Method for Establishing Device-Level Radiation Hardening Prediction Model Based on Orthogonal Design and Neural Network

Abstract

The present invention provides a method for establishing a device-level anti-radiation hardening prediction model based on orthogonal design and neural network to solve the problem that the anti-radiation hardening design of aerospace electronic devices requires a large number of process exploratory tests to obtain a relatively optimized hardening scheme, resulting in a long development cycle and high development cost of aerospace anti-radiation hardened products. Based on the device TCAD simulation model calibrated by device ground radiation test data and device key electrical parameter test data, this method uses simulation to obtain the changes in device anti-radiation performance and key electrical parameters under different hardening conditions based on the orthogonal design method, and uses neural network fitting to establish a prediction model for the changes in device anti-radiation performance and key electrical parameters under complex hardening conditions. Based on the established prediction model, further obtain the optimal or approximate optimal hardening scheme through search algorithms such as genetic algorithm and simulated annealing algorithm, so as to guide the preparation of aerospace anti-radiation hardened products.

Description

Technical Field

[0001] The present invention relates to a method for radiation hardening of electronic devices for aerospace applications, and particularly to a method for establishing a device-level radiation hardening prediction model based on orthogonal design and neural network. Background Art

[0002] There are a large number of high-energy particles and rays in cosmic space, such as neutrons, protons, high-energy electrons, heavy ions, X-rays, γ-rays, etc. When a spacecraft operates in space, it may be irradiated by these particles and rays, causing disturbances in the electronic system of the spacecraft and even leading to the failure of the electronic system of the spacecraft, seriously endangering the on-orbit safe operation of the spacecraft. In order to ensure the long-term stable on-orbit operation of the spacecraft, electronic devices for aerospace applications must meet certain radiation resistance index requirements. For example, for long-life satellites in geostationary orbit, it is required that the linear energy transfer (LET) of the power device against single event burnout (SEB) is greater than 75 MeV·cm 2 / mg. Therefore, electronic devices for aerospace applications need to be subjected to corresponding radiation hardening designs before being carried on spacecraft.

[0003] However, radiation hardening measures may have an adverse impact on the electrical performance of the device. For example, for an N-channel VDMOS (Vertical Double-Diffused Metal-Oxide-Semiconductor Field-Effect Transistor), increasing the doping concentration in the body region can reduce the base transport factor of the parasitic BJT (Bipolar Junction Transistor), suppress the current amplification characteristics of the parasitic BJT, and thus improve the single event burnout resistance of the N-channel VDMOS device. However, at the same time, increasing the doping concentration in the body region will also cause an increase in the threshold voltage of the device and reduce the forward drive ability of the device. Another example is that reducing the minority carrier lifetime will accelerate the recombination of carriers induced by high-energy particles, thereby reducing the local current density inside the device and reducing the local heat accumulation inside the device, achieving the purpose of improving the single event burnout resistance of the N-channel VDMOS device. However, this hardening measure will cause a significant increase in the leakage current in the off state of the device.

[0004] In the anti - radiation hardening design of aerospace electronic devices, if only one hardening measure is used to improve the anti - radiation ability of the device, it may cause a serious degradation of a certain electrical parameter of the device, and even fail to meet the application requirements of spacecraft. In order to balance the technical requirements of the aerospace electronic system for the anti - radiation index and electrical performance index of the device, it may be necessary to use several hardening means simultaneously to achieve an optimal compromise between the electrical performance of the device and the anti - radiation hardening design. During the actual development process of anti - radiation hardened products, a relatively optimized hardening scheme can often only be obtained through a large number of process exploration tests, resulting in a relatively long development cycle and high development cost for aerospace anti - radiation hardened products. Summary of the Invention

[0005] The object of the present invention is to solve the technical problem that when anti - radiation hardening is carried out on aerospace electronic devices at present, in order to balance the technical requirements of the aerospace electronic system for the anti - radiation index and electrical performance index of the device, a relatively optimized hardening scheme can only be obtained through a large number of process exploration tests, resulting in a long development cycle and high development cost for aerospace anti - radiation hardened products. The present invention provides a method for establishing a device - level anti - radiation hardening prediction model based on orthogonal design and neural network.

[0006] The design idea of the present invention is as follows: First, based on the device TCAD (Semiconductor Process and Device Simulation Software) simulation model calibrated by the device ground irradiation test data and the device key electrical parameter test data, the changes of the anti - radiation performance and key electrical parameters of the device under different hardening conditions are obtained by TCAD simulation using the orthogonal design method. Then, the neural network fitting technology is used to establish a prediction model for the changes of the anti - radiation performance and key electrical parameters of the device under complex hardening conditions. This prediction model can cover three or more anti - radiation hardening technologies simultaneously and consider the interaction between these hardening technologies. Based on the prediction model established in the present invention, further search algorithms such as genetic algorithm and simulated annealing algorithm are used to search for the optimal or approximately optimal hardening scheme, so that while the anti - radiation performance of the hardened device is significantly improved, the key electrical parameters of the device can also be fully optimized, thereby maximizing the overall performance of aerospace electronic devices.

[0007] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0008] A method for establishing a device - level anti - radiation hardening prediction model based on orthogonal design and neural network, which is characterized by including the following steps:

[0009] Step 1: For the initial device structure to be hardened, establish a TCAD simulation model, and calibrate the TCAD simulation model according to the actual test data of the key electrical parameters of the device to be hardened and the ground irradiation test data.

[0010] Step 2: Select at least three common radiation hardening technologies and conduct preliminary TCAD simulations respectively to obtain the influence laws of different hardening technologies on the radiation resistance performance and key electrical parameters of the device;

[0011] Step 3: According to the influence laws obtained in Step 2, select m feasible hardening technologies, divide the parameter variation range of a single hardening technology into n - 1 equal parts, and establish an orthogonal design table with m factors and n levels using the orthogonal design method; where m ≥ 3, n ≥ 5, and both m and n are integers;

[0012] Step 4: Conduct TCAD simulations according to the orthogonal design table to obtain discrete data on the changes in the radiation resistance performance and key electrical parameters of the device under the action of a single hardening technology or multiple hardening technologies;

[0013] Step 5: Randomly divide the discrete data obtained from the simulation in Step 4 into a training set, a validation set, and a test set;

[0014] Step 6: Construct a prediction model for the changes in the radiation resistance performance and key electrical parameters of the device under complex hardening conditions based on a neural network, use the training set data for iterative learning, and at the same time use the validation set data for accuracy verification to obtain the prediction results of the prediction model for the changes in the radiation resistance performance and key electrical parameters of the device to be hardened;

[0015] Step 7: Use the test set data to verify whether the error between the prediction results of the prediction model established in Step 6 and the TCAD simulation results in Step 4 is lower than the set threshold; if so, the training is completed and the prediction model is obtained; if not, return to Step 6 and modify the network parameters of the prediction model for the changes in the radiation resistance performance and key electrical parameters of the device under complex hardening conditions based on the neural network until the error between the prediction results and the simulation results is lower than the set threshold to obtain the prediction model.

[0016] Further, in Step 6, the neural network is a multi - layer feed - forward neural network, including an input layer, an output layer, and at least one hidden layer; the number of neurons in the input layer is the same as the number m of the selected hardening technologies in Step 3; there is at least 1 neuron in the hidden layer; the number of neurons in the output layer is the sum of the number of device key electrical parameter indicators and the number of radiation resistance performance indicators.

[0017] Further, in Step 6, the neural network further includes a normalization module arranged between the input layer and the hidden layer, and a denormalization module arranged after the output layer.

[0018] Further, in Step 6, the hidden layer uses the tansig function as the transfer function, and the output layer uses the pureline function as the transfer function, and their specific formulas are as follows:

[0019]

[0020] f pureline (x) = x

[0021] where: f tansi g(x) is the tansig function; f pureline (x) is the pureline function, and x is the function input variable.

[0022] Furthermore, in step 6, the Levenberg–Marquardt algorithm is adopted when learning using the training set data.

[0023] Furthermore, in step 5, the setting ratios of the training set, validation set, and test set are as follows: the training set is 65% - 75%, the validation set is 10% - 20%, and the test set is 10% - 20%.

[0024] Furthermore, in step 7, the error is the mean square error, and its set threshold is 0.0001.

[0025] Furthermore, in step 1, the device to be strengthened is an N-channel VDMOS device, and the performance to be strengthened is the single-event burnout resistance performance; the key electrical parameters include the breakdown voltage, threshold voltage, and on-resistance parameter.

[0026] Furthermore, in step 2, the multiple anti-irradiation strengthening technologies include 4 strengthening technologies: optimizing the doping concentration and thickness of the buffer layer, increasing the doping concentration of the body region, and increasing the lateral length of the P+ region.

[0027] Compared with the prior art, the beneficial technical effects of the present invention are as follows:

[0028] 1. The method for establishing a device-level anti-irradiation strengthening prediction model based on orthogonal design and neural network disclosed by the present invention first calibrates the device TCAD simulation model based on the device ground irradiation test data and the device key electrical parameter test data, and then uses the orthogonal design method to obtain the changes in the device anti-irradiation performance and key electrical parameters under different strengthening conditions through TCAD simulation, and then uses the neural network fitting technology to establish a prediction model for the changes in the device anti-irradiation performance and key electrical parameters under complex strengthening conditions.

[0029] 2. Based on calibrating the device TCAD simulation model, the present invention uses TCAD simulation to obtain the changes in the device anti-irradiation performance and key electrical parameters under different strengthening conditions, effectively avoiding the problems of time tension and high test costs of the ground irradiation test machine, and improving the modeling efficiency.

[0030] 3. The present invention studies the changes in the radiation resistance performance and key electrical parameters of devices under different reinforcement conditions based on an orthogonal design table. Orthogonality is used to select some representative reinforcement conditions for TCAD simulation, which greatly reduces the number of required TCAD simulations and saves TCAD simulation time.

[0031] 4. The present invention uses a multi-layer feedforward neural network for learning and fitting, introduces multiple hidden layers, and can improve the model fitting accuracy and ultimately the accuracy of the prediction model by adjusting the number of hidden layers and the number of neurons in each layer.

[0032] 5. The prediction model established by the present invention for the changes in the radiation resistance performance and key electrical parameters of devices under complex reinforcement conditions based on a neural network can cover three or more radiation resistance reinforcement technologies at the same time and consider the interactions between these reinforcement technologies, so as to predict the changes in the radiation resistance performance and key electrical parameters of a device under a certain reinforcement technology or the combined action of these reinforcement technologies.

[0033] 6. The input layer of the multi-layer feedforward neural network adopted by the present invention is composed of different reinforcement parameters, and the units of the input variables may be different. Therefore, a normalization operation is introduced between the input layer and the hidden layer, and a denormalization operation is introduced after the output layer, which simplifies the calculation method of the multi-layer feedforward neural network and improves the modeling efficiency.

[0034] 7. Based on the prediction model established by the present invention, search algorithms such as genetic algorithms and simulated annealing algorithms can be used to search for the optimal or approximately optimal reinforcement scheme, so that while the radiation resistance performance of aerospace electronic devices is significantly improved, the key electrical parameters of the devices can also be fully optimized. The optimized reinforcement scheme can be more effectively directly used to guide the actual preparation of aerospace radiation resistance reinforcement products, which greatly improves the development efficiency of radiation resistance reinforcement products and reduces the development cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 Schematic diagram of the structure of a typical N-channel VDMOS device selected for an embodiment of the present invention;

[0036] Figure 2 Flowchart of the method for establishing a device-level radiation resistance reinforcement prediction model based on orthogonal design and neural network for an embodiment of the present invention;

[0037] Figure 3 Schematic diagram of the structure of a neural network in an embodiment of the present invention; DETAILED DESCRIPTION OF THE EMBODIMENTS

[0038] To make the objectives, advantages and features of the present invention clearer, the following further elaborates in detail a method for establishing a device-level anti-radiation hardening prediction model based on orthogonal design and neural network in combination with the accompanying drawings and specific embodiments. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0039] The method for establishing a device-level anti-radiation hardening prediction model based on orthogonal design and neural network proposed by the present invention includes the following steps:

[0040] Step 1: For the initial device structure to be hardened, establish a TCAD simulation model, and calibrate the TCAD simulation model according to the actual test data of the key electrical parameters of the device to be hardened and the ground radiation test data.

[0041] Step 2: Select a variety of common anti-radiation hardening technologies to conduct preliminary TCAD simulations respectively, and obtain the influence laws of different hardening technologies on the anti-radiation performance and key electrical parameters of the device.

[0042] Step 3: According to the influence laws obtained in Step 2, select m (m≥3) feasible hardening technologies, and divide the parameter variation range of a single hardening technology into n-1 equal parts (for subsequent machine learning mapping), and use the orthogonal design method to establish an orthogonal design table of m factors and n levels (n≥5). Here, the larger the value of n, the more accurate the established prediction model, but at the same time, the amount of TCAD simulation data is also larger;

[0043] Step 4: Conduct TCAD simulations according to the orthogonal design table to obtain discrete data on the changes in the anti-radiation performance and key electrical parameters of the device under the action of a single hardening technology or multiple hardening technologies combined.

[0044] Step 5: Randomly divide the discrete data obtained from the simulation in Step 4 into a training set, a validation set, and a test set;

[0045] The setting ratios of the training set, the validation set, and the test set are: the training set is 65% - 75%, the validation set is 10% - 20%, and the test set is 10% - 20%.

[0046] Step 6: Construct a prediction model for the changes in the anti-radiation performance and key electrical parameters of the device under complex hardening conditions based on a neural network, use the training set data for iterative learning, and at the same time use the validation set data for verification to obtain a data set of the prediction results of the changes in the anti-radiation performance and key electrical parameters of the device.

[0047] The neural network is a multi-layer feedforward neural network, which includes an input layer, an output layer, and at least one hidden layer. The number of neurons in the input layer is the same as the number of input variables of the prediction model, that is, the number of reinforcement parameters is equal to m. There is at least 1 neuron in the hidden layer. When the hidden layer is two layers, each hidden layer includes 8 neurons. The number of neurons in the output layer is the sum of the number of device key electrical parameter indicators and the number of radiation resistance performance indicators. By adjusting the number of hidden layers and the number of neurons in each hidden layer, the accuracy of the prediction model is improved.

[0048] A normalization operation is introduced between the input layer and the hidden layer, and a denormalization operation is introduced after the output layer. The normalization operation simplifies the calculation method of the multi-layer feedforward neural network and can improve the modeling efficiency.

[0049] The Levenberg–Marquardt algorithm is used when learning with the training set data.

[0050] The tansig function is used as the transfer function for the hidden layer, and the pureline function is used as the transfer function for the output layer. The specific formulas are as follows:

[0051]

[0052] f pureline (x) = x

[0053] where: f tansig (x) is the tansig function; f pureline (x) is the pureline function, and x is the function input variable.

[0054] Step 7: Use the test set data to verify whether the error between the prediction result of the prediction model established in Step 6 and the TCAD simulation result in Step 4 is lower than the set threshold; if so, the training is completed and the prediction model is obtained; if not, return to Step 6 and modify the network parameters of the prediction model of the radiation resistance performance and key electrical parameter changes of the device under complex reinforcement conditions based on the neural network until the error between the prediction result and the simulation data is lower than the set threshold, and the prediction model is completed.

[0055] The set threshold can be the mean square error between the prediction result of the prediction model and the TCAD simulation data, and its value is 0.0001.

[0056] Common aerospace electronic devices include VDMOS, power diodes, LDMOS (Laterally-Diffused Metal-Oxide-Semiconductor Field-Effect Transistor), FinFET (Fin Field-Effect Transistor), IGBT (Insulated Gate Bipolar Transistor), etc. Taking the single-event burnout hardening of N-channel VDMOS devices as an example, the specific modeling steps are as follows:

[0057] Select a typical N-channel VDMOS product. The device structure is as Figure 1 shown. Establish a TCAD simulation model according to the device structure, and calibrate the device TCAD simulation model based on the actual test results of its key electrical parameters (including breakdown voltage, threshold voltage, on-resistance) and ground heavy ion test data (the single-event burnout failure threshold voltage under 1044 MeV Ta ion irradiation). The doping concentration of the device epitaxial layer is 2.2e14 cm -3 , and the epitaxial layer thickness is 60 μm.

[0058] Select four hardening techniques for introducing a buffer layer, including optimizing the buffer layer doping concentration and thickness, increasing the body region doping concentration, and increasing the P+ region lateral length, to improve the single-event burnout resistance of the device, and obtain the influence laws of the four hardening techniques on the radiation resistance and key electrical parameters of the device.

[0059] Set four hardening parameters in the TCAD simulation, namely the body region doping concentration denoted as factor 1, the buffer layer thickness denoted as factor 2, the buffer layer doping concentration denoted as factor 3, and the P+ region lateral length denoted as factor 4. Divide the parameter change ranges of each factor into 8 equal parts, and the values of each level corresponding to each factor are shown in Table 1.

[0060] Table 1 Variation ranges of four hardening parameters

[0061]

[0062] Use the orthogonal design method to establish a 4-factor 9-level orthogonal design table, as shown in Table 2.

[0063] Table 2 4-factor 9-level orthogonal design table

[0064]

[0065]

[0066]

[0067] A total of 81 groups of TCAD simulations are required. The reinforcement parameters of the first group of TCAD simulations are as follows: factor 1 takes level 6, factor 2 takes level 2, factor 3 takes level 9, and factor 4 takes level 9. By comparing the values ​​of each level corresponding to each factor in Table 1, it can be seen that the reinforcement parameters in the first group of TCAD simulations are set as follows: the body region doping concentration is 3.1625e18 cm -3 , the buffer layer thickness is 20 μm, and the buffer layer doping concentration is 1e16 cm -3 The lateral length of the P+ region is 25 μm. The reinforcement parameter settings in the TCAD simulation of the other groups are similar.

[0068] Through TCAD simulation, the changes in the radiation resistance of VDMOS devices (single-particle burnout failure threshold voltage under 1044MeVTa ion irradiation) and key electrical parameters of the devices (including breakdown voltage, threshold voltage, and on-resistance) under various reinforcement conditions in the orthogonal design table are obtained.

[0069] Based on the TCAD simulation data, Figure 2 The modeling method shown establishes a prediction model for the radiation resistance performance and key electrical parameter changes of VDMOS devices under complex reinforcement conditions.

[0070] The TCAD simulation data is randomly divided into a training set, a validation set, and a test set. The ratios are usually set as follows: 65% to 75% for the training set, 10% to 20% for the validation set, and 10% to 20% for the test set. In this embodiment, the ratios are set as 70% for the training set, 15% for the validation set, and 15% for the test set.

[0071] A multi-layer feedforward neural network is used for learning and fitting, and the learning algorithm adopts the Levenberg–Marquardt algorithm. The multi-layer feedforward neural network includes an input layer, at least one hidden layer, and an output layer. Generally, the more hidden layers, the higher the fitting accuracy. Each hidden layer includes at least one neuron. The more neurons there are, the higher the accuracy of the prediction model. The accuracy of the prediction model is verified using the test set data. In this embodiment, the threshold value of the prediction model fitting error is set to a mean square error of 0.0001.

[0072] The hidden layer uses the tansig function as the transfer function, and the output layer uses the pureline function as the transfer function. The specific formulas of the hidden layer and output layer functions are as follows:

[0073]

[0074] f pureline (x) = x

[0075] Where: f tansig (x) is the tansig function; fpureline (x) is a pureline function, where x is the input variable of the function.

[0076] In this embodiment, a multi-layer feedforward neural network with 8×8×4 is used to implement a prediction model for the radiation resistance performance and the changes of key electrical parameters of VDMOS devices under complex hardening conditions. The multi-layer feedforward neural network is as Figure 3 shown, including 2 hidden layers, each hidden layer contains 8 neurons, 1 output layer, and the number of neurons in the output layer is the sum of the number of device key electrical parameter indexes (in this embodiment, it is 3, including breakdown voltage, threshold voltage, and on-resistance) and the number of radiation resistance performance indexes (in this embodiment, it is 1, including the single-event burnout failure threshold voltage under 1044 MeV Ta ion irradiation). Since the input layer is composed of hardening parameters such as body doping concentration, buffer layer thickness, buffer layer doping concentration, and P+ region lateral length, and the units of the input variables are different, a normalization operation is introduced between the input layer and the hidden layer. Similarly, a denormalization operation is introduced after the output layer.

[0077] Through multiple learning iterations, a prediction model based on the neural network is finally obtained. In this embodiment, after 1000 learning iterations, a prediction model for the radiation resistance performance and the changes of key electrical parameters of VDMOS devices under complex hardening conditions is obtained.

[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the present invention.

Claims

1. A method for establishing a device-level radiation hardening prediction model based on orthogonal design and neural network, characterized in that, it includes the following steps: Step 1: For the initial device structure to be hardened, establish a TCAD simulation model, and calibrate the TCAD simulation model according to the actual test data and ground radiation test data of the key electrical parameters of the device to be hardened; Step 2: Select at least three radiation hardening technologies to conduct preliminary TCAD simulations respectively, and obtain the influence laws of different hardening technologies on the radiation resistance performance and key electrical parameters of the device; Step 3: According to the influence laws obtained in Step 2, select m feasible hardening technologies, and divide the parameter change range of a single hardening technology into n-1 equal parts, and establish an m-factor n-level orthogonal design table by using the orthogonal design method; where m≥3, n≥5, and both m and n are integers; Step 4: Conduct TCAD simulations according to the orthogonal design table to obtain discrete data on the changes in the radiation resistance performance and key electrical parameters of the device under the action of a single hardening technology or multiple hardening technologies; Step 5: Randomly divide the discrete data obtained from the simulation in Step 4 into a training set, a validation set, and a test set; Step 6: Construct a prediction model for the changes in the radiation resistance performance and key electrical parameters of the device under complex hardening conditions based on a neural network, use the training set data for iterative learning, and at the same time use the validation set data for accuracy verification to obtain the prediction results of the prediction model for the changes in the radiation resistance performance and key electrical parameters of the device to be hardened; Step 7: Use the test set data to verify whether the error between the prediction results of the prediction model established in Step 6 and the TCAD simulation results in Step 4 is lower than the set threshold; if so, the training is completed and the prediction model is obtained; if not, return to Step 6 and modify the network parameters of the prediction model for the changes in the radiation resistance performance and key electrical parameters of the device under complex hardening conditions based on the neural network until the error between the prediction results and the simulation results is lower than the set threshold, and the prediction model is obtained.

2. The method for establishing a device-level radiation hardening prediction model based on orthogonal design and neural network according to claim 1, characterized in that: In Step 6, the neural network is a multi-layer feedforward neural network, including an input layer, an output layer, and at least one hidden layer; the number of neurons in the input layer is the same as the number m of the selected hardening technologies in Step 3; there is at least 1 neuron in the hidden layer; the number of neurons in the output layer is the sum of the number of device key electrical parameter indicators and the number of radiation resistance performance indicators.

3. The method for establishing a device-level radiation hardening prediction model based on orthogonal design and neural network according to claim 2, characterized in that: In Step 6, the neural network further includes a normalization module arranged between the input layer and the hidden layer, and a denormalization module arranged after the output layer.

4. The method for establishing a device-level radiation hardening prediction model based on orthogonal design and neural network according to claim 3, characterized in that: In Step 6, the hidden layer uses the tansig function as the transfer function, and the output layer uses the pureline function as the transfer function, and the specific formulas are as follows: f pureline (x) = x where: f tansig (x) is the tansig function; f pureline (x) is the pureline function, and x is the function input variable.

5. The method for establishing a device-level radiation hardening prediction model based on orthogonal design and neural network according to any one of claims 1-4, characterized in that: In step 6, the Levenberg–Marquardt algorithm is used for iterative learning with the training set data.

6. The method for establishing a device-level radiation hardening prediction model based on orthogonal design and neural network according to claim 5, characterized in that: In step 5, the setting ratios of the training set, validation set and test set are: training set 65% - 75%, validation set 10% - 20%, and test set 10% - 20%.

7. The method for establishing a device-level radiation hardening prediction model based on orthogonal design and neural network according to claim 6, characterized in that: In step 7, the error is the mean square error, and its set threshold is 0.0001.

8. The method for establishing a device-level radiation hardening prediction model based on orthogonal design and neural network according to claim 1, characterized in that: In step 1, the device to be hardened is an N-channel VDMOS device, and the performance to be hardened is the single-event burnout resistance performance; the key electrical parameters include breakdown voltage, threshold voltage, and on-resistance parameters.

9. The method for establishing a device-level radiation hardening prediction model based on orthogonal design and neural network according to claim 8, characterized in that: In step 2, the multiple radiation hardening techniques include 4 hardening techniques of optimizing the doping concentration and thickness of the buffer layer, increasing the doping concentration of the body region, and increasing the lateral length of the P+ region.

Images