A reliability prediction system for electronic components

Abstract

A reliability prediction system for electronic components belongs to the technical field of electronic component life prediction. To solve the problems of low efficiency, poor flexibility, and difficult calculation of complex models in existing electronic component reliability prediction methods. It includes: a data processing module for uploading test data on a Web interface; a component model selection module that selects different component degradation models according to the reliability prediction target and constructs them based on the test data; a reliability model construction module that uses the sparrow optimization algorithm to optimize the parameters of the component degradation model, uses an interactive setting algorithm to construct an iterative search minimization model for the number of search individuals and the number of search iterations, and solves the parameter fitting by the difference between the iterative search minimization model and the actual data to construct a reliability model; a reliability prediction module that uses the lognormal cumulative distribution function for life cycle prediction to obtain the reliability prediction result of the component. It is used for the reliability prediction of electronic components.

Description

Technical Field

[0001] The present invention relates to a reliability prediction system for electronic components and belongs to the technical field of electronic component life prediction. Background Art

[0002] With the rapid development of electronic component technology, the evaluation of its performance and reliability has become increasingly important. Especially in the aerospace field, aerospace equipment such as satellites usually operates in extreme environments, such as high radiation, drastic temperature changes, etc. These harsh conditions pose extremely high requirements on the reliability of electronic components, and the performance and reliability of electronic components are crucial. Therefore, accurately predicting and analyzing the performance degradation and failure rate of key electronic components of these devices has become an important task in the aerospace field.

[0003] The traditional reliability prediction methods for electronic components mainly have the following problems:

[0004] 1. Reliability prediction methods often rely on cumbersome experimental processes and time-consuming data analysis. These methods usually include manual data entry, statistical analysis using statistical methods, and simulation using engineering software, etc., which result in low efficiency in the rapid iteration product development cycle and large data scale;

[0005] 2. Aerospace equipment often faces the influence of variable environmental factors. Traditional methods lack sufficient flexibility and adaptability in simulating and analyzing these influences, and have problems such as insufficient flexibility, low reliability calculation integration, and complex operation;

[0006] 3. Traditional reliability model parameter fitting methods such as the gradient descent method and the least squares method are difficult to adapt to complex models and handle non-linear problems.

[0007] Considering the above problems, a method capable of quickly and accurately predicting the reliability of electronic components is of great significance for improving data processing efficiency, achieving adaptive high-precision parameter fitting, conducting reliability modeling and prediction, contributing to the performance and reliability analysis of electronic components in the aerospace field, and shortening the device development cycle, improving device quality, and reducing costs. Summary of the Invention

[0008] The purpose of the present invention is to solve the problems of low efficiency, poor flexibility, and difficult calculation of complex models existing in the existing reliability prediction methods for electronic components, and provide a reliability prediction system for electronic components.

[0009] A reliability prediction system for electronic components according to the present invention includes a data processing module, a component model selection module, a reliability model construction module, and a reliability prediction module;

[0010] The data processing module is used to realize the upload of test data on the Web interface;

[0011] The component model selection module is used to select different component degradation models according to the reliability prediction target and construct them based on the test data;

[0012] The reliability model construction module is used to optimize the parameters of the component degradation model by using the sparrow optimization algorithm, construct an iterative search minimization model for the number of search individuals and the number of search iterations by using the interactive setting algorithm, and solve the parameter fitting by the difference between the iterative search minimization model and the actual data, so as to construct a reliability model;

[0013] The reliability prediction module is used to calculate the mean and variance of the normal distribution of the initial data based on the reliability model, calculate the logarithmic normal cumulative distribution function, and use the logarithmic normal cumulative distribution function to predict the life cycle to obtain the reliability prediction result of the component.

[0014] Preferably, the test data uploaded by the data processing module on the Web interface is in Excel format;

[0015] The upload of the test data by the data processing module includes: data input, data reading, data processing and data online editing;

[0016] The data processing includes: data cleaning and formatting;

[0017] The data online editing includes: adjusting data on the Web interface, removing outliers by the quartile method, and reducing data noise by the exponentially weighted moving average method;

[0018] The data reading and data processing are implemented by using the Pandas library; the online editing of the data is implemented by using the Streamlit-aggrid library.

[0019] Preferably, the component degradation models include Schottky barrier diode Au-Si contact degradation model, Schottky barrier diode metallization electromigration model, PN junction rectifier diode PN junction characteristic degradation model, PN junction rectifier diode hot carrier injection model, bipolar transistor hot carrier injection model and bipolar transistor PN junction characteristic degradation model.

[0020] Preferably, the specific method for constructing the PN junction rectifier diode PN junction characteristic degradation model includes:

[0021] Obtain the test data of the electronic component under four groups of stress conditions; upload the test data on the Web interface and perform data cleaning and formatting;

[0022] Construct the PN junction rectifier diode PN junction characteristic degradation model as:

[0023]

[0024] Among them, I R represents the reverse leakage current of the PN junction rectifier diode measured at different times t, and I R0 represents the initial value of the reverse leakage current, A represents the degradation scale factor, which is used to quantify the influence degree of stress on degradation, and V R represents the reverse voltage applied to the diode, m represents the voltage exponent, which is used to describe the influence degree of reverse voltage on the degradation process, and E a is the activation energy, which represents the energy barrier that needs to be overcome during the degradation process, T is the absolute temperature, with the unit of K, p is the time exponent, which describes the influence of time on the degradation process, and k is the Boltzmann constant, which is used to convert energy into temperature;

[0025] Among them, the parameters A, m, and E a and p are the parameters to be solved;

[0026] The sparrow optimization algorithm is used to optimize the parameters of the component degradation model, and its specific process includes:

[0027] Define the objective function as:

[0028]

[0029] Among them, MSE m represents the mean square error under condition m, m are four different groups of stress conditions, n represents the number of data points, and I R (i) represents the actual reverse leakage current value of the i-th data point, and I' R (i) represents the predicted reverse leakage current value of the i-th data point calculated using the current parameter set and the model formula;

[0030] Define the fitness function as:

[0031] MSE = MSE1 + MSE2 + MSE3 + MSE4

[0032] Among them, MSE1, MSE2, MSE3, and MSE4 respectively represent the mean square errors under four groups of stress conditions;

[0033] Substitute the data into the sparrow algorithm, and randomly initialize the positions of each individual in the sparrow population. Since the individual is a sparrow, for each sparrow i:

[0034] X′ i =(A i , m i , E ai , p i), i = 1, 2, 3, ..., N

[0035] Among them, X' i represents the parameter vector of the position of each sparrow i, representing a set of potential solutions, namely the parameters A, m, E a and p, N represents the size of the sparrow population;

[0036] The update of the sparrow position depends on the behavioral strategy. Update the discoverer i. If there is no predator, that is, r2 < 0.8, then:

[0037]

[0038] Among them, I t represents the current iteration number, I tmax represents the maximum number of iterations, determining the number of iteration rounds of the algorithm operation. r1 and r2 represent random numbers in [0, 1], introducing random behavior to simulate the natural behavior of sparrows. exp represents the exponential function to simulate the change of the sparrow flight trajectory;

[0039] If there is a predator, then:

[0040] X' i = X' i + Q × 1

[0041] Among them, Q represents a random number of the normal distribution, and 1 represents a vector of all 1s;

[0042] Update the position of the followers. For each follower i, if i is greater than half of the population, then:

[0043]

[0044] Among them, X worst represents the position of the sparrow with the worst fitness in the current population;

[0045] Otherwise:

[0046]

[0047] Among them, X best represents the position of the previous optimal sparrow, that is, the position of the sparrow with the best performance in the population, and A represents a random matrix;

[0048] Update the position of the early warning sparrows. While maintaining vigilance against the current best position, by introducing randomness, the sparrows can explore new possible areas, thereby increasing the chance of finding better solutions. For each early warning sparrow i, if its fitness is greater than the global best, then:

[0049] X' i = X gbest + Q × |X' i - Xgbest |

[0050] Among them, X gbest represents the best position found among all sparrows so far;

[0051] Otherwise:

[0052]

[0053] Among them, r3 represents a random number in [0, 1], and MSE worst represents the worst fitness in the population, and ε is a very small positive number used to prevent division-by-zero errors;

[0054] Repeat the above steps until the maximum number of iterations is reached, and the maximum number of iterations is set manually interactively.

[0055] Preferably, the test data is the forward voltage drop degradation data from 0 to 1000 h, with each time interval point being 100 h, and 10 samples are tested under each stress condition.

[0056] Preferably, the four stress conditions are respectively:

[0057] Ambient temperature 398 K, reverse bias voltage 800 V;

[0058] Ambient temperature 398 K, reverse bias voltage 1000 V;

[0059] Ambient temperature 398 K, reverse bias voltage 1300 V;

[0060] Ambient temperature 423 K, reverse bias voltage 1300 V.

[0061] Preferably, the specific methods of removing outliers using the quartile method and data denoising using the exponentially weighted moving average method include:

[0062] IQR = Q3 - Q1

[0063] B L = Q1 - 1.5 × IQR

[0064] B U = Q3 + 1.5 × IQR

[0065] B L <X i <B U

[0066] X t = α·X i +(1 - α)·X t-1

[0067] Among them, IQR represents the interquartile range, Q3 represents the third quartile, that is, the 75% quantile, Q1 represents the first quartile, that is, the 25% quantile, and B L represents the lower boundary of the outlier, and B U represents the upper boundary of the outlier, and X i represents the data after removing the outliers, α represents the smoothing coefficient, and X t represents the smoothed value at time point t, and X t-1 represents the smoothed value at time t - 1.

[0068] Preferably, the maximum number of iterations is manually set to 100 interactively, and thus the values of parameters A, m, and E a are as follows:

[0069] A = -24, m = 4, and E a = 0.76, p = 0.2;

[0070] The PN - junction characteristic degradation model of the PN - junction rectifier diode obtained is:

[0071]

[0072] Preferably, the specific method for the reliability prediction module to obtain the reliability prediction result of the component includes:

[0073] Define the failure threshold of the electronic component, calculate the initial data, that is, at t = 0, the mean and standard deviation of the normal distribution of all data, and use the cumulative distribution function of the log - normal distribution to predict the reliability of the component and conduct the life - cycle prediction:

[0074] R(t)=1 - F(t)

[0075]

[0076]

[0077] Among them, R(t) represents the reliability, F(t) represents the cumulative distribution function of the log - normal distribution, μ represents the mean of the data after logarithmic transformation, and σ represents the standard deviation of the data after logarithmic transformation.

[0078] Preferably, the defined failure threshold of the electronic component is 50 mA.

[0079] Advantages of the present invention: A reliability prediction system for electronic components proposed by the present invention, compared with traditional parameter fitting methods (such as the least squares method and the gradient descent method), uses the sparrow search algorithm (SSA algorithm) with a custom fitness function to achieve efficient optimization of model parameters, and interactively sets the number of search individuals and the number of search iterations of the algorithm. This process iteratively searches to minimize the difference between the model and the actual data, solves for parameter fitting, improves the accuracy of parameter fitting, and thus obtains a high-precision reliability model for electronic components. Its advantages include:

[0080] 1. High efficiency: Compared with traditional manual data processing, the present invention improves the data processing speed, can reduce the data processing time by about 50%-70%, and through an automated and integrated process, the solution efficiency is increased by about 40-60%, significantly improving the overall efficiency of parameter fitting and reliability prediction.

[0081] 2. Ease of use: Compared with traditional reliability prediction methods that require professional programming and statistical analysis knowledge, the user interface of the present invention is friendly and the operation is simple. The learning and operation cost for non-professional users is reduced by about 60%, enabling a wider user group to easily get started and improve efficiency. Through interactive settings, it has a wider applicability for electronic components.

[0082] 3. High integration: Integrates data uploading, processing, model fitting, visualization, and reliability assessment into one, making the process more coherent, reducing the efficiency loss caused by using multiple tools by about 30-50%, and achieving a system response time of seconds, significantly superior to the processing time of traditional software.

[0083] 4. High calculation accuracy: Uses the sparrow search algorithm to handle complex, multi-dimensional, and non-linear problems in model parameter fitting, balances global search and local search, effectively avoids the model parameter solution from falling into local optimal solutions, and improves the accuracy of reliability prediction by about 20-40%, depending on the specific data set and the choice of solution model. Brief Description of the Drawings

[0084] Figure 1 is the method flow block diagram of reliability prediction using the reliability prediction system for electronic components described in the present invention;

[0085] Figure 2 is the interactive data processing result, showing the comparison between the original data and the smoothed data of the data set;

[0086] Figure 3 is the training iteration loss in the model parameter fitting process, where the abscissa represents the iteration, the ordinate represents the loss function, and the curve represents the optimization process;

[0087] Figure 4It is a schematic diagram of the model fitting results under four conditions;

[0088] Figure 5 It is a schematic diagram of the reliability solution results. Specific implementation manners

[0089] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0090] It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments may be combined with each other.

[0091] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, but it is not limited to the present invention.

[0092] Embodiment 1:

[0093] The following is combined with Figure 1 This embodiment is described. The reliability prediction system for an electronic component described in this embodiment includes a data processing module, a component model selection module, a reliability model construction module, and a reliability prediction module;

[0094] The data processing module is used to upload test data on the Web interface;

[0095] The component model selection module is used to select different component degradation models according to the reliability prediction target and construct them based on the test data;

[0096] The reliability model construction module is used to optimize the parameters of the component degradation model by using the sparrow optimization algorithm, construct an iterative search minimization model for the number of search individuals and the number of search iterations by using the interactive setting algorithm, and solve the parameter fitting by the difference between the iterative search minimization model and the actual data, so as to construct a reliability model;

[0097] The reliability prediction module is used to calculate the mean and variance of the normal distribution of the initial data based on the reliability model, calculate the logarithmic normal cumulative distribution function, and use the logarithmic normal cumulative distribution function to predict the life cycle to obtain the reliability prediction result of the component.

[0098] Furthermore, the test data uploaded by the data processing module on the Web interface is in Excel format;

[0099] The data processing module uploads the test data, including: data input, data reading, data processing, and data online editing;

[0100] The data processing includes: data cleaning and formatting;

[0101] The data online editing includes: adjusting data on the Web interface, removing outliers using the quartile method, and performing data denoising using the exponentially weighted moving average method;

[0102] The data reading and data processing are implemented using the Pandas library; the data online editing is implemented using the Streamlit-aggrid library.

[0103] Furthermore, the component degradation model includes a Schottky barrier diode Au-Si contact degradation model, a Schottky barrier diode metallization electromigration model, a PN junction rectifier diode PN junction characteristic degradation model, a PN junction rectifier diode hot carrier injection model, a bipolar transistor hot carrier injection model, and a bipolar transistor PN junction characteristic degradation model.

[0104] Furthermore, the specific method for constructing the PN junction rectifier diode PN junction characteristic degradation model includes:

[0105] Obtain the test data of electronic components under four groups of stress conditions; upload the test data on the Web interface and perform data cleaning and formatting;

[0106] Construct the PN junction rectifier diode PN junction characteristic degradation model as:

[0107]

[0108] where, I R represents the reverse leakage current of the PN junction rectifier diode measured at different times t, I R0 represents the initial value of the reverse leakage current, A represents the degradation scaling factor, which is used to quantify the influence degree of stress on degradation, V R represents the reverse voltage applied to the diode, m represents the voltage exponent, which is used to describe the influence degree of reverse voltage on the degradation process, E a is the activation energy, which represents the energy barrier that needs to be overcome during the degradation process, T is the absolute temperature, with the unit of K, p is the time exponent, which describes the influence of time on the degradation process, and k is the Boltzmann constant, which is used to convert energy to temperature;

[0109] where, the parameters A, m, E a and p are the parameters to be solved;

[0110] Use the sparrow optimization algorithm to optimize the parameters of the component degradation model, and its specific process includes:

[0111] Define the objective function as:

[0112]

[0113] where MSE m represents the mean square error under condition m, m are four different groups of stress conditions, n represents the number of data points, and I R (i) represents the actual reverse leakage current value of the i-th data point, and I' R (i) represents the predicted reverse leakage current value of the i-th data point calculated using the current parameter set and the model formula;

[0114] Define the fitness function as:

[0115] MSE = MSE1 + MSE2 + MSE3 + MSE4

[0116] where MSE1, MSE2, MSE3, and MSE4 represent the mean square errors under the four groups of stress conditions respectively;

[0117] Substitute the data into the sparrow algorithm, and randomly initialize the positions of each individual in the sparrow population. Since the individual is a sparrow, for each sparrow i:

[0118] X′ i = (A i , m i , E ai , p i ), i = 1, 2, 3,..., N

[0119] where X′ i represents the parameter vector of the position of each sparrow i, representing a set of potential solutions, namely the parameters A, m, E a and p, and N represents the size of the sparrow population;

[0120] The update of the sparrow position depends on the behavior strategy. Update the discoverer i. If there is no predator, that is, r2 < 0.8, then:

[0121]

[0122] where I t represents the current iteration number, I tmax represents the maximum number of iterations, which determines the number of iteration rounds of the algorithm. r1 and r2 represent random numbers in [0, 1], introducing random behavior to simulate the natural behavior of sparrows, and exp represents the exponential function to simulate the change of the sparrow flight trajectory;

[0123] If there is a predator, then:

[0124] X′i = X' i + Q × 1

[0125] Where Q represents a random number of the normal distribution, and 1 represents a vector of all 1s;

[0126] Update the positions of the followers. For each follower i, if i is greater than half of the population, then:

[0127]

[0128] Where X worst represents the position of the sparrow with the worst fitness in the current population;

[0129] Otherwise:

[0130]

[0131] Where X best represents the position of the previous optimal sparrow, i.e., the position of the sparrow with the best performance in the population, and A represents a random matrix;

[0132] Update the positions of the early warning sparrows. While maintaining vigilance against the current best position, by introducing randomness, the sparrows can explore new possible areas, thus increasing the chance of finding a better solution. For each early warning sparrow i, if its fitness is greater than the global best, then:

[0133] X' i = X gbest + Q × |X' i - X gbest |

[0134] Where X gbest represents the best position found among all sparrows so far;

[0135] Otherwise:

[0136]

[0137] Where r3 represents a random number in [0, 1], and MSE worst represents the worst fitness in the population, and ε is a very small positive number used to prevent division by zero errors;

[0138] Repeat the above steps until the maximum number of iterations is reached. The maximum number of iterations is set manually through interaction.

[0139] Furthermore, the test data is the forward voltage drop degradation data from 0 to 1000 h, with each time interval point being 100 h, and 10 samples are tested under each stress condition.

[0140] Furthermore, the four stress conditions are respectively:

[0141] The ambient temperature is 398 K and the reverse bias voltage is 800 V;

[0142] The ambient temperature is 398 K and the reverse bias voltage is 1000 V;

[0143] The ambient temperature is 398 K and the reverse bias voltage is 1300 V;

[0144] The ambient temperature is 423 K and the reverse bias voltage is 1300 V.

[0145] Furthermore, the specific methods for removing outliers using the interquartile range method and for data denoising using the exponentially weighted moving average method include:

[0146] IQR = Q3 - Q1

[0147] B L = Q1 - 1.5 × IQR

[0148] B U = Q3 + 1.5 × IQR

[0149] B L <X i <B U

[0150] X t = α·X i + (1 - α)·X t-1

[0151] where IQR represents the interquartile range, Q3 represents the third quartile, i.e., the 75% quantile, Q1 represents the first quartile, i.e., the 25% quantile, B L represents the lower boundary of the outliers, B U represents the upper boundary of the outliers, X i represents the data after removing the outliers, α represents the smoothing coefficient, X t represents the smoothed value at time point t, X t-1 represents the smoothed value at time t - 1.

[0152] Furthermore, by interactively setting the maximum number of iterations to 100 manually, the values of parameters A, m, E a and p are as follows:

[0153] A = -24, m = 4, E a = 0.76, p = 0.2;

[0154] The PN - junction characteristic degradation model of the PN - junction rectifier diode is obtained as:

[0155]

[0156] Furthermore, the obtained degradation model of the PN junction characteristics of the PN junction rectifier diode is as follows:

[0157]

[0158] Furthermore, the specific method for the reliability prediction module to obtain the reliability prediction result of the component includes:

[0159] Define the failure threshold of the electronic component, calculate the initial data, that is, at the moment of t = 0, the mean and standard deviation of the normal distribution of all data, and use the cumulative distribution function of the lognormal distribution to predict the reliability of the component and conduct the life cycle prediction:

[0160] R(t) = 1 - F(t)

[0161]

[0162]

[0163] Among them, R(t) represents the reliability, F(t) represents the cumulative distribution function of the lognormal distribution, μ represents the mean of the data after logarithmic transformation, and σ represents the standard deviation of the data after logarithmic transformation.

[0164] Furthermore, the defined failure threshold of the electronic component is 50 mA.

[0165] In the present invention, a reliability prediction system for electronic components is proposed, as Figure 1 shown, including:

[0166] Data processing: Upload test data in Excel format through the Web interface, and the application uses the pandas library to read and process the data. At the same time, use the streamlit - aggrid library for online data editing, directly adjust the data on the interface, use the interquartile range method (IQR) to remove outliers, and then perform data noise reduction processing through the exponentially weighted moving average method, enhancing the flexibility of data processing;

[0167] Selection of electronic component models: Multiple component degradation models can be established, such as the Schottky barrier diode Au - Si contact degradation model, the Schottky barrier diode metallization electromigration model, the PN junction rectifier diode PN junction characteristic degradation model, the PN junction rectifier diode hot carrier injection model, the bipolar transistor hot carrier injection model, and the bipolar transistor PN junction characteristic degradation model, etc.;

[0168] Parameter Optimization and Reliability Model Establishment: Compared with traditional parameter fitting methods (such as the least squares method and the gradient descent method), the present invention uses a sparrow search algorithm (SSA algorithm) with a custom fitness function to achieve efficient optimization of model parameters, and interactively sets the number of search individuals and the number of search iterations of the algorithm. This process solves the parameter fitting by iteratively searching to minimize the difference between the model and the actual data, improves the accuracy of parameter fitting, and thus obtains a high-precision reliability model for electronic components;

[0169] Reliability Calculation: Based on the optimized parameters, interactively set the failure thresholds of different electronic components, the mean and standard deviation of the initial data normal distribution, use the lognormal distribution to predict the reliability of the components, and perform life cycle prediction.

[0170] The key devices involved in the performance degradation of the chip include Schottky barrier diodes, PN junction rectifier diodes, and bipolar transistors, etc. Here, taking the PN junction rectifier diode as an example, the performance degradation analysis of the PN junction rectifier diode is carried out.

[0171] First, collect the test data of electronic components under different working conditions, select four groups of stress conditions, namely: ① ambient temperature 398K, reverse bias voltage 800V, ② ambient temperature 398K, reverse bias voltage 1000V, ③ ambient temperature 398K, reverse bias voltage 1300V, and ④ ambient temperature 423K, reverse bias voltage 1300V, extract the forward voltage drop degradation data from 0 to 1000h, with a time interval of 100h each time, and test 10 samples under each group of stress conditions.

[0172] Subsequently, use the Web application provided by the present invention to upload the test data and perform data cleaning and formatting. The Web application is written in the python language, defines data processing functions, including outlier removal and data smoothing processing of the data:

[0173] IQR = Q3 - Q1

[0174] B L = Q1 - 1.5 × IQR

[0175] B U = Q3 + 1.5 × IQR

[0176] B L <X i <B U

[0177] X t =α·X i +(1 - α)·X t-1

[0178] Wherein, IQR is the interquartile range, Q3 is the third quartile (75% quantile), Q1 is the first quartile (25% quantile), and B L is the lower boundary of the outlier, and B U is the upper boundary of the outlier, X i is the data after removing the outliers, α is the smoothing coefficient, X t is the smoothed value at time point t, and X t-1 is the smoothed value at time t-1.

[0179] For the PN junction rectifier diode, a PN junction characteristic degradation model of the PN junction rectifier diode is selected to be established:

[0180]

[0181] Wherein, I R is the reverse leakage current of the PN junction rectifier diode measured at different times t, and I R0 is the initial value of the reverse leakage current. A represents the degradation scaling factor, which is used to quantify the influence degree of stress on degradation. V R represents the reverse voltage applied to the diode, m represents the voltage exponent, which is used to describe the influence degree of the reverse voltage on the degradation process, and E a is the activation energy, which represents the energy barrier that needs to be overcome during the degradation process. T is the absolute temperature, with the unit of K. p is the time exponent, which describes the influence of time on the degradation process. k is the Boltzmann constant, which is used to convert energy into temperature.

[0182] Among them, the parameters A, m, and E a and p are unknown parameters to be solved. The present invention innovatively uses the Sparrow Search Algorithm (SSA) to solve the parameters, and the objective function is defined as:

[0183]

[0184] Among them, MSE m represents the mean square error under the condition of m, n is the number of data points, and I R (i) is the actual reverse leakage current value of the i-th data point, and I' R (i) is the predicted reverse leakage current value of the i-th data point calculated using the current parameter set and the model formula. m refers to four different stress conditions.

[0185] The fitness function is defined as:

[0186] MSE = MSE1 + MSE2 + MSE3 + MSE4

[0187] Substitute the data into the sparrow algorithm, and randomly initialize the positions of each individual (sparrow) in the sparrow population. For each sparrow i:

[0188] X′ i =(A i ,m i ,E ai ,p i ), i = 1, 2, 3, ..., N

[0189] Wherein, X′ i is the parameter vector of the position of each sparrow i, representing a set of potential solutions, namely the parameters A, m, E a and p, and N represents the size of the sparrow population.

[0190] The update of the sparrow position depends on their behavioral strategies. Update the discoverer i. If there is no predator, that is, r2 < 0.8, then:

[0191]

[0192] Wherein, I t represents the current iteration number, I tmax represents the maximum number of iterations, which determines the number of iteration rounds of the algorithm operation. r1 and r2 represent random numbers in [0, 1], introducing random behavior to simulate the natural behavior of sparrows. exp represents the exponential function, simulating the change of the sparrow flight trajectory.

[0193] If there is a predator, then:

[0194] X′ i = X′ i + Q × 1

[0195] Wherein, Q represents a random number of the normal distribution, and 1 represents a vector of all 1s.

[0196] Subsequently, update the position of the followers. For each follower i, if i is greater than half of the population, then:

[0197]

[0198] Wherein, X worst represents the position of the sparrow with the worst fitness in the current population.

[0199] Otherwise:

[0200]

[0201] Wherein, X best represents the position of the previous optimal sparrow, that is, the position of the best-performing sparrow in the population, and A represents a random matrix.

[0202] Subsequently, update the position of the early warning agents. While maintaining "vigilance" over the current best position (i.e., not moving too far away from the best position), introduce randomness to enable the sparrows to explore new possible areas, thereby increasing the chance of finding a better solution. For each early warning agent i, if its fitness is greater than the global best, then:

[0203] X′ i = X gbest + Q × |X′ i - X gbest |

[0204] where X gbest represents the best position found among all sparrows so far.

[0205] Otherwise:

[0206]

[0207] where r3 represents a random number in [0, 1], MSE worst represents the worst fitness in the population, and ε is a very small positive number used to prevent division-by-zero errors.

[0208] Finally, repeat the above steps until the maximum number of iterations is reached. The maximum number of iterations can be set manually interactively, and here it is taken as 100. Finally, find the numerical values of parameters A, m, E a and p. A = -24, m = 4, E a = 0.76, p = 0.2.

[0209] As Figure 2 shown, it is the interactive data processing result, showing the comparison of the original data and the smoothed data in the dataset.

[0210] As Figure 3 shown, it is the training iteration loss during the model parameter fitting process. Among them, the abscissa represents the iteration, the ordinate represents the loss function, and the curve represents the optimization process.

[0211] Finally, obtain the PN junction characteristic degradation model of the PN junction rectifier diode:

[0212]

[0213] The model fitting results under four conditions are as Figure 4 shown, and the 95% confidence interval is used to verify the model accuracy.

[0214] Subsequently, continue to use the model to solve the reliability of electronic components, as Figure 5As shown, it is a schematic diagram of the reliability solution result. The failure threshold of the device is defined as 50 mA. Calculate the initial data, that is, at the moment of t = 0, the mean and standard deviation of the normal distribution of all data. Use the cumulative distribution function of the log-normal distribution to predict the reliability of the components and conduct life cycle prediction:

[0215] R(t) = 1 - F(t)

[0216]

[0217]

[0218] In the formula, R(t) represents the reliability, F(t) represents the cumulative distribution function of the log-normal distribution, μ represents the mean of the data after logarithmic transformation, and σ represents the standard deviation of the data after logarithmic transformation.

[0219] In the present invention, an interactive application system is created using Python. Innovatively, the reliability prediction of electronic components is interactively modeled, and the sparrow optimization algorithm is innovatively used for parameter fitting in reliability modeling, including data model selection, online data editing and preprocessing, component reliability modeling, component reliability solution calculation, and result visualization, etc.

[0220] Although the present invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the present invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed, as long as they do not deviate from the spirit and scope of the present invention defined by the appended claims. It should be understood that different dependent claims and the features described herein can be combined in a manner different from that described in the original claims. It should also be understood that the features described in connection with a single embodiment can be used in other described embodiments.

Claims

1. A reliability prediction system for electronic components, characterized in that: It includes data processing module, component model selection module, reliability model building module and reliability prediction module; The data processing module is used to upload the test data on the Web interface; The component model selection module is used to select different component degradation models according to the reliability prediction target and construct them based on the test data; The reliability model building module is used to optimize the parameters of the component degradation model using the sparrow optimization algorithm, to build an iterative search minimization model for the number of search individuals and the number of search iterations using the interactive setting algorithm, to solve the parameter fitting by the difference between the iterative search minimization model and the actual data, and to build a reliability model; The reliability prediction module is used to calculate the mean variance of the normal distribution of the initial data based on the reliability model, calculate the log-normal cumulative distribution function, use the log-normal cumulative distribution function to perform life cycle prediction, and obtain the reliability prediction results of the components; The component degradation models include Schottky barrier diode gold semiconductor contact degradation model, Schottky barrier diode metallization electromigration model, PN junction rectifier diode PN junction characteristic degradation model, PN junction rectifier diode hot carrier injection model, bipolar transistor hot carrier injection model and bipolar transistor PN junction characteristic degradation model; The specific method of constructing the PN junction characteristic degradation model of the PN junction rectifier diode includes: Obtain test data of electronic components under four sets of stress conditions; upload the test data on the Web interface, and clean and format the data; The PN junction characteristic degradation model of the PN junction rectifier diode is constructed as follows: Among them, I R It indicates the reverse leakage current of the PN junction rectifier diode measured at different time t, I R0 represents the initial value of the reverse leakage current, A represents the proportionality factor of the degradation, which is used to quantify the influence of stress on the degradation, V R represents the reverse voltage applied to the diode, m represents the voltage index, which is used to describe the influence of the reverse voltage on the degradation process, E a is the activation energy, which represents the energy barrier that needs to be overcome during the degradation process, T is the absolute temperature in K, p is the time index, which describes the effect of time on the degradation process, and k is the Boltzmann constant, which is used to convert energy into temperature; Among them, parameters A, m, E a and p are the parameters to be solved; The parameters of the component degradation model are optimized by using the sparrow optimization algorithm, and the specific process includes: The objective function is defined as: Among them, MSE m represents the mean square error under condition m, where m is four different sets of stress conditions, n is the number of data points, and I R (i) represents the actual reverse leakage current value of the i-th data point, I' R (i) represents the predicted reverse leakage current value of the i-th data point calculated using the current parameter set and model formula; The fitness function is defined as: MSE=MSE1+MSE2+MSE3+MSE4 Among them, MSE1, MSE2, MSE3, and MSE4 represent the mean square errors under four groups of stress conditions; Bring the data into the sparrow algorithm and randomly initialize the position of each individual in the sparrow population. The individual is the sparrow, so for each sparrow i: X i '=(A i ,m i ,E ai ,p i ),i=1,2,3,...,N Among them, X i ' represents the parameter vector of each sparrow i's position, representing a set of potential solutions, namely parameters A, m, E a and p, N represents the size of the sparrow colony; The update of the sparrow's position depends on the behavioral strategy. If there is no predator, that is, r2 < 0.8, then: Among them, I t Indicates the current iteration number, I tmax It represents the maximum number of iterations, which determines the number of iterations of the algorithm. r1 and r2 represent random numbers in the range [0,1], which introduce random behavior and simulate the natural behavior of sparrows. exp represents an exponential function, which simulates the changes in the flight trajectory of sparrows. If a predator is present, then: X i '=X i '+Q×1 Among them, Q represents a random number of normal distribution, and 1 represents a vector of all 1s; Update the follower positions, for each follower i, if i is greater than half of the population, then: Among them, X worst Indicates the position of the sparrow with the worst fitness in the current group; otherwise: Among them, X best represents the position of the previous optimal sparrow, that is, the position of the sparrow with the best performance in the group, and A represents a random matrix; Update the position of the early warning person, keep alert to the current best position, and introduce randomness to enable the sparrow to explore new possible areas, thereby increasing the chance of finding a better solution. For each early warning person i, if its fitness is greater than the global best, then: X i '=X gbest +Q×|X i '-X gbest | Among them, X gbest represents the best position found among all the sparrows so far; otherwise: Among them, r3 represents a random number [0,1], MSE worst Represents the worst fitness in the population, ε is a very small positive number used to prevent division by zero errors; Repeat the above steps until the maximum number of iterations is reached, which is manually set interactively; By interactively manually setting the maximum number of iterations to 100, the parameters A, m, and E are obtained. a The values ​​of and p are: A=-24,m=4,E a =0.76,p=0.2; The degradation model of the PN junction characteristics of the PN junction rectifier diode is obtained as follows:

2. The reliability prediction system of an electronic component according to claim 1, characterized in that: The test data uploaded by the data processing module on the Web interface is in Excel format; The data processing module uploads the test data including: data input, data reading, data processing and data online editing; The data processing includes: cleaning and formatting of data; The online data editing includes: adjusting data on a web interface, removing outliers using the quartile method, and reducing data noise using the exponential weighted average method; The data reading and data processing are implemented using the Pandas library; the online editing of the data is implemented using the Streamlit-aggrid library.

3. The reliability prediction system of an electronic component according to claim 1, characterized in that: The test data is the forward voltage drop degradation data of 0-1000h, each time interval is 100h, and 10 samples are tested for each set of stress conditions.

4. The reliability prediction system of an electronic component according to claim 1, characterized in that: The four sets of stress conditions are: Ambient temperature 398K, reverse bias voltage 800V; Ambient temperature 398K, reverse bias voltage 1000V; Ambient temperature 398K, reverse bias voltage 1300V; Ambient temperature 423K, reverse bias voltage 1300V.

5. The reliability prediction system of an electronic component according to claim 1, characterized in that: The specific methods of using the quartile method to eliminate outliers and the exponential weighted average method to reduce data noise include: IQR=Q3-Q1 B L =Q1-1.5×IQR <h2 style=";text-align:left;direction:ltr">B<h2 style=";text-align:left;direction:ltr"> U <h2 style=";text-align:left;direction:ltr"> =Q3+1.5×IQR B L <X i <B U X t =α·X i +(1-a)·X t-1 Among them, IQR represents the interquartile range, Q3 represents the third quartile, i.e. the 75% quartile, Q1 represents the first quartile, i.e. the 25% quartile, and B L represents the lower boundary of the outlier, B U represents the upper boundary of the outlier, X i represents the data after removing outliers, α represents the smoothing coefficient, and X t represents the smoothed value at time point t, X t-1 Represents the smoothed value at time t-1.

6. The reliability prediction system of an electronic component according to claim 1, characterized in that: The specific method for the reliability prediction module to obtain the reliability prediction result of the component includes: Define the failure threshold of electronic components, calculate the initial data, that is, the mean and standard deviation of the normal distribution of all data at time t = 0, and use the cumulative distribution function of the log-normal distribution to predict the reliability of components and perform life cycle prediction: R(t)=1-F(t) Where R(t) represents reliability, F(t) represents the cumulative distribution function of the log-normal distribution, μ represents the mean of the data after logarithmic transformation, and σ represents the standard deviation of the data after logarithmic transformation.

7. The reliability prediction system of electronic components according to claim 6, characterized in that: The failure threshold of the electronic components is defined as 50 mA.

Images