Multi-mode space navigation static random access memory irradiation fault diagnosis method and platform

Abstract

The invention provides a multi-mode aerospace static random access memory irradiation fault diagnosis method and platform, and relates to the technical field of anti-radiation integrated circuit design, and the method comprises the following steps: collecting the voltage, current, electromagnetic wave and temperature data of a to-be-tested device through an analog end circuit module; carrying out data preprocessing based on a digital signal recovery fixed-point decimal algorithm and wavelet transform filtering; the method comprises the following steps: establishing a fault diagnosis system, establishing a reverse neural network and a Pytorch end feedforward neural network, training and testing the neural network through a fault data set, adaptively adjusting the learning rate of each parameter in combination with an Adam algorithm, extracting weight and bias parameters of the trained neural network, transmitting the weight and bias parameters to an FPGA end feedforward neural network, and detecting the diagnosis precision. According to actually measured irradiation fault data, a static random access memory is used as a chip to be detected, and current and voltage parameters in two irradiation fault states of a single event effect and a total dose effect are collected and detected.

Description

Technical Field

[0001] The present application relates to the technical field of radiation-resistant integrated circuit design, and more specifically, to a multi-modal aerospace static random access memory radiation fault diagnosis method and platform. Background Art

[0002] As a nanoscale integrated circuit, static random access memory is very sensitive to the radiation environment, and the radiation effect may cause temporary or permanent changes in the function of the integrated circuit in the spacecraft, or even cause the device to fail completely. Among the failures of spacecraft integrated circuits, about 45% of the failures are caused by various radiation failures, which shows that further research on how to improve the radiation resistance of spacecraft integrated circuits is very necessary. Integrated circuit radiation fault diagnosis is the basis for improving the reliability of integrated circuits. In integrated circuits, due to the complex circuit structure and the large number of static random access memories, the radiation faults are highly hidden. Therefore, improving the accuracy of radiation fault diagnosis will help to ensure the performance and stability of the circuit, improve the quality and reliability of spacecraft, and reduce the cost and time of device maintenance. In the prior art, the solutions to improve radiation fault diagnosis are: The Chinese invention patent with publication number CN109003645A discloses an on-orbit single-particle effect verification system for an electronic system, including a space environment detector and an FPGA controller, wherein the FPGA controller is connected to two or more electronic systems under test and the space environment detector respectively; the two or more electronic systems under test have a self-detection function, and through self-detection, they count data errors caused by single-particle effects, and send the counted data errors as monitoring data to the FPGA controller; at the same time, the two or more electronic systems under test send data errors caused by single-particle effects in the process of data exchange as monitoring data to the FPGA controller; the space environment detector identifies valid particle events by detecting the space environment, and sends them to the FPGA controller as monitoring data; by comparing the monitoring data sent by the electronic system under test and the space environment detector, it is confirmed whether a single-particle effect occurred at the time when the data error occurred to perform single-particle effect verification.

[0003] However, the solutions adopted by the existing technology are mainly real-time verification and detection, which rely on equipment such as LET probes and proton probes to confirm whether a single particle effect occurred at the moment when the data error occurred to perform single particle effect verification, and the acquisition and analysis of data rely on special equipment. Summary of the invention

[0004] To solve the above problems, the technical solution adopted in this application is a multi-modal aerospace static random access memory irradiation fault diagnosis method, comprising the following steps: Data acquisition: Collect voltage, current, electromagnetic wave and temperature data of the device under test through the analog circuit module; Data preprocessing: preprocess the data collected by the analog circuit module based on the digital signal recovery fixed-point decimal algorithm and wavelet transform filtering algorithm; Algorithm training: Build a reverse neural network, a Pytorch-side feedforward neural network, and an FPGA-side feedforward neural network with the same number of layers, neurons, and forward propagation activation functions. Train the reverse neural network after preprocessing the fault data set. Use the Adam algorithm to adaptively adjust the learning rate of each parameter. Extract the weights and bias parameters of the trained neural network and pass them to the Pytorch-side feedforward neural network for testing. After the test accuracy meets expectations, pass the weights and bias parameters of the neural network to the FPGA-side feedforward neural network. Diagnostic accuracy test: Under the condition of fault injection, after data acquisition and data preprocessing, the fault diagnosis system including the FPGA-side feedforward neural network and the data preprocessing module conducts a radiation simulation fault test to verify the feasibility of the FPGA-side feedforward neural network radiation fault diagnosis and obtain the diagnostic accuracy.

[0005] Optionally, the wavelet transform filtering includes using Daubechies wavelet as a wavelet basis to perform multi-scale decomposition of the signal processed by the digital signal recovery fixed-point decimal algorithm into wavelet coefficients through continuous wavelet transform, setting a hard threshold to remove noise in the wavelet coefficients, and reconstructing the signal using the denoised wavelet coefficients.

[0006] Optionally, the learning rate of each parameter is adaptively adjusted in combination with the Adam algorithm, including calculating the first-order moment and the second-order moment of the gradient of the loss function relative to the weight and bias parameters, performing bias correction on the first-order moment and the second-order moment to reduce the deviation, and finally using the bias-corrected first-order moment and the second-order moment to update the weight and bias parameters.

[0007] Optionally, an SPI receiving protocol is used as a transmission protocol between the analog end circuit module and the fault diagnosis system, and the SPI receiving protocol, digital signal recovery fixed-point decimal algorithm, wavelet transform filtering and FPGA-end feedforward neural network are designed based on the Verilog hardware description language.

[0008] Optionally, the hardware of the fault diagnosis system is an FPGA chip, the analog end circuit module is a signal acquisition component, and the device to be tested is an irradiation fault static random access memory.

[0009] Optionally, the FPGA-side feedforward neural network is a single hidden layer feedforward neural network with a 4-4-3 structure, and the neural network inputs the current, voltage, temperature and electromagnetic wave parameters of the device under test after data preprocessing; and outputs single particle effect, total dose effect or normal.

[0010] Optionally, the feedforward neural network on the FPGA side classifies the output values ​​of the output layer nodes, compares the output values, and converts the outputs into one-hot codes; the activation function of the neurons of the feedforward neural network on the FPGA side adopts a ReLU activation function.

[0011] Optionally, it also includes designing, board making, welding and debugging a multi-modal aerospace static random access memory radiation fault diagnosis platform composed of a module under test, an analog end circuit module and an FPGA diagnostic module; debugging includes current detection capability debugging, voltage detection resolution debugging, temperature detection capability debugging, electromagnetic wave detection capability debugging, signal detection broadband debugging and FPGA diagnostic module debugging.

[0012] Optionally, during testing, the program is burned into the FPGA chip through the UART serial port.

[0013] The present application also provides a multi-modal aerospace static random access memory radiation fault diagnosis platform, which is suitable for implementing any of the aforementioned multi-modal aerospace static random access memory radiation fault diagnosis methods.

[0014] The beneficial effects of a multi-modal aerospace static random access memory radiation fault diagnosis method and platform provided by the present application are: 1. Design an aerospace static random access memory irradiation fault diagnosis platform consisting of a circuit module to be tested, an analog end circuit module and an FPGA diagnosis module. Under the condition of fault injection, perform an irradiation simulation fault test on the aerospace static random access memory irradiation fault diagnosis platform. The tested part of the static random access memory irradiation fault diagnosis platform is based on the measured irradiation fault data. The static random access memory is used as the chip to be tested. The current and voltage parameters of two types of irradiation fault states, single particle effect and total dose effect, are collected and detected. Compared with using equipment equipped with LET probes and proton probes, the test data cost is greatly reduced.

[0015] 2. The fault diagnosis algorithm is implemented using a neural network algorithm. Compared with the traditional aerospace static random access memory radiation fault diagnosis method, it has the advantages of low cost, low power consumption, high precision, and fast detection speed. At the same time, the overall platform has a simple structure, which effectively reduces the burden of the detection system on the spacecraft during the space mission. The neural network theory is applied to failure analysis in the field of integrated circuit reliability, and more types of radiation faults can be detected, with a shorter detection cycle and higher detection accuracy. After training, it can independently detect and identify radiation faults based on the collected information.

[0016] 3. Wavelet transform filtering, which provides localized information of signals in time and frequency domains and is suitable for analyzing non-stationary signals, is used to pre-process the voltage, current, temperature, and electromagnetic wave data. The Adam algorithm is used to adaptively adjust the learning rate of each parameter to perform multi-modal aerospace static random access memory irradiation fault diagnosis, thereby improving the accuracy and sensitivity of fault identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art.

[0018] Figure 1 It is the algorithm module division and data flow diagram provided in the embodiment of the present application; Figure 2 It is a block diagram of the fault diagnosis algorithm design provided in the embodiment of the present application. DETAILED DESCRIPTION

[0019] In order to make the technical problems, technical solutions and beneficial effects to be solved by 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.

[0020] Example 1 The aerospace static random access memory irradiation fault diagnosis platform includes a tested circuit module, an analog end circuit module and an SRAM (static random access memory) type FPGA (field-programmable gate array) diagnosis module.

[0021] The analog circuit module mainly includes signal acquisition components to complete the electrical signal acquisition of the faulty device and monitor the current, voltage, temperature and electromagnetic waves of the static random access memory of the module under test.

[0022] The module under test is a static random access memory, and the radiation failure types include the two most common types of radiation failures of static random access memory under radiation: single particle effect and total dose effect. The power supply in the module under test is interconnected with the external input channel reserved by the analog end circuit module to monitor the current and voltage changes of the module under test in real time.

[0023] like Figure 1 As shown, Figure 1The algorithm module division and data flow diagram are as follows: Control module: After the fault diagnosis system is completed, it starts the next diagnosis cycle, obtains the operation completion signal from the feedforward neural network module, and sends a restart signal to the data preprocessing module; Data preprocessing module: Receives ADC acquisition data, performs data preprocessing, and transmits the processed data to the neural network algorithm module; Feedforward neural network module: The input signal is the output data of the data preprocessing module, performs fault diagnosis, and outputs the fault diagnosis result to the diagnosis result module; Diagnosis result module: The input is the diagnosis output of the feedforward neural network module, and the diagnosis result is saved; Through the SPI interface, the ADC transmission data is received and preprocessed, and then transmitted to the neural network algorithm module for fault diagnosis. The diagnosis results are saved, and a restart signal is given for the next diagnosis. The input signals include SCK (clock signal), SDI (input data), SDO (output data), and the output signals include the diagnosis completion signal DONE and the output diagnosis result OUT.

[0024] like Figure 2 As shown in the figure, the aerospace static random access memory irradiation fault diagnosis method is based on neural network implementation. The algorithm module includes SPI receiving protocol algorithm, data preprocessing algorithm and neural network algorithm. The design steps of the multi-modal aerospace element static random access memory irradiation fault diagnosis method are as follows: S1: Collect the voltage, current, electromagnetic wave and temperature data of the device under test through the analog circuit module, determine the transmission protocol between the ADC acquisition signal in the acquisition module and the fault diagnosis system, and design the SPI receiving protocol using the Verilog language; The system under test includes a device under test and an analog terminal circuit module for collecting voltage, current, electromagnetic wave and temperature data of the device under test; Step S1 is specifically as follows: adopting unidirectional transmission, the master device is ADC, receiving the voltage signal of the thermocouple, the induced current of the near-field probe and the current and voltage of the device under test, and adopting 1SPI protocol bus. Mainly including SCK (clock signal), SDI (input data), SDO (output data). The SPI bus interface works in a master / slave communication mode, and the two devices that communicate with each other are one as a master device and one as a slave device. The clock signal line SCK is generated and controlled by the master device, and the slave device receives the clock and sends and receives data. When the device is in master-slave SPI protocol communication, the data line connecting the data output terminal SDO of the master device and the data input terminal SDI of the slave device is set to MOSI (master device output slave device input), and the data line connecting the data input terminal SDI of the master device and the data output terminal SDO of the slave device is set to MISO (master device input slave device output).

[0025] The acquisition method is: collect the 4 pins of the chip under test through ADC, collect each data 10 times continuously, and collect the next data when the acquisition times reach 10 times. After the data is collected, stop receiving data and proceed to the next step of calculation.

[0026] The input signals of the SPI receiving algorithm are SCK (clock signal), SDI (input data), SDO (output data), the frequency is 1MHz, and the FPGA operating frequency is 50MHz.

[0027] Since the change of electromagnetic emission characteristics is not necessarily completely caused by the change of voltage and current, there are cases where the electromagnetic emission characteristics change but the voltage and current do not change significantly. For example, electromagnetic interference can affect the electromagnetic emission characteristics through common impedance coupling, capacitive coupling and inductive coupling, as well as the harmonic current influence that may be generated by nonlinear loads. At the same time, there is a correlation between the temperature of integrated circuits and single particle effects and total dose effect radiation failures. For example, under high temperature and high voltage, the single particle effect of the device is more likely to be triggered. The total dose effect will affect the oxide layer and silicon interface of the device, causing the thermal dissipation of the device to change, thereby affecting the operating temperature. Therefore, this application adopts multimodal information analysis to improve the radiation fault diagnosis capability.

[0028] S2: Preprocess the data collected by the analog circuit module based on the digital signal recovery fixed-point decimal algorithm and wavelet transform filter, determine the data preprocessing process of collecting the voltage, current, electromagnetic wave and temperature data parameters of the device under test, and design the digital signal recovery fixed-point decimal algorithm and wavelet transform filter using Verilog language; Step S2 is specifically as follows: the digital signal recovery fixed-point decimal algorithm is based on the conversion principle of the successive approximation analog / digital converter, and the recovery algorithm is reversed by reverse thinking to obtain a universal binary number representing the true value of the characteristic parameter. The signal acquisition component uses a 16-bit SAR ADC, and the conversion process of the analog signal to the 16-bit digital signal is performed bit by bit from the highest bit to the lowest bit, and one clock pulse converts one bit. Assume that the 16-bit digital signal outputs are s16-s1 respectively, and the initial value of the input voltage Vin is set to Vin- (minimum range). The first clock pulse of the conversion process compares the size of Vin and Vin- + Vref / 2. If Vin is greater than the latter, s16 is high, otherwise s16 is low; the second clock pulse of the conversion process compares the size of Vin and Vin- + (s16 ×Vref) / 2+ (s15×Vref) / 4. If Vin is greater than the latter, s15 is high, otherwise s15 is low. Similarly, the level values ​​of s14-s1 are output in sequence. The maximum value record compares the size of the collected signals, stores the larger data, and then compares the larger data with the next collected data to obtain the maximum value of the collected signal. The average value is obtained by shifting. Since one channel of data is collected 10 times, the average value is obtained by summing the data and then shifting.

[0029] The input of the data preprocessing module is the SPI received data ADC received data input adc_data_i. After data preprocessing, the output is data_out_1~data_out_4, which serves as the input data of the feedforward neural network module.

[0030] Using Daubechies wavelet as the wavelet basis, through continuous wavelet transform, the signal processed by the digital signal recovery fixed-point decimal algorithm is decomposed into wavelet coefficients at multiple scales, and the hard threshold is set to remove the noise in the wavelet coefficients, and the denoised wavelet coefficients are used to reconstruct the signal. Daubechies wavelet has a high degree of time localization ability and can detect local characteristics in the signal, especially transient characteristics or unstable signals in temperature, current, voltage and electromagnetic wave data. In order to cope with the simplified fault diagnosis module, continuous wavelet transform is adopted in the case of less data source to enhance the sensitivity to changes in time and identify small changes in temperature and electromagnetic wave detection.

[0031] The advantages of wavelet transform filtering in processing voltage, current, temperature, and electromagnetic emission characteristic signals are mainly reflected in its excellent localization characteristics and its ability to provide signal information in both the time domain and the frequency domain. This means that wavelet transform can effectively analyze the local characteristics of the signal, such as mutation points and singular points, and can capture the characteristics of the signal at different frequency levels through multi-scale analysis, which has obvious advantages in identifying and processing electromagnetic emission characteristics. In processing voltage, current, temperature, and electromagnetic emission characteristic signals, it provides higher flexibility, local feature representation capabilities, multi-scale analysis capabilities, and adaptability to mutation and singular signals.

[0032] S3: Build a fault diagnosis system. The fault diagnosis system includes a data preprocessing module and an FPGA-side feedforward neural network. The feedforward neural network is designed using the Verilog hardware description language.

[0033] Step S3 is specifically as follows: Verilog is used to design a double hidden layer feedforward neural network with a 4-4-3 structure. The neural network inputs IN1-IN4 are voltage, current, temperature, and electromagnetic wave parameters after data preprocessing; the outputs are OUT1-OUT3, which are single particle effect, total dose effect, and normal, respectively; the output values ​​of the output layer nodes are classified, the output values ​​are compared, the largest output value becomes 1, and the rest become 0, and the output is converted into a unique hot code. The activation function of the feedforward neural network neuron adopts the ReLU activation function, and the calculation form of the ReLU activation function is f(x) = max(0,x). Since it only needs to compare whether the input is greater than zero, its delay is very small, which can effectively improve the processing speed, and does not require multiplication, division, or other complex operations, so a lot of hardware resources and power consumption are saved in implementation. The module receives the 4-way data data_out_1-data_out_4 processed by the data preprocessing module, and OUT_1-OUT_3 are the diagnostic output results.

[0034] S4: Build a reverse neural network in Pytorch with the same number of layers, neurons, and forward propagation activation function as the FPGA-side feedforward neural network in step S3. Combined with the data preprocessing module, build a neural network training platform, train and test the neural network using the fault data set, and adaptively adjust the learning rate of each parameter using the Adam algorithm. Extract the weights and bias parameters of the trained neural network and pass them to the FPGA-side feedforward neural network.

[0035] Step S4 is specifically as follows: according to the structural framework of the feedforward neural network, determine the number of neural network layers, the number of nodes in each layer, the activation function of each layer of neurons, the initial learning rate and the initial weight bias parameters; according to the measured static random access memory irradiation failure data, including single particle effect, total dose effect and normal, prepare the data as training sets and test sets, and use torch.utils.data.DataLoader for batch loading; use PyTorch's nn.Module class to define the BP neural network structure; use the gradient descent algorithm for training; the neural network training process includes determining the number of training rounds, outputting predicted values, and calculating the error between predicted values ​​and expected values, and updating weights and bias parameters according to the error; if the prediction results of the mature neural network after training meet the accuracy of the expected results, the weights and bias parameters of the mature neural network are extracted and transmitted to the feedforward neural network built on the FPGA through a script to achieve fast and real-time fault diagnosis, which is particularly important for application scenarios that require fast response.

[0036] Combined with the Adam algorithm, it is specifically: 1. Network initialization Parameter initialization: Randomly initialize the weights and biases of the neural network.

[0037] Adam parameter initialization: Initialize the first-order moment estimate and second-order moment estimate of the Adam algorithm to 0, set the learning rate, first-order moment decay rate, second-order moment decay rate and a small constant.

[0038] 2. Training process Forward propagation: The pre-processed voltage, current, electromagnetic wave, and temperature input data are passed through the input layer to the first hidden layer. Each hidden layer neuron calculates the sum of the weighted input and bias, applies an activation function, and then passes the result to the next layer. This process continues until the data reaches the output layer, which generates a prediction result.

[0039] Calculate the loss: The loss function value is calculated based on the predicted results and the true label.

[0040] Backward Propagation: The gradients of the loss function with respect to the output layer parameters are calculated and then these gradients are back-propagated to the input layer.

[0041] For each layer, the gradient of the loss function with respect to the weights and biases is calculated.

[0042] Adam's update: For each parameter, the Adam algorithm is used to update the first-order moment estimate and the second-order moment estimate.

[0043] First-order moment (i.e., mean of the gradient): mt =β1·m t-1 +(1-β1)·g t , where m t is the first-order moment estimate of the time step, g t is the gradient of the time step, β1 is the decay rate of the first-order moment estimate, m t-1 is the first-order moment estimate at time step t-1, that is, the first-order moment estimate of the previous iteration.

[0044] Second moment (i.e. the uncentered variance of the gradient): v t =β2·v t-1 +(1-β2)·g t 2 , where v t is the second-order moment estimate of the time step, β2 is the decay rate of the second-order moment estimate, and v t-1 is the second-order moment estimate at time step t-1, that is, the second-order moment estimate of the previous iteration.

[0045] Bias-corrected first and second order moments.

[0046] Bias correction of the first-order moment: = ; Bias correction of the second-order moment: = ; In the formula, is the first-order moment estimate after bias correction at time step t, that is, the exponentially weighted moving average of the corrected gradient, is the bias-corrected second-order moment estimate at time step t, i.e., the exponentially weighted moving average of the corrected squared gradient.

[0047] Use the first-order moment and second-order moment after bias correction to update each parameter, and adaptively adjust the learning rate of each parameter. In the formula, θ is the parameter, representing the weight and bias, η is the learning rate, Represents a small constant to prevent the denominator from being zero: .

[0048] 3. Parameter Update Weight and bias updates: The update values ​​calculated by the Adam algorithm are used to update the weights and biases of the neural network.

[0049] 4. Iterative training Repeat steps 3 and 4: Repeat the forward propagation, loss calculation, backpropagation and parameter update for each batch of data in the training set.

[0050] S5: Under the condition of completing the design of feedforward neural network, design, make boards, weld and debug the aerospace static random access memory irradiation fault diagnosis platform composed of the circuit module to be tested, the analog end circuit module and the FPGA diagnosis module; The design, board making and welding of the system in step S5 are slightly modified from the prototype verification board block diagram; the debugging includes current detection capability debugging, voltage detection resolution debugging, signal detection bandwidth debugging and SRAM type FPGA diagnostic module debugging.

[0051] The current detection capability debugging is as follows: the bias voltage of SRAM FPGA is 1.0V, 1.5V, 1.8V, and 3.3V. The current detection capability required in the index is: 500uA-1A, the sampling resistor is set to 100Ω, 10Ω, 1Ω, and 0.1Ω, the 0.5mA-1mA sampling resistor is 100Ω, the 1mA-10mA sampling resistor is 10Ω, the 10mA-100mA sampling resistor is 1Ω, and the 0.1A-1A sampling resistor is 0.1Ω. Therefore, the typical voltage difference value of the analog current value is 0.01V, 0.05V, and 0.1V.

[0052] The specific debugging of voltage detection resolution is as follows: the prototype verification board continuously changes the input voltage difference, the input typical voltage difference is from 0.05V-0.1V, stepping 0.01V, FPGA samples 1000 points and reads the results, and FPGA reads the processed results. The voltage detection resolution debugging ensures that the voltage value (V) must meet the two decimal places (0.01V) to remain stable to meet the index requirements.

[0053] The temperature detection capability debugging and electromagnetic wave detection capability debugging are carried out simultaneously with the current detection capability debugging and voltage detection resolution debugging. The temperature and electromagnetic waves are detected synchronously under the same voltage and current conditions. The accuracy within the detection range is required to fully meet the possible detection range requirements of voltage and current.

[0054] The signal detection broadband debugging is as follows: the ADC test range is 0-1.125V. During the test, a sine wave with a frequency of 200KHz and a signal peak-to-peak value Vpp=0-1.125V is input at the SMA port. The FPGA reads the test results and compares them with the signal peak-to-peak value Vpp=0-0.795V with a -3db bandwidth to verify whether the signal detection bandwidth reaches 200KHz.

[0055] S6: Under the condition of fault injection, the aerospace static random access memory irradiation fault diagnosis platform composed of the circuit module under test, the analog end circuit module and the SRAM type FPGA diagnosis module is jointly tested to conduct irradiation simulation fault test, verify the feasibility of the irradiation fault diagnosis system and obtain the diagnostic accuracy of the failure diagnosis system.

[0056] Step S6 is specifically as follows: The board 1 to be tested continuously runs the specified program in a normal interference-free environment; periodically samples and restores the characteristic parameter values ​​at 100 sampling moments, a total of 100 samples, and records the label 1 outside the system as a normal state test sample.

[0057] Boards 2 and 3 under test continuously run the specified program in the radiation simulation environment to simulate the single particle effect and the total dose effect; periodically sample and restore the voltage, current, electromagnetic wave and temperature data values, run different programs, and each board under test periodically samples and restores the characteristic parameter values ​​of 100 sampling moments, for a total of 300 samples, which are recorded outside the system as 2 and 3.

[0058] The irradiation failure diagnosis of the board under test requires the collection of voltage, current, electromagnetic wave and temperature data parameters of the SRAM-type FPGA under test as characteristic parameters, and the periodic collection of each characteristic parameter 10 times as a sample. After the fault diagnosis system is turned on, the FPGA receives the ADC transmission data, and the system performs periodic irradiation fault detection on the SRAM-type FPGA under test. A total of 600 failure samples were detected, of which 597 were successfully detected, with an accuracy rate of 99.5%.

[0059] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A multi-modal aerospace static random access memory irradiation fault diagnosis method, characterized in that: The following steps are involved: Data acquisition: Collect voltage, current, electromagnetic wave and temperature data of the device under test through the analog circuit module; Data preprocessing: preprocessing the data collected by the analog end circuit module based on a digital signal recovery fixed-point decimal algorithm and a wavelet transform filtering algorithm; Algorithm training: Build a reverse neural network, a Pytorch-side feedforward neural network, and an FPGA-side feedforward neural network with the same number of layers, neurons, and forward propagation activation functions. Train the reverse neural network after data preprocessing of the fault data set. Adaptively adjust the learning rate of each parameter in combination with the Adam algorithm. Extract the weights and bias parameters of the trained neural network and pass them to the Pytorch-side feedforward neural network for testing. After the test accuracy meets expectations, pass the weights and bias parameters of the neural network to the FPGA-side feedforward neural network. Diagnostic accuracy test: Under the condition of fault injection, after data acquisition and data preprocessing, the fault diagnosis system including the FPGA-side feedforward neural network and the data preprocessing module conducts a radiation simulation fault test to verify the feasibility of the FPGA-side feedforward neural network radiation fault diagnosis and obtain the diagnostic accuracy.

2. The multi-modal aerospace static random access memory radiation fault diagnosis method according to claim 1, characterized in that: The wavelet transform filtering includes using Daubechies wavelet as wavelet basis through continuous wavelet transform, performing multi-scale decomposition of the signal processed by the digital signal recovery fixed-point decimal algorithm into wavelet coefficients, setting a hard threshold to remove noise in the wavelet coefficients, and reconstructing the signal of the denoised wavelet coefficients.

3. The multi-modal aerospace static random access memory radiation fault diagnosis method according to claim 1, characterized in that: The adaptive adjustment of the learning rate of each parameter in combination with the Adam algorithm includes calculating the first-order moment and the second-order moment of the gradient of the loss function relative to the weight and bias parameters, performing bias correction on the first-order moment and the second-order moment to reduce the deviation, and finally using the bias-corrected first-order moment and the second-order moment to update the weight and bias parameters.

4. The multi-modal aerospace static random access memory radiation fault diagnosis method according to claim 2, characterized in that: The SPI receiving protocol is used as the transmission protocol between the analog end circuit module and the fault diagnosis platform. The SPI receiving protocol, digital signal recovery fixed-point decimal algorithm, wavelet transform filtering and FPGA end feedforward neural network are designed based on Verilog hardware description language.

5. The multi-modal aerospace static random access memory radiation fault diagnosis method according to claim 3, characterized in that: The hardware of the fault diagnosis system is an FPGA chip, the analog end circuit module is a signal acquisition component, and the device to be tested is an irradiation fault static random access memory.

6. The multi-modal aerospace static random access memory radiation fault diagnosis method according to claim 1, characterized in that: The FPGA-side feedforward neural network is a single hidden layer feedforward neural network with a 4-4-3 structure. The neural network inputs the current, voltage, temperature and electromagnetic wave parameters of the device under test after data preprocessing; and outputs single particle effect, total dose effect or normal.

7. The multi-modal aerospace static random access memory radiation fault diagnosis method according to claim 5, characterized in that: The FPGA-side feedforward neural network classifies the output values ​​of the output layer nodes, compares the output values, and converts the outputs into one-hot codes; the activation function of the neurons of the FPGA-side feedforward neural network adopts the ReLU activation function.

8. The multi-modal aerospace static random access memory radiation fault diagnosis method according to claim 1, characterized in that: It also includes the design, board making, welding and debugging of a multi-modal aerospace static random access memory radiation fault diagnosis platform composed of a module under test, an analog end circuit module and an FPGA diagnostic module; the debugging includes current detection capability debugging, voltage detection resolution debugging, temperature detection capability debugging, electromagnetic wave detection capability debugging, signal detection broadband debugging and FPGA diagnostic module debugging.

9. The multi-modal aerospace static random access memory radiation fault diagnosis method according to claim 8, characterized in that: During testing, the program is burned into the FPGA chip through the UART serial port.

10. A multi-modal aerospace static random access memory radiation fault diagnosis platform, characterized by: The method is suitable for implementing the multi-modal aerospace static random access memory radiation fault diagnosis method as described in any one of claims 1 to 9.

Images