 and w physics2 These are the weights of the neural network calculation error and the weights of the two physical constraint errors, respectively; MSE() represents the mean square error function, which represents the average of the sum of squares of the numerical errors of the two types of data.

[0034] Furthermore, the acquisition of various data under normal operating conditions of the MOSFET device includes:

[0035] Obtain the MOSFET device at different drain-source voltages V ds Under conditions ranging from minimum to maximum, the gate-source voltage V gs The data obtained during the device's change from minimum to maximum value includes raw data and characteristic data of the raw data; among which, the raw data includes temperature T. emp MOSFET gate oxide thickness T ox The semiconductor doping concentration N of MOSFET d The raw data includes voltage, current, capacitance, transconductance, on-resistance, and surface potential. The characteristic data of the raw data include the mean, standard deviation, kurtosis, skewness, and slope of the raw data.

[0036] Furthermore, the preprocessing of the obtained data includes:

[0037] The data is preprocessed through data cleaning and standardization methods, erroneous data from the data collection stage is deleted, similar data are adjusted to the same unit of measurement, and then all data are adjusted to the appropriate units of measurement.

[0038] The second technical solution adopted in this invention is:

[0039] A MOSFET device modeling system based on a multi-physics constraint neural network includes:

[0040] The data acquisition module is used to acquire various data under normal operating conditions of MOSFET devices and preprocess the acquired data to obtain the dataset required for training a multi-physics constraint neural network model.

[0041] The model building module is used to build multi-physics-constrained neural network models. It incorporates charge balance and transport equations as physical constraints into the model design and training process to achieve physical constraint neural network modeling.

[0042] The model training module is used to iteratively train a multi-physics-constrained neural network model using a dataset until the model's performance metrics meet preset requirements.

[0043] The third technical solution adopted in this invention is:

[0044] An electronic device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set, or an instruction set, and the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement a MOSFET device modeling method based on a multi-physics constraint neural network as described above.

[0045] The fourth technical solution adopted in this invention is:

[0046] A computer-readable storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement a MOSFET device modeling method based on a multi-physics constraint neural network as described above.

[0047] The fifth technical solution adopted in this invention is:

[0048] A computer program product or computer program includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium and execute the computer instructions, causing the computer device to perform the aforementioned MOSFET device modeling method based on a multi-physics constraint neural network.

[0049] The beneficial effects of this invention are as follows: This invention introduces the charge balance condition and carrier transport equation in the vertical channel direction as physical constraints into the design and training process of MOSFET neural network models, enabling accurate fitting of the behavioral characteristics of MOSFET devices during operation. Furthermore, this invention is simple in principle, highly portable, and consumes few computational resources, making it widely applicable in academic research and engineering practice. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following description is provided with accompanying drawings of the relevant technical solutions in the embodiments of the present invention or the prior art. It should be understood that the accompanying drawings described below are only for the purpose of clearly illustrating some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0051] Figure 1 This is a diagram of the MOSFET model architecture based on a multi-physics constraint neural network in an embodiment of the present invention;

[0052] Figure 2 This is a flowchart of the MOSFET device modeling method based on a multi-physics constraint neural network in an embodiment of the present invention;

[0053] Figure 3 This is a schematic diagram illustrating how the error of a MOSFET device model based on a multi-physics constraint neural network is reduced to the expected level in an embodiment of the present invention;

[0054] Figure 4 This is a flowchart of the steps of the MOSFET device modeling method based on a multi-physics constraint neural network in an embodiment of the present invention. Detailed Implementation

[0055] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application. The step numbers in the following embodiments are set only for ease of explanation, and there is no limitation on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0056] The terminology used in the embodiments of this application is for the purpose of describing specific embodiments only and is not intended to limit the embodiments of this application. The singular forms "a," "described," and "the" used in the embodiments of this application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise. Furthermore, unless otherwise expressly limited, terms such as "set," "install," and "connect" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this invention in conjunction with the specific content of the technical solution.

[0057] In the description of this application, it should be understood that the orientation descriptions, such as up, down, front, back, left, right, etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

[0058] In the description of this application, "several" means one or more, "more than" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" in the description is merely for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0059] In the description of this application, "and / or" describes the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the related objects before and after it are in an "or" relationship.

[0060] Terminology Explanation:

[0061] MOSFET: an abbreviation for Metal-Oxide-Semiconductor Field-Effect Transistor.

[0062] Artificial Neural Networks (ANNs) are widely recognized as an effective alternative to traditional modeling techniques. Neural networks have the ability to approximate arbitrary nonlinear functions, making them an unconventional yet useful tool for device modeling and design. During the model design phase, a large amount of data is obtained through device testing and simulation. This data is then cleaned to generate a high-quality training dataset to support model training. By training the neural network, an accurate and efficient neural network model can be obtained.

[0063] In recent years, with the rapid development of technologies such as computational science, artificial intelligence, and computing power, device modeling techniques based on neural networks have received increasing attention. To address issues such as poor modeling stability and unstable convergence speed of neural network models, researchers have conducted research on optimizing neural networks using intelligent algorithms such as genetic algorithms (GA), Bayesian inference (BI), particle swarm optimization (PSO), and support vector machines (SVM). Compared to traditional modeling techniques, semiconductor device modeling techniques based on neural networks can efficiently and accurately fit the behavioral characteristics of devices. However, these models rely on data-driven modeling based on device operating data, almost completely disregarding the device's operating mechanisms and underlying physical principles, resulting in models that cannot describe the physical mechanisms and deep physical meaning of device operation. In recent years, physically constrained neural networks have begun to demonstrate considerable academic research value in the fields of fluid mechanics and thermodynamics-related physical computations. The idea of ​​constraining neural network training through physical process description equations can also be applied to device modeling research.

[0064] Based on this, the present invention provides a modeling method that incorporates the charge balance condition and carrier transport equation in the vertical channel direction as physical constraints into the design and training process of a MOSFET neural network model. The charge balance physical constraint is that the MOSFET operates at a given gate-source voltage V... gs Drain-source voltage V ds Under certain conditions, the fundamental electrostatic conditions that must be satisfied at the semiconductor-oxide interface are defined. The input quantities in these constraints relate to the input and output quantities of the surface potential solving neural network, including but not limited to input quantities, output quantities, or process calculation variables. The aforementioned transport equation is for the MOSFET at a given drain-source voltage V. ds Under certain conditions, the equations describing the distribution and motion of charge carriers within a semiconductor, and the input quantities in the constraints involve the input and output quantities of the neural network solving for the surface potential, including but not limited to input quantities, output quantities, or process calculation variables. Using two or more physical conditions to constrain the training process of the neural network model improves training effectiveness.

[0065] Example 1

[0066] like Figure 4As shown, this embodiment provides a MOSFET device modeling method based on a multi-physics-constraint neural network. Compared to traditional physical process modeling, this method avoids the predicament of solving complex physical processes. Compared to general data-driven modeling methods, this method uses multiple physical constraints in the modeling and training process, which is closer to the real physical process. Overall, it offers advantages such as strong physical interpretability, high modeling accuracy, and low computational resource consumption. The method specifically includes the following steps:

[0067] S1. Obtain various data under normal operating conditions of the MOSFET device, and preprocess the obtained data to obtain the dataset required for training the multi-physics constraint neural network model.

[0068] S2. Construct a multi-physics-constrained neural network model, incorporating charge balance and transport equations as physical constraints into the model design and training process to achieve physical constraint neural network modeling.

[0069] S3. Iteratively train the multi-physics-constrained neural network model using the dataset until the model's performance metrics meet preset requirements. In some embodiments, model training is successfully completed and a physical constraint neural network model is generated only when the model training metrics reach the expected level and remain stable in multiple repeated testing processes. The training metrics include, but are not limited to, model error, learning rate, MAE, MSE, and other data metrics.

[0070] The above method will be explained in detail below with reference to the accompanying drawings and specific embodiments.

[0071] like Figure 2 As shown, this embodiment provides a MOSFET device modeling method based on a multi-physics constraint neural network, including the following steps:

[0072] Step 1: Obtain the MOSFET device at different drain-source voltages V ds Under conditions ranging from minimum to maximum, the gate-source voltage V gs The device's data during the change process (from minimum to maximum value), including temperature and MOSFET gate oxide thickness T. ox The semiconductor doping concentration N of MOSFET d The raw data such as voltage, current, capacitance, transconductance, on-resistance, and surface potential, as well as the characteristic data such as mean, standard deviation, kurtosis, skewness, and slope of the raw data, are used as the dataset required for training the MOSFET physical constraint neural network model.

[0073] Step 2: Preprocess the obtained dataset using data cleaning and standardization methods, delete erroneous data from the data collection stage, adjust similar data to the same unit of measurement, and then adjust all data to appropriate units of measurement using data processing methods such as standardization and normalization.

[0074] Step 3: According to Figure 1 As shown, a multi-physics-constrained neural network model is constructed. This model includes a main neural network and two physical constraint neural networks (i.e., the first neural network, the second neural network, and the third neural network), which are used to solve the surface potential at the MOSFET channel, semiconductor charge, and transport equations, respectively. The neural network model includes an input layer, an output layer, and multiple hidden layers. Each layer contains several neurons. Each neuron calculates the inner product of the input vector and the weights, which is then processed by an activation function and connected to other neurons. The number of hidden layers and the number of neurons in each hidden layer can be adjusted to avoid overfitting and underfitting problems in the neural network model.

[0075] In some embodiments, see Figure 1 , Figure 1 This is a diagram of the MOSFET model architecture based on a physically constrained neural network. The first neural network serves as the main neural network of the model, and its descriptive equation is:

[0076] NN(V gs V ds ,I ds ,T emp ,T ox N d )=φ s_pred

[0077] The input parameters of this neural network include: gate-source voltage V gs Drain-source voltage V ds Drain-source current I ds Temperature T emp MOSFET gate oxide thickness T ox The semiconductor doping concentration N of MOSFET d The output parameter is the calculated value of the channel surface potential φ. s_pred .

[0078] As one implementation method, the modeling method of this embodiment uses charge balance and transport equations as constraints that must be followed in the design and training of the neural network model.

[0079] Charge balance refers to the charge neutrality condition Q at the oxide-semiconductor interface in the vertical direction during normal operation of a MOSFET. gate +Q semi =0, where Q gate Q is the gate charge.semi This refers to semiconductor charge. The specific calculation methods for gate charge and semiconductor charge are as follows:

[0080]

[0081] Q semi =f(φ s_pred ,T emp N d )=NN(φ s_pred ,T emp N d )

[0082] Where, ε ox V is the dielectric constant of the oxide layer. fb For flat-band voltage, φ s_pred For the surface potential calculated by the neural network, f(φ) s_pred ,T emp N d φ is a precise calculation function for semiconductor charge. This function is difficult to solve; therefore, this implementation uses a neural network NN(φ). s_pred ,T emp N d Fit the function.

[0083] The descriptive equation for the second neural network is:

[0084] Q gate +NN(φ s_pred ,T emp N d ) = 0

[0085] The input parameters of this neural network model include: surface potential φ s_pred Temperature T emp The semiconductor doping concentration N of MOSFET d The output parameter is the calculated semiconductor charge value Q. semi_pred .

[0086] The transport equation describes the relationship between current and carrier movement within a semiconductor, including drift and diffusion. Analytical calculations require consideration of carrier distribution and electric field distribution. This implementation uses a neural network NN(φ). s_pred V ds ,T emp N d ) = I ds_pred Fitting the drain-source current I of the MOSFET ds .

[0087] Step 4: This implementation scheme uses the charge balance and carrier transport equations in the vertical channel direction during MOSFET operation as physical constraints. The multi-physics-constrained neural network model measures the model's fit to the MOSFET state before and after training by calculating the error function value Loss. The weights and biases of neurons are adjusted by sequentially calculating the partial derivatives of the output with respect to the input. In the physical constraint neural network, the overall computational error of the model includes the main neural network computational error Loss. data And physical constraint calculation error Loss physics1 and Loss physics2 The specific calculation method is as follows:

[0088] The mean square error of the charge balance residual is used as the physical loss. physics1 :

[0089] Loss physics1 =MSE(Q gate +Q semi_pred ,0)

[0090] The mean square error between the predicted and measured drain-source currents is calculated as the physical loss. physics2 :

[0091] Loss physics2 =MSE(I ds_pred -I ds ,0)

[0092] The total loss of the model is:

[0093] Loss = w data Loss data +w physics1 Loss physics1 +w physics2 Loss physics2

[0094] Where Loss is the total model error, Loss data Specifically, the calculation error of the channel surface potential neural network, Loss physics1 Loss represents the calculation error of the charge balance condition in the vertical channel direction of the MOSFET. physics2 For the physical loss in the MOSFET delivery equation, w data w physics1 and w physics2 These are the weights for the neural network calculation error and the weights for the two physical constraint errors, respectively, and their magnitudes are adjustable.

[0095] Step 5: Iteratively train the neural network model using the preprocessed data until model evaluation metrics such as model error reach the expected levels. The model training is then complete. Figure 3As shown.

[0096] In summary, compared with the prior art, this embodiment has at least the following advantages and beneficial effects:

[0097] (1) The modeling method based on multi-physics constraint neural network provided by the present invention can accurately fit the behavior characteristics of MOSFET devices during operation.

[0098] (2) Unlike analytical physics-based modeling methods, the method of this invention can complete the MOSFET device model design using measurable physical quantities. Furthermore, the model training and usage consumes low computational resources and achieves high accuracy. Unlike single-physics-constrained neural network modeling methods, multi-physics-constrained neural network modeling methods can describe the behavior of MOSFETs from multiple physical law perspectives. Using multiple constraint-limited neural networks for solving (fitting) results in a more accurate representation of real physical laws and achieves better modeling performance.

[0099] (3) Unlike pure data-driven modeling methods, the method of this invention can introduce the charge balance condition and carrier transport equation in the vertical channel direction as physical constraints into the neural network model design process, so that the model satisfies objective physical laws and the model training speed and fitting effect are better than pure data-driven modeling methods.

[0100] (4) The method of the present invention is applicable to modeling of various types of MOSFET devices. The structure and input / output parameters of the model are adjustable. The method is simple in principle, highly portable, and consumes little computing resources. It can be widely applied in academic research and engineering practice.

[0101] Example 2

[0102] This embodiment provides a MOSFET device modeling system based on a multi-physics constraint neural network, including:

[0103] The data acquisition module is used to acquire various data under normal operating conditions of MOSFET devices and preprocess the acquired data to obtain the dataset required for training a multi-physics constraint neural network model.

[0104] The model building module is used to build multi-physics-constrained neural network models. It incorporates charge balance and transport equations as physical constraints into the model design and training process to achieve physical constraint neural network modeling.

[0105] The model training module is used to iteratively train a multi-physics-constrained neural network model using a dataset until the model's performance metrics meet preset requirements.

[0106] Since this system is a MOSFET device modeling system based on a multi-physics constraint neural network according to an embodiment of the present invention, and the principle of solving the problem in this system is similar to that of this method, the implementation of this system can refer to the implementation process of the above method embodiment, and the repeated parts will not be described again.

[0107] Example 3

[0108] This invention also provides an electronic device, which includes a processor and a memory. The memory stores at least one instruction, at least one program, a code set, or an instruction set. The at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to achieve the following: Figure 4 This paper presents a MOSFET device modeling method based on a multi-physics constraint neural network.

[0109] It is understood that the memory may include random access memory (RAM) or read-only memory. Optionally, the memory may include non-transitory computer-readable storage medium. The memory can be used to store instructions, programs, code, code sets, or instruction sets. The memory may include a stored program area and a stored data area, wherein the stored program area may store instructions for implementing an operating system, instructions for at least one function, instructions for implementing the various method embodiments described above, etc.; the stored data area may store data created according to the use of the server, etc.

[0110] A processor may include one or more processing cores. The processor connects to various parts of the server via various interfaces and lines, executing instructions, programs, code sets, or instruction sets stored in memory, and accessing data stored in memory to perform various server functions and process data. Optionally, the processor may be implemented using at least one of the following hardware forms: Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), and Programmable Logic Array (PLA). The processor may integrate one or more of the following: Central Processing Unit (CPU) and Modem. The CPU primarily handles the operating system and applications; the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip without being integrated into the processor.

[0111] Since this electronic device is the electronic device corresponding to the MOSFET device modeling method based on multi-physics constraint neural network in the embodiment of the present invention, and the principle of solving the problem by this electronic device is similar to that of the method, the implementation of this electronic device can refer to the implementation process of the above method embodiment, and the repeated parts will not be described again.

[0112] Example 4

[0113] This invention also provides a computer-readable storage medium storing at least one instruction, at least one program, a code set, or an instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to achieve the following: Figure 4 This paper presents a MOSFET device modeling method based on a multi-physics constraint neural network.

[0114] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0115] Since the storage medium is the storage medium corresponding to the MOSFET device modeling method based on multi-physics constraint neural network in the embodiment of the present invention, and the principle of the storage medium in solving the problem is similar to that of the method, the implementation of the storage medium can refer to the implementation process of the above method embodiment, and the repeated parts will not be described again.

[0116] Example 5

[0117] In some possible implementations, various aspects of the methods of the embodiments of the present invention can also be implemented as a program product comprising program code that, when run on a computer device, causes the computer device to perform the steps of a MOSFET device modeling method based on a multi-physics constraint neural network according to various exemplary embodiments of this application as described above. The executable computer program code or "code" for performing the various embodiments can be written in high-level programming languages ​​such as C, C++, Python, Smalltalk, Java, JavaScript, Visual Basic, Structured Query Language (e.g., Transact-SQL), Perl, or in various other programming languages.

[0118] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0119] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0120] The above embodiments are merely illustrative of the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent changes or modifications made based on the essence of the content of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A MOSFET device modeling method based on a multi-physics constraint neural network, characterized in that, Includes the following steps: Acquire various data under normal operating conditions of MOSFET devices, and preprocess the acquired data to obtain the dataset required for training a multi-physics constrained neural network model; A multi-physics-constrained neural network model is constructed, incorporating charge balance and transport equations as physical constraints into the model design and training process to achieve physical constraint neural network modeling. The multi-physics-constrained neural network model is trained iteratively using the dataset until the model's performance metrics meet the preset requirements. The multi-physics constraint neural network model includes three multi-input single-output neural networks, which are used to solve the surface potential at the MOSFET channel, the semiconductor charge, and the drain-source current of the MOSFET, respectively. The first neural network is used to solve for the surface potential at the MOSFET channel, and its describing equation is: The input parameters of the first neural network include: gate-source voltage. Drain-source voltage Drain source current ,temperature MOSFET gate oxide thickness Semiconductor doping concentration of MOSFETs The output parameter is the calculated value of the channel surface potential. ; The second neural network is used to solve for semiconductor charge, and its describing equation is: The input parameters of the second neural network include: calculated channel surface potential. ,temperature Semiconductor doping concentration of MOSFETs The output parameter is the calculated value of semiconductor charge. ; The third neural network is used to solve for the drain-source current of the MOSFET, and its describing equation is: The input parameters of the third neural network include: calculated channel surface potential. Drain-source voltage ,temperature Semiconductor doping concentration of MOSFETs The output parameter is the calculated drain-source current of the MOSFET. ; The charge balance refers to the charge neutrality condition in the vertical direction at the oxide-semiconductor interface when the MOSFET is operating normally, that is: In the formula, Gate charge, Semiconductor charge; Q semi =f(φ) s_pred ,T emp ,N d )=NN(φ s_pred ,T emp ,N d ) In the formula, ε ox V is the dielectric constant of the oxide layer. fb For flat-band voltage, φ s_pred The surface potential calculated for a neural network; f(φ) s_pred ,T emp N d The charge on a semiconductor is a precise calculation function, which is difficult to solve; therefore, a neural network model is used. Fit the function; The transport equation describes the relationship between current and carrier movement within a semiconductor, including drift and diffusion. The analytical calculation process needs to consider carrier distribution and electric field distribution. Among them, neural networks are used Fitting the drain-source current of the MOSFET .

2. The MOSFET device modeling method based on a multi-physics constraint neural network according to claim 1, characterized in that, The loss function during the training process of the multi-physics constrained neural network model is: Loss=w data Loss data +w physics1 Loss physics1 +w physics2 Loss physics2 Loss data =MSE(φ s_pred -f s ,0) Loss physics1 =MSE(Q gate +Q semi_pred ,0) Loss physics2 =MSE(I ds_pred -I ds ,0) In the formula, This represents the total error of the model. This refers to the computational error of the neural network. Physical losses due to MOSFET charge balance For the physical losses in the MOSFET delivery equation, , and These are the weights for the neural network calculation error and the weights for the two physical constraint errors, respectively. The mean squared error function represents the average of the sum of squares of the numerical errors of the two types of data.

3. The MOSFET device modeling method based on a multi-physics constraint neural network according to claim 1, characterized in that, The acquisition of various data under normal operating conditions of the MOSFET device includes: Obtain MOSFET devices at different drain-source voltages Under the condition, gate-source voltage During the change process, various data from the device were obtained, including raw data and characteristic data of the raw data; among them, the raw data included temperature. MOSFET gate oxide thickness Semiconductor doping concentration of MOSFETs The raw data includes voltage, current, capacitance, transconductance, on-resistance, and surface potential. The characteristic data of the raw data include the mean, standard deviation, kurtosis, skewness, and slope of the raw data.

4. A MOSFET device modeling system based on a multi-physics constraint neural network, characterized in that, include: The data acquisition module is used to acquire various data under normal operating conditions of MOSFET devices, and to preprocess the acquired data to obtain the dataset required for training the multi-physics constraint neural network model. The model building module is used to build multi-physics-constrained neural network models. It incorporates charge balance and transport equations as physical constraints into the model design and training process to achieve physical constraint neural network modeling. The model training module is used to iteratively train a multi-physics-constrained neural network model using a dataset until the model's performance metrics meet preset requirements. The multi-physics constraint neural network model includes three multi-input single-output neural networks, which are used to solve the surface potential at the MOSFET channel, the semiconductor charge, and the drain-source current of the MOSFET, respectively. The first neural network is used to solve for the surface potential at the MOSFET channel, and its describing equation is: The input parameters of the first neural network include: gate-source voltage. Drain-source voltage Drain source current ,temperature MOSFET gate oxide thickness Semiconductor doping concentration of MOSFETs The output parameter is the calculated value of the channel surface potential. ; The second neural network is used to solve for semiconductor charge, and its describing equation is: The input parameters of the second neural network include: calculated channel surface potential. ,temperature Semiconductor doping concentration of MOSFETs The output parameter is the calculated value of semiconductor charge. ; The third neural network is used to solve for the drain-source current of the MOSFET, and its describing equation is: The input parameters of the third neural network include: calculated channel surface potential. Drain-source voltage ,temperature Semiconductor doping concentration of MOSFETs The output parameter is the calculated drain-source current of the MOSFET. ; The charge balance refers to the charge neutrality condition in the vertical direction at the oxide-semiconductor interface when the MOSFET is operating normally, that is: In the formula, Gate charge, Semiconductor charge; Q semi =f(φ) s_pred ,T emp ,N d )=NN(φ s_pred ,T emp ,N d ) In the formula, ε ox V is the dielectric constant of the oxide layer. fb For flat-band voltage, φ s_pred The surface potential calculated for a neural network; f(φ) s_pred ,T emp N d φ is a precise calculation function for semiconductor charge. This function is difficult to solve, therefore a neural network model NN(φ) is used. s_pred ,T emp N d Fit the function; The transport equation describes the relationship between current and carrier movement within a semiconductor, including drift and diffusion. The analytical calculation process needs to consider carrier distribution and electric field distribution. Among them, the neural network NN(φ) is used s_pred V ds ,T emp N d ) = I ds_pred Fitting the drain-source current of the MOSFET .

5. An electronic device, characterized in that, The electronic device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set, or an instruction set, and the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the method as described in any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that, The storage medium stores at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the method as described in any one of claims 1 to 3.

7. A computer program product, characterized in that, The computer program product includes computer instructions that, when executed by a processor, are used to perform the method as described in any one of claims 1 to 3.

Images