Transient dose rate effect simulation modeling method based on neural network
Through the transient dose rate effect simulation modeling method based on neural network, the simulation process is simplified, the simulation efficiency and accuracy are improved, and the accuracy of radiation resistance performance evaluation in the existing technology is solved, so that the simulation can be better The circuit's response under radiation.
Patent Information
- Application Number
- CN202410968099.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-18
- Publication Date
- 2024-11-19
AI Technical Summary
Existing instantaneous dose rate effect simulation methods have shortcomings in accuracy and efficiency, especially when rapidly evaluating and verifying the radiation resistance of advanced process devices, the existing methods are costly, time-consuming and limited in accuracy.
Using the transient dose rate effect simulation modeling method based on neural network, the photocurrent pulse curve is obtained by obtaining the radiation dose rate and circuit bias conditions, and the photocurrent pulse curve is obtained by performing radiation simulation or experiment, sampling to form a training data set, and using a multi-layer feedforward neural network Structural training of models, simplify modeling process, and improve simulation accuracy.
This method simplifies the modeling process, improves simulation efficiency and accuracy, and can better simulate the true response of the circuit under instantaneous dose rate effect irradiation, reducing the dependence on complex physical models.
Smart Images

Figure CN118981941A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of computer technology, and in particular to a method for simulating and modeling instantaneous dose rate effects based on a neural network. Background Art
[0002] With the continuous advancement of integrated circuit technology, the feature size is getting smaller and smaller, the speed is getting faster and faster, and the scale of integrated circuits is getting larger and larger, which brings challenges to the radiation hardening of integrated circuits. For example, the high-dose rate pulsed gamma rays produced by nuclear explosions interact with silicon materials, generating a large amount of charge in silicon materials through the Compton effect. The additional charge in the material is collected under the drift and diffusion mechanism to form a transient photocurrent. When the transient photocurrent is generated and propagated in the integrated circuit, it will induce disturbances, latching, flipping and even burning in specific structures. Due to the high penetration of the pulsed gamma ray, its energy loss in the material is evenly distributed, and the charge generated by ionization is also evenly distributed. Therefore, transient photocurrents will be generated simultaneously at various sensitive locations of the integrated circuit, thus forming a global transient photocurrent. In order to improve the working performance of semiconductors in space or ground radiation environments, it is necessary to first clarify the transient dose rate characteristics of integrated circuits in radiation environments.
[0003] At present, radiation source simulation test and computer simulation test are the main methods to simulate the transient dose rate effect. Radiation source simulation test is to use artificial radiation source to irradiate the chip, simulate the situation of electronic devices being irradiated by high-energy particles in the ground or space radiation environment, and observe the response of the chip to the radiation particles. However, this experiment is costly and time-consuming, and it is difficult to study the propagation law of the transient dose rate effect in the circuit. Computer simulation test mainly includes three methods: device level simulation, circuit level simulation, and device / circuit level mixed simulation. Designers generally use computer simulation test method to evaluate the radiation resistance of the circuit in the design stage, that is, use computer software to simulate the transient dose rate transient pulse, add it to the circuit to be simulated in the form of fault injection, and evaluate the circuit's radiation resistance by evaluating the circuit's resistance to the fault injection model, but the accuracy is limited. In order to obtain an accurate model of the instantaneous dose rate effect, one modeling method is to model the instantaneous dose rate effect based on physical mechanisms. For example, starting from the one-dimensional semiconductor control equation, the classic PN junction instantaneous dose rate effect pulse photocurrent model is obtained after reasonable simplification. Although the constructed instantaneous dose rate effect model has high accuracy, for advanced process devices, the development of a simple model that describes the internal physical process of semiconductor devices in a simplified mathematical form is difficult and time-consuming, which is not conducive to the rapid evaluation and verification of device performance; another modeling method is an instantaneous dose rate effect model based on empirical specific functions (such as: double exponential function, Weibull function, etc.), which requires a large number of radiation tests. In addition, as an independent source, the model cannot accurately reflect the actual response of circuit nodes to instantaneous dose rate effect pulses. Summary of the invention
[0004] The main purpose of the embodiments of the present application is to propose a simulation modeling method of instantaneous dose rate effect based on a neural network, aiming to improve simulation accuracy and simplify calculation difficulty.
[0005] To achieve the above object, one aspect of an embodiment of the present application proposes a method for simulating and modeling the instantaneous dose rate effect based on a neural network, comprising the following steps:
[0006] Obtain radiation dose rates and circuit bias conditions;
[0007] Inputting the radiation dose rate and the circuit bias condition into an instantaneous dose rate effect model based on a neural network to obtain a device photocurrent response result;
[0008] Wherein, the instantaneous dose rate effect model is obtained by the following steps:
[0009] By performing radiation simulation or radiation experiment on the target device, a photocurrent pulse curve is obtained;
[0010] Sampling the photocurrent pulse curve to obtain a training data set, wherein the training data set includes discrete data of photocurrent pulses under different times, different bias voltages, and different dose rates;
[0011] The training data set is input into the initialized neural network for training to obtain an instantaneous dose rate effect model.
[0012] In some embodiments, the radiation simulation process of the target device includes the following steps:
[0013] Use computer-aided design technology to build a three-dimensional device physical model of the target device;
[0014] The radiation time and the radiation dose rate are set, and the instantaneous dose rate effect simulation is performed using the three-dimensional device physical model to obtain a photocurrent pulse curve.
[0015] In some embodiments, sampling the photocurrent pulse curve to obtain a training data set includes the following steps:
[0016] Dividing the photocurrent pulse curve into a front section of the pulse curve and a rear section of the pulse curve;
[0017] uniformly sampling the front section of the pulse curve using a first sampling frequency to obtain a plurality of first discrete data;
[0018] uniformly sampling the latter part of the pulse curve using a second sampling frequency to obtain a plurality of second discrete data, wherein the first sampling frequency is greater than the second sampling frequency;
[0019] A training data set is formed according to a plurality of the first discrete data and a plurality of the second discrete data.
[0020] In some embodiments, the instantaneous dose rate effect model adopts a multi-layer feedforward neural network structure, and the instantaneous dose rate effect model includes an input layer, a normalization layer, a plurality of hidden layers and an output layer connected in sequence;
[0021] The input layer is used to obtain multiple input features, wherein the multiple input features include radiation time, circuit bias conditions and radiation dose rate;
[0022] The normalization layer is used to perform a normalization operation on the input features;
[0023] The hidden layer is used to extract the normalized input features to obtain multiple feature representations;
[0024] The output layer is used to perform photoelectric pulse discrimination prediction based on the multiple feature representations, and normalize the prediction results to obtain the device photocurrent response results.
[0025] In some embodiments, inputting the training data set into the initialized neural network for training to obtain the instantaneous dose rate effect model comprises the following steps:
[0026] Inputting the training data set into the initialized neural network for prediction to obtain a photoelectric pulse prediction result;
[0027] Calculating model loss according to the photoelectric pulse prediction result and the corresponding photocurrent pulse discrete data in the training data set;
[0028] An adaptive moment estimation optimizer and a preset learning rate are used to continuously update the parameters of the neural network according to the model loss until the photoelectric pulse prediction result output by the neural network reaches a preset accuracy, thereby obtaining an instantaneous dose rate effect model.
[0029] In some embodiments, the number of neurons in the hidden layer is determined by the following formula:
[0030]
[0031] Where n is the number of neurons in the input layer, m is the number of neurons in the output layer, and a is the adjustment factor, which ranges from [1-10].
[0032] In some embodiments, the neural network-based transient dose rate effect simulation modeling method further includes the following steps:
[0033] The instantaneous dose rate effect model is saved in a pkl format file, wherein the pkl format file includes the weight and bias of each neuron node of the instantaneous dose rate effect model;
[0034] The instantaneous dose rate effect model is converted into a Verilog-A current source model according to the information in the pkl format file.
[0035] The instantaneous dose rate effect simulation modeling method based on a neural network proposed in the present application obtains a photocurrent pulse curve of the circuit response by performing radiation simulation or radiation experiment on the target device, samples the photocurrent pulse curve to obtain discrete data of photocurrent pulses under different times, different bias voltages, and different dose rate conditions, forms a training data set, and inputs the training data set into the initialized neural network for training to obtain an instantaneous dose rate effect model. It is not necessary to use a complex physical model to model each device, simplifies the modeling process, and improves the modeling efficiency. The model considers the influence of the instantaneous dose rate effect under various bias conditions, and can better simulate the real response of the circuit after being irradiated by the instantaneous dose rate effect. A more accurate instantaneous dose rate effect model is obtained by learning and fitting through a neural network, and the circuit is simulated for radiation using the model of the present application to improve the accuracy of the device photocurrent response results. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] Figure 1 It is a flow chart of a method for simulating and modeling instantaneous dose rate effect based on a neural network provided in an embodiment of the present application;
[0037] Figure 2 yes Figure 1 The instantaneous dose rate effect model construction flow chart in step S102;
[0038] Figure 3 It is a schematic diagram of the process of constructing the transient dose rate effect model provided in the embodiment of the present application;
[0039] Figure 4 It is a schematic diagram of the structure of the instantaneous dose rate effect model provided in the embodiment of the present application;
[0040] Figure 5 It is a schematic diagram comparing the effects of the instantaneous dose rate effect model provided in the embodiments of the present application. DETAILED DESCRIPTION
[0041] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and 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.
[0042] It should be noted that the terms "first", "second", etc. in the specification, claims and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.
[0043] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.
[0044] First, some nouns involved in this application are analyzed:
[0045] TCAD (Technology Computer Aided Design) is a semiconductor process and device simulation tool that can simulate and optimize the performance and characteristics of semiconductor devices. It is based on the principles of semiconductor physics and uses numerical simulation technology to simulate the physical processes inside the device, thereby evaluating and optimizing different process conditions, device structures and circuit performance. TCAD simulation can simulate the manufacturing process of the device, including material growth, processing technology, doping, chemical mechanical polishing and other processes, as well as the working principle and performance of the device, such as current-voltage characteristics, frequency characteristics, temperature characteristics, etc. Through simulation, the performance of the device can be predicted and the design parameters can be optimized, reducing expensive and time-consuming experimental processes, improving development efficiency and reducing development costs.
[0046] The pkl format is the pickle format. Pickle is a Python-specific serialization format that converts Python objects into a format that can be stored or transmitted. The main purpose of pickle is to serialize Python objects into a binary format so that they can be saved to a file or sent over the network. With pickle, Python objects can be converted into byte streams so that they can be stored on disk or sent over the network to another program or computer. When needed, they can be deserialized back to the original Python object from the byte stream. Pickle can handle almost all Python data types, including numbers, strings, lists, dictionaries, sets, custom objects, and more.
[0047] Verilog-A is a high-level description language used to simulate electronic systems. It can describe the behavior and performance of electronic systems. In Verilog-A, a current source model is a model that describes the behavior of a current source element.
[0048] An HSPICE netlist is a text file that describes a circuit and is commonly used for analog circuit design and analysis. It is an extension of SPICE (Simulation Program with Integrated Circuit Emphasis), a simulation tool widely used to simulate electronic systems. An HSPICE netlist describes the various components of a circuit and their connections, as well as the parameters and properties of the components. It also contains information such as the stimulus source, the operating conditions of the circuit, and test points, which are used to simulate and analyze the performance and behavior of the circuit. HSPICE netlist files are usually written in text format, which is easy to read, modify, and generate.
[0049] The instantaneous dose rate effect simulation modeling method based on a neural network provided in the embodiment of the present application relates to the field of computer technology. The instantaneous dose rate effect simulation modeling method based on a neural network provided in the embodiment of the present application can be applied to a terminal, can also be applied to a server side, and can also be software running in a terminal or a server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or can be configured as a server cluster or a distributed system composed of multiple physical servers, and can also be configured as a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements the instantaneous dose rate effect simulation modeling method based on a neural network, etc., but is not limited to the above forms.
[0050] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.
[0051] Figure 1 is an optional flow chart of the instantaneous dose rate effect simulation modeling method based on neural network provided in the embodiment of the present application, Figure 1 The method may include but is not limited to steps S101 to S102.
[0052] Step S101, obtaining radiation dose rate and circuit bias conditions;
[0053] Step S102, inputting the radiation dose rate and circuit bias conditions into the transient dose rate effect model based on the neural network to obtain the device photocurrent response result.
[0054] Among them, refer to Figure 2 , Figure 2 It is a flow chart for constructing the instantaneous dose rate effect model provided in the embodiment of the present application, including but not limited to steps S201 to S203.
[0055] Step S201, obtaining a photocurrent pulse curve by performing radiation simulation or radiation experiment on the target device;
[0056] Step S202, sampling the photocurrent pulse curve to obtain a training data set, wherein the training data set includes discrete data of photocurrent pulses under different time, different bias voltage, and different dose rate conditions;
[0057] Step S203, inputting the training data set into the initialized neural network for training to obtain an instantaneous dose rate effect model.
[0058] In the steps S101 to S102 and S201 to S203 shown in the embodiment of the present application, the photocurrent pulse curve of the circuit response is obtained by performing radiation simulation or radiation experiment on the target device, and the photocurrent pulse curve is sampled to obtain discrete data of photocurrent pulses under different times, different bias voltages, and different dose rates to form a training data set. The training data set is input into the initialized neural network for training to obtain an instantaneous dose rate effect model. There is no need to use a complex physical model to model each device, which simplifies the modeling process and improves the modeling efficiency. The model considers the influence of the instantaneous dose rate effect under various bias conditions, and can better simulate the actual response of the circuit after being irradiated by the instantaneous dose rate effect. A more accurate instantaneous dose rate effect model is obtained by learning and fitting through a neural network, and the circuit is simulated for radiation using the model of the present application to improve the accuracy of the device photocurrent response results.
[0059] In step S101 of some embodiments, the radiation dose rate represents the intensity of radiation per unit time, and the circuit bias condition may refer to the bias voltage of the semiconductor circuit. Both the radiation dose rate and the circuit bias condition may be obtained through a visual interface, that is, the developer inputs the radiation dose rate and the circuit bias condition on the visual interface, and the input parameters are passed to the transient dose rate effect model for simulation. Furthermore, the input parameters may also include a preset circuit radiation time.
[0060] In step S102 of some embodiments, the instantaneous dose rate effect model is constructed based on a neural network architecture. The neural network can be a multi-layer feedforward neural network to simplify the modeling difficulty. The instantaneous dose rate effect model uses a neural network architecture to more accurately simulate the changes in the photoelectric characteristics of each node of the circuit under a real radiation environment, thereby analyzing the photocurrent response results of the device after radiation.
[0061] Next, combine Figure 3 , the construction process of the transient dose rate effect model of the embodiment of the present application is described in detail.
[0062] In some embodiments of step S201, to train the instantaneous dose rate effect model, a certain amount of sample data needs to be obtained. The sample data can be obtained through a real radiation experiment or through a simulation tool based on a physical model.
[0063] Specifically, the radiation simulation process of the target device may include but is not limited to step S301 to step S302:
[0064] Step S301, constructing a three-dimensional device physical model of a target device using computer-aided design technology;
[0065] Step S302 , setting the radiation time and radiation dose rate, and using a three-dimensional device physical model to simulate the instantaneous dose rate effect to obtain a photocurrent pulse curve.
[0066] In this embodiment, after completing the modeling of the three-dimensional device physical model of the target device and the process experiment calibration using computer-aided design technology, a radiation effect model is added to the three-dimensional device physical model, and parameters such as radiation time and dose rate are set. The three-dimensional device physical model is used to simulate the device / circuit mixed mode instantaneous dose rate effect to obtain the response of the circuit / device to the instantaneous dose rate effect, that is, the primary and secondary photocurrent pulse curves generated by the instantaneous dose rate effect under different conditions are obtained.
[0067] In some embodiments of step S202, since the input data of the instantaneous dose rate effect model is in the form of discrete data points, it is necessary to sample the photocurrent pulse curve according to the accuracy requirements to obtain discrete data of photocurrent pulses under different times, different bias voltages, and different dose rates. Then the sampled discrete data is used as sample data for model training, and the sample data is divided into training data and test data, for example, 70% of the sample data is divided into a training set, and 30% of the sample data is divided into a test set. The training data is mainly used to train model parameters, and the test data is mainly used to verify the accuracy of the model.
[0068] Specifically, in step S202, the step of sampling the photocurrent pulse curve to obtain a training data set may include but is not limited to steps S401 to S404:
[0069] Step S401, dividing the photocurrent pulse curve into a front section of the pulse curve and a rear section of the pulse curve;
[0070] Step S402, uniformly sampling the front section of the pulse curve using a first sampling frequency to obtain a plurality of first discrete data;
[0071] Step S403, uniformly sampling the latter part of the pulse curve using a second sampling frequency to obtain a plurality of second discrete data, wherein the first sampling frequency is greater than the second sampling frequency;
[0072] Step S404: forming a training data set according to the plurality of first discrete data and the plurality of second discrete data.
[0073] In this embodiment, the current change rate at each point of the photocurrent pulse curve is used as a judgment basis to identify the node in the photocurrent pulse curve where the current change rate begins to be less than a preset value, and the photocurrent pulse curve is divided into the front section of the pulse curve and the back section of the pulse curve. The current in the front section of the pulse curve changes faster, and a higher frequency is used for uniform sampling. The current in the back section of the pulse curve changes slower, and a lower frequency is used for uniform sampling. Uniform sampling is performed to reduce the amount of calculation while taking into account the sampling accuracy.
[0074] In some embodiments of step S203, a neural network-based instantaneous dose rate effect model is constructed, and the instantaneous dose rate effect model is trained using a training data set through learning and fitting. Then, the trained instantaneous dose rate effect model is tested for photocurrent accuracy using a test data set to determine the gap between the results output by the instantaneous dose rate effect model and the experimental data or physical simulation data. The smaller the gap, the higher the accuracy of the instantaneous dose rate effect model.
[0075] Specifically, in step S203, the training data set is input into the initialized neural network for training to obtain the instantaneous dose rate effect model, which may include but is not limited to steps S501 to S503:
[0076] Step S501, inputting the training data set into the initialized neural network for prediction to obtain the photoelectric pulse prediction result;
[0077] Step S502, calculating the model loss according to the photoelectric pulse prediction result and the corresponding photocurrent pulse discrete data in the training data set;
[0078] Step S503, using an adaptive moment estimation optimizer and a preset learning rate, the parameters of the neural network are continuously updated according to the model loss until the photoelectric pulse prediction result output by the neural network reaches a preset accuracy, thereby obtaining an instantaneous dose rate effect model.
[0079] In this embodiment, the neural network is initialized first, including initializing the neural network structure and parameters.
[0080] Specifically, the neural network structure can be a multi-layer feedforward neural network, and the neural network structure is as follows:
[0081] Reference Figure 4, the neural network structure (i.e., the structure of the instantaneous dose rate effect model) is constructed as an input layer, a normalization layer, at least one hidden layer, and an output layer connected in sequence. The input layer is used to obtain multiple input features, wherein the multiple input features include radiation time, circuit bias conditions, and radiation dose rate; the normalization layer is used to perform normalization operations on the input features; the hidden layer is used to perform feature extraction on the normalized input features to obtain multiple feature representations; the output layer is used to perform photoelectric pulse discrimination prediction based on the multiple feature representations, and normalize the prediction results to obtain the device photocurrent response results. Since the input features include time, dose rate, and bias voltage, and the units of the input variables are different, a normalization operation is introduced between the input layer and the hidden layer, and a denormalization operation is introduced after the output layer to simplify the calculation process of the multi-layer feedforward neural network and improve the modeling efficiency.
[0082] The number of neurons and layers of a neural network has an important impact on the performance of the network. Too many neurons and layers will make the calculation complicated and seriously slow down the training speed of the network. If the number of neurons and layers is too small, the network learning will not be comprehensive, which will lead to poor prediction results. Therefore, the design of each layer is as follows:
[0083] Input layer: Based on the research and analysis of the instantaneous dose rate effect, it can be seen that the larger the dose rate, the larger the irradiation photocurrent, and the irradiation photocurrent is related to the simulation time, so the dose rate, radiation time and bias voltage are used as input features of the input layer. When there are many input features, correlation analysis is usually used to exclude certain redundant input features and improve the efficiency of model training, which is suitable for cases with many input features. Combined with the actual needs of the embodiment of the present application, the feature quantity of the input layer of the embodiment of the present application is 3, and there is no need to perform correlation analysis. Therefore, the number of input layers is one layer and the number of neurons is 3.
[0084] Hidden layer: The embodiment of the present application can use an empirical formula to estimate the number of neurons in the hidden layer, and slowly adjust it through experiments, and finally adjust it to a value that can make the system reach the optimal value. The above empirical formula is as follows: Where M is the number of neurons, n is the number of neurons in the input layer, m is the number of neurons in the output layer, and a is an adjustment factor, which ranges from [1-10]. By adjusting the number of hidden layers and the number of neurons in each layer, the model fitting accuracy can be improved, and the accuracy of the instantaneous dose rate effect model can be improved. For example, after multiple training adjustments, it is determined that the number of hidden layers is 3, and the number of neurons in each layer is 15, 10, and 8 respectively.
[0085] Output layer: The prediction target of the embodiment of the present application is the photocurrent response of the circuit. Therefore, the number of output layers selected in the embodiment of the present application is 1, and the number of neurons in the output layer is 1, that is, the photocurrent response value is used as the output of the network.
[0086] In addition to initializing the neural network structure, it is also necessary to initialize the neuron parameters and activation functions (weights and biases) of the neural network itself, as well as the related adjustment parameters used in the initialization training process, as follows:
[0087] Neuron parameters can be initialized through pre-training or randomly initialized.
[0088] Activation function, in order to improve the nonlinearity of the neural network, the activation function is introduced into the establishment of the model. Common activation functions include sigmoid, tanh function and ReLU. The sigmoid function is mainly used in classification and has high computational complexity. The tanh function is prone to gradient saturation and has high computational complexity, while the ReLU converges faster than the other two algorithms and has low computational complexity. Its function form is f(x)=max(0,x). Therefore, the embodiment of the present application uses ReLU as the hidden layer activation function, and the activation function of the output layer is f(x)=x.
[0089] Optimizer, the optimizer determines the optimization direction and learning rate update in the neural network, so that the loss function continuously approaches the global minimum. Considering that the BGD (Batch gradient descent) optimizer is globally optimal but converges slowly; the SGD (Stochastic gradient descent) optimizer converges quickly but is prone to fall into local minima; the Adam (Adaptive moment estimation) optimizer converges stably and quickly. In the embodiment of the present application, the Adam optimizer with relatively stable convergence and fast speed can be selected as the optimizer of the neural network.
[0090] The learning rate is also called the step size. It can be used to measure the amount of each update of the neural network's weights, thresholds and other parameters. The range is between 0 and 1. If the learning rate is too large, the weight update will be large and it will not be easy to converge at the minimum value. If the learning rate is too small, the convergence speed will be too slow. Through many experiments, the learning rate can be determined to be 0.001.
[0091] After initializing the structure and related parameters of the neural network, the training data set can be input into the neural network for forward prediction to obtain the photoelectric pulse prediction result. Then, the loss function is used to calculate the model loss based on the photoelectric pulse prediction result and the corresponding photocurrent pulse discrete data in the training data set (i.e., the real photoelectric pulse response). When the model loss is too large, an adaptive moment estimation optimizer and a preset learning rate are used to update the parameters of the neural network according to the model loss. Repeat the above process based on different sample data until the training reaches the number of iterations or the model loss is small, and then use the test data set to verify the model. Input the test data into the instantaneous dose rate effect model to obtain a test curve, and compare the test curve with the simulated photocurrent pulse curve to determine whether the error is lower than the set threshold: if not, adjust the parameters of the instantaneous dose rate effect model of the neural network and continue training; if so, the training is completed and the instantaneous dose rate effect model is obtained. After multiple learning iterations, the instantaneous dose rate effect current source model based on the neural network is finally obtained. Figure 5 As shown, after 1000 learning iterations, the photocurrent test curve output by the instantaneous dose rate effect model (left figure) is compared with the measured result (right figure). It can be seen that at different dose rates, the fitting accuracy of each group of instantaneous dose rate effect current source model test curve and the measured photocurrent pulse curve is high.
[0092] In some embodiments, the instantaneous dose rate effect simulation modeling method based on a neural network in the embodiment of the present application may also include but is not limited to steps S601 to S602:
[0093] Step S601, using a pkl format file to save the instantaneous dose rate effect model, wherein the pkl format file includes the weight and bias of each neuron node of the instantaneous dose rate effect model;
[0094] Step S602: converting the instantaneous dose rate effect model into a Verilog-A current source model according to the information in the pkl format file.
[0095] In this embodiment, please continue to refer to Figure 3 ,After the training and obtaining the instantaneous dose rate effect model expressed in Python language, the instantaneous dose rate effect model is automatically saved in a pkl format file, which contains the weight and bias of each node, the weight and bias information of each node saved in the pkl format file is read, and converted into a Verilog-A current source model suitable for circuit simulation.
[0096] In the application process, if the injection of instantaneous dose rate is realized based on the HSPICE netlist simulation method, it is necessary to read the converted va format file, instantiate the bias-dependent Verilog-A current source model, define the two ports of the model and the parameters involved in the model, such as the dose rate. If the circuit-level netlist with the instantaneous dose rate effect model is simulated using the Spectre simulation tool integrated in the Cadence ADE simulation environment, it is necessary to complete the compilation of the Verilog-A model and generate a symbol. The generated current source symbol can be added to the node where you want to inject through the graphical operation interface.
[0097] The Verilog-A current source of the embodiment of the present application is a controlled current source that can change with the voltage bias, and can better simulate the actual response of the circuit after being irradiated with the transient dose rate effect, which is consistent with the experiment.
[0098] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.
[0099] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.
[0100] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above may be implemented as software, firmware, hardware, or a suitable combination thereof.
[0101] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein.
[0102] The technical solution of the present application, or the part that contributes to the prior art, or the whole or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes multiple instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store programs.
[0103] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.
Claims
1. A method for simulating and modeling transient dose rate effects based on a neural network, characterized in that: The following steps are involved: Obtain radiation dose rate and circuit bias conditions; Inputting the radiation dose rate and the circuit bias condition into an instantaneous dose rate effect model based on a neural network to obtain a device photocurrent response result; Wherein, the instantaneous dose rate effect model is obtained by the following steps: By performing radiation simulation or radiation experiment on the target device, a photocurrent pulse curve is obtained; Sampling the photocurrent pulse curve to obtain a training data set, wherein the training data set includes discrete data of photocurrent pulses under different times, different bias voltages, and different dose rates; The training data set is input into the initialized neural network for training to obtain an instantaneous dose rate effect model.
2. The method for simulating and modeling the transient dose rate effect based on a neural network according to claim 1, characterized in that: The radiation simulation process of the target device comprises the following steps: Use computer-aided design technology to build a three-dimensional device physical model of the target device; The radiation time and the radiation dose rate are set, and the instantaneous dose rate effect simulation is performed using the three-dimensional device physical model to obtain a photocurrent pulse curve.
3. The method for simulating and modeling the transient dose rate effect based on a neural network according to claim 1, characterized in that: The step of sampling the photocurrent pulse curve to obtain a training data set comprises the following steps: Dividing the photocurrent pulse curve into a front section of the pulse curve and a rear section of the pulse curve; uniformly sampling the front section of the pulse curve using a first sampling frequency to obtain a plurality of first discrete data; uniformly sampling the latter part of the pulse curve using a second sampling frequency to obtain a plurality of second discrete data, wherein the first sampling frequency is greater than the second sampling frequency; A training data set is formed according to a plurality of the first discrete data and a plurality of the second discrete data.
4. The method for simulating and modeling transient dose rate effects based on a neural network according to claim 1, characterized in that: The instantaneous dose rate effect model adopts a multi-layer feedforward neural network structure, and the instantaneous dose rate effect model includes an input layer, a normalization layer, a plurality of hidden layers and an output layer connected in sequence; The input layer is used to obtain multiple input features, wherein the multiple input features include radiation time, circuit bias conditions and radiation dose rate; The normalization layer is used to perform a normalization operation on the input features; The hidden layer is used to extract the normalized input features to obtain multiple feature representations; The output layer is used to perform photoelectric pulse discrimination prediction based on the multiple feature representations, and normalize the prediction results to obtain the device photocurrent response results.
5. The method for simulating and modeling the transient dose rate effect based on a neural network according to claim 4, characterized in that: The step of inputting the training data set into the initialized neural network for training to obtain an instantaneous dose rate effect model comprises the following steps: Inputting the training data set into the initialized neural network for prediction to obtain a photoelectric pulse prediction result; Calculating model loss according to the photoelectric pulse prediction result and the corresponding photocurrent pulse discrete data in the training data set; An adaptive moment estimation optimizer and a preset learning rate are used to continuously update the parameters of the neural network according to the model loss until the photoelectric pulse prediction result output by the neural network reaches a preset accuracy, thereby obtaining an instantaneous dose rate effect model.
6. The method for simulating and modeling transient dose rate effects based on a neural network according to claim 4, characterized in that: The number of neurons in the hidden layer is determined by the following formula: Where n is the number of neurons in the input layer, m is the number of neurons in the output layer, and a is the adjustment factor, which ranges from [1-10].
7. The method for simulating and modeling transient dose rate effects based on a neural network according to claim 1, characterized in that: The instantaneous dose rate effect simulation modeling method based on neural network also includes the following steps: The instantaneous dose rate effect model is saved in a pkl format file, wherein the pkl format file includes the weight and bias of each neuron node of the instantaneous dose rate effect model; The instantaneous dose rate effect model is converted into a Verilog-A current source model according to the information in the pkl format file.
