A reliability prediction method, device and system for a field effect transistor

Abstract

The present invention belongs to the technical field of field effect transistor performance prediction, and particularly relates to a reliability prediction method, device and system for a field effect transistor. By using the correlation analysis method, the correlation between the changes of environmental parameters, electrical parameters and performance parameters in the time series data of historical operation records is determined, and the key performance parameters are screened out. Then, before using the screened key performance parameters for life prediction, a virtual operation model of the field effect transistor is constructed, and the historical data is used to verify the effectiveness of the virtual operation model. Finally, when using the successfully verified virtual operation model, the dynamic changes of environmental parameters and electrical parameters are simulated through historical data, the degradation trend of the key performance parameters is recorded, and the remaining life prediction is carried out based on this, which not only improves the prediction efficiency, but also enables the life prediction to more accurately reflect the actual working conditions and improves the accuracy of the remaining life prediction.

Description

Technical Field

[0001] The present invention belongs to the technical field of field effect transistor performance prediction, and particularly relates to a reliability prediction method, device and system for field effect transistors. Background Art

[0002] As a key component in modern electronics, the field effect transistor has significant advantages such as high input impedance, low noise coefficient, and fast switching speed. These characteristics have enabled its wide application in fields such as analog circuits and digital circuits. In many applications, especially in industries with strict requirements for safety and reliability (such as aerospace, medical equipment, and the automotive industry), the failure of field effect transistors can lead to serious consequences. Therefore, it is crucial to evaluate their service life through reliability prediction.

[0003] In the prior art, various reliability prediction schemes for field effect transistors have been proposed. For example, the Chinese invention patent with the publication number CN118095182B proposes a reliability prediction method for power field effect transistors, and the specific steps are as follows: S1: Collect initial data through sensors and preprocess it; S2: Optimize the hyperparameters of the Transformer model using a genetic algorithm to construct a GA-Transformer model; S3: Input the data obtained in step S1 into the GA-Transformer model constructed in step S2, evaluate the accuracy of the prediction based on performance indicators, predict the remaining service life, and finally output the reliability prediction result.

[0004] Although the above scheme shows certain advantages in using advanced models for life prediction, it mainly focuses on improving prediction accuracy through model optimization and data processing, lacking the screening and differentiation of key performance indicators. Specifically, field effect transistors have multiple performance indicators, such as threshold voltage (Vth), on-resistance (Rds(on)), transconductance (gm), etc. However, not all of these performance indicators have a significant impact on the device's life. If prediction is carried out without screening key performance indicators, the model will contain a large amount of redundant information, resulting in an increase in model complexity. This will not only consume more computing resources but also may reduce the model's operating efficiency and prediction accuracy.

[0005] In addition, even if certain performance indicators are considered important under general circumstances, their importance may change under specific actual operating conditions. If key performance indicators are not screened according to actual operating conditions for prediction, it is easy to cause deviations in the prediction results and fail to accurately reflect the true health status of the device. Summary of the Invention

[0006] The object of the present invention is to improve the deficiencies existing in the prior art, and to provide a method, device and system for predicting the reliability of a field effect transistor, which realizes targeted prediction of the life of the field effect transistor by introducing the screening of key performance indicators under actual operating conditions.

[0007] The object of the present invention can be achieved by the following technical solutions: In the first aspect of the present invention, a method for predicting the reliability of a field effect transistor is provided, including the following steps: Step1: Retrieve the historical operation records of the field effect transistor from the operation database, including the time series data of environmental parameters, power parameters and performance parameters.

[0008] Step2: Determine the correlation between the changes in environmental parameters or power parameters and the changes in performance parameters through cross-correlation analysis of the historical operation records, and screen out the key performance parameters.

[0009] Step3: Construct a virtual operation model based on the material characteristics and device structure of the field effect transistor. The input of the virtual operation model is environmental parameters and power parameters, and the output is performance parameters.

[0010] Step4: Input the time series data of environmental parameters and electrical parameters in the historical operation records into the virtual operation model, and verify the effectiveness of the model by comparing the performance parameters output by the model with the corresponding performance parameters in the historical operation records.

[0011] Step5: When the model is verified to be effective, simulate the dynamic changes of environmental parameters and electrical parameters in the virtual operation model using historical data, and record the degradation trend of the key performance parameters.

[0012] Step6: Predict the remaining life and accelerated degradation time of the field effect transistor according to the degradation trend, and generate a life evaluation report.

[0013] In the second aspect of the present invention, a device for predicting the reliability of a field effect transistor is proposed, and the device includes a method for predicting the reliability of a field effect transistor according to the present invention.

[0014] In the third aspect of the present invention, a system for predicting the reliability of a field effect transistor is proposed, including the following modules: Historical operation data retrieval module: Retrieve the historical operation records of the field effect transistor from the operation database, including the time series data of environmental parameters, power parameters and performance parameters.

[0015] Key performance parameter identification module: Determine the correlation between the changes in environmental parameters or power parameters and the changes in performance parameters through cross-correlation analysis of the historical operation records, and screen out the key performance parameters.

[0016] Virtual operation model construction module: Construct a virtual operation model based on the material characteristics and device structure of the field-effect transistor. The input of the virtual operation model is environmental parameters and power parameters, and the output is performance parameters.

[0017] Model validity verification module: Input the time-series data of environmental parameters and electrical parameters in the historical operation record into the virtual operation model, and verify the validity of the model by comparing the performance parameters output by the model with the corresponding performance parameters in the historical operation record.

[0018] Degradation dynamic simulation module: When verifying the effectiveness of the model, simulate the dynamic changes of environmental parameters and electrical parameters in the virtual operation model using historical data, and record the degradation trend of key performance parameters.

[0019] Remaining life prediction module: Predict the remaining life and accelerated degradation time of the field-effect transistor according to the degradation trend, and generate a life assessment report.

[0020] Combining all the above technical solutions, the positive effects of the present invention are as follows: 1. By analyzing the historical operation record of the field-effect transistor and using the correlation analysis method, the present invention determines the correlation between its changes and the changes in performance parameters from the time-series data of environmental parameters, electrical parameters, and performance parameters, and screens out key performance parameters, ensuring the relevance and effectiveness of the input data of the subsequent life prediction model, and significantly improving the prediction efficiency and the matching with the actual working conditions.

[0021] 2. Before predicting the life using the selected key performance parameters, the present invention constructs a virtual operation model of the field-effect transistor and uses historical data to verify the effectiveness of the virtual operation model. Only when the verification is effective, the model is used for life prediction, avoiding prediction deviations caused by inaccurate models, and improving the prediction accuracy and credibility to a certain extent.

[0022] 3. When using the verified virtual operation model, the present invention simulates the dynamic changes of environmental parameters and electrical parameters using historical data, records the degradation trend of key performance parameters, and predicts the remaining life based on this, enabling the life prediction to more accurately reflect the actual working conditions and improving the accuracy of the remaining life prediction.

[0023] 4. When predicting the remaining life using the verified virtual operation model, the present invention also adds the prediction of the accelerated degradation time based on the degradation trend curve, which can detect in advance the stage when the device is about to enter the rapid degradation stage, provide early warning signals, and help the operation and maintenance personnel take preventive measures in time. Description of the Drawings

[0024] The present invention will be further described with reference to the accompanying drawings. However, the embodiments in the drawings do not constitute any limitation to the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on the following drawings without creative work.

[0025] Figure 1 This is the flowchart of the method implementation steps of the present invention.

[0026] Figure 2 This is the implementation flowchart for screening key performance parameters in the present invention.

[0027] Figure 3 This is the schematic diagram of the system module connection of the present invention. Detailed implementation manners

[0028] 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. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work belong to the scope of protection of the present invention.

[0029] Embodiment 1

[0030] Refer to Figure 1 As shown, the present invention provides a method for predicting the reliability of a field effect transistor, including the following steps: Step 1: Retrieve the historical operation records of the field effect transistor from the operation database, including the time series data of environmental parameters, electrical parameters, and performance parameters.

[0031] It should be noted that the present invention is for predicting the lifespan of the field effect transistor under actual use conditions, rather than under the factory state. Manufacturers usually predict the lifespan under the factory state based on specific ideal conditions, but there may be significant differences between the actual use conditions and these ideal conditions, resulting in the difficulty of applying the lifespan prediction under the factory state to various complex actual application scenarios. In contrast, the lifespan prediction under the use state is more practical.

[0032] Furthermore, it should be noted that when predicting the lifespan of the field effect transistor under actual use conditions, it is necessary to rely on the data during its actual operation. These data are obtained from the actual operation environment of the field effect transistor through a data acquisition terminal, and detailed operation records are generated and stored in the operation database. These records contain the time series data of environmental parameters, electrical parameters, and performance parameters during the actual operation of the device.

[0033] The above environmental parameters include but are not limited to temperature, humidity, etc., which reflect the external environmental conditions where the field effect transistor is located.

[0034] Power parameters include but are not limited to voltage, current, load, etc., which describe the electrical conditions of the field - effect transistor during actual operation.

[0035] Performance parameters include but are not limited to the following: Threshold voltage (Vth): The voltage that determines the transistor to turn on.

[0036] On - resistance (Rds(on)): The resistance of the transistor when it is in the on - state.

[0037] Transconductance (gm): An important parameter that measures the amplification ability of the transistor.

[0038] Drain - source breakdown voltage (Vds(max)): The maximum voltage that the transistor can withstand.

[0039] Switching delay time (td(on), td(off)): The time required for the transistor to turn from off to on or from on to off.

[0040] During the actual operation of the field - effect transistor, environmental parameters and power parameters will have a significant impact on performance parameters. Specifically, the impact of environmental parameters: Temperature changes will affect the threshold voltage and on - resistance of the transistor, resulting in changes in its performance. Other environmental factors such as humidity may also affect the long - term reliability of the device.

[0041] The impact of power parameters: Changes in voltage and current directly affect the operating state of the transistor. Excessive voltage may cause breakdown, and excessive current may cause overheating and damage. Load changes will also affect the operating efficiency and stability of the transistor.

[0042] It should be added that the application circuit of the field - effect transistor is also recorded in the historical operation records. This application circuit information reflects the specific topology of the application circuit, including the connection method of the device with other components (such as resistors, capacitors, diodes, etc.) and its position in the network.

[0043] Step2. Determine the correlation between changes in environmental parameters or power parameters and changes in performance parameters through cross - correlation analysis of historical operation records, and screen out key performance parameters.

[0044] Please refer to Figure 2 , in the preferred implementation of the above - mentioned scheme, the correlation between changes in environmental parameters or power parameters and changes in performance parameters determined through correlation analysis is as follows: Plot the environmental parameter change curve, power parameter change curve, and performance parameter change curve in a coordinate system with time as the horizontal axis and environmental parameters, power parameters, and performance parameters as the vertical axis for the time - series data of environmental parameters, power parameters, and performance parameters in each historical operation record.

[0045] It should be emphasized that when screening key performance parameters using historical operation records, the historical operation records of the same application circuit must be used for screening. This is because different application circuits have unique topological structures, operating modes, and electrical stress conditions, and these factors will affect the performance and degradation modes of field-effect transistors. By using the data of the same application circuit, it can be ensured that the screened performance parameters match the operating conditions of this specific circuit and can more accurately reflect the degradation of the device in the actual application scenario.

[0046] Mark the inflection points on the environmental parameter change curve, power parameter change curve, and performance parameter change curve in each historical operation record, and form an environmental parameter inflection point set, a power parameter inflection point set, and a performance parameter inflection point set with the timestamps corresponding to the marked inflection points.

[0047] It should be pointed out that the inflection point in the curve refers to the point where the curve changes from one change trend to another, usually manifested as a change in slope, reflecting the change in the fluctuation characteristics of the curve at this point.

[0048] Use the cross-correlation function to calculate the cross-correlation coefficients between each performance parameter and the environmental parameters and power parameters in the historical operation record for the environmental parameter inflection point set and power parameter inflection point set corresponding to each historical operation record, respectively.

[0049] It should be added that when performing cross-correlation analysis using the inflection point set, the inflection points in the inflection point set need to be aligned. This is because aligning the inflection points can ensure the synchronization of the changes in environmental parameters, power parameters, and performance parameters in time and avoid analysis errors caused by time offsets.

[0050] As an example of inflection point alignment, for missing timestamps, linear interpolation is used to fill them.

[0051] Furthermore, it should be added that the cross-correlation function is used to calculate the cross-correlation coefficient between two parameter time series to determine whether there is a lag correlation between them. Specifically, the cross-correlation function can identify whether the change of one parameter will cause the following change of another parameter, thereby revealing the potential time series relationship between the two. The absolute value of the obtained cross-correlation coefficient reflects the strength of the cross-correlation. When the cross-correlation coefficient is large, it indicates a higher cross-correlation strength between the two time series, meaning that the change of one parameter has a more significant impact on the other parameter.

[0052] It should be pointed out that calculating the cross-correlation coefficient using the cross-correlation function belongs to the category of existing technologies, and its specific calculation steps have been described in detail in relevant literature. In view of this, the specific details of this calculation process will not be elaborated in this article.

[0053] Applied to the present invention, calculating the cross-correlation coefficients between each performance parameter in the historical operation records and environmental parameters and power parameters using the cross-correlation function can help identify whether changes in environmental parameters (such as temperature, humidity) or power parameters (such as voltage, current) will cause the follow-up changes in performance parameters (such as threshold voltage, on-resistance), and quantify the strength of this association.

[0054] In a further preferred implementation of the above solution, the key performance parameters are screened as follows: Compare the cross-correlation coefficients between each performance parameter in each historical operation record and environmental parameters and power parameters with the set correlation coefficient threshold. Select the performance parameters whose cross-correlation coefficients reach the correlation coefficient threshold and mark them as associated performance parameters, and record the relevant environmental parameters or power parameters as sensitive factors.

[0055] As an illustration of the above preferred implementation, comparing the cross-correlation coefficients between each performance parameter in each historical operation record and environmental parameters and power parameters with the set correlation coefficient threshold is actually comparing the absolute value of the calculated cross-correlation coefficient with the correlation coefficient threshold.

[0056] In an example, since the value range of the cross-correlation coefficient is between, the purpose of setting the correlation coefficient threshold is to effectively distinguish strongly correlated and weakly correlated performance parameters. The specific threshold can be set according to actual needs. Usually, a higher correlation coefficient threshold may select fewer associated performance parameters, but the relationships between these parameters are more significant and reliable. On the contrary, a lower correlation coefficient threshold may select more associated performance parameters, which can provide more prediction dimensions for subsequent life prediction. For example, it can be 0.6, which can ensure a certain number of associated performance parameters while ensuring that the relationships between these parameters have sufficient strength and reliability, thus providing a solid foundation for subsequent in-depth analysis.

[0057] In a specific embodiment, the cross-correlation coefficients between some performance parameters and some environmental parameters and power parameters in a certain historical operation record are shown in Table 1.

[0058] Table 1: Cross-correlation coefficients between some performance parameters and some environmental parameters and power parameters

[0059]

[0060] As can be seen from Table 1, the cross-correlation coefficients of the threshold voltage with voltage, the threshold voltage with temperature, the on-resistance with voltage, the transconductance with temperature, and the transconductance with voltage all reach the correlation coefficient threshold of 0.6. Therefore, the threshold voltage, on-resistance, and transconductance are all used as the associated performance parameters of the corresponding historical operation record, and the sensitive factors of the threshold voltage are voltage and temperature, the sensitive factor of the on-resistance is voltage, and the sensitive factors of the transconductance are voltage and temperature.

[0061] Classify the associated performance parameters of each historical operation record, count the occurrence proportion of the associated performance parameters, and compare it with the configured effective occurrence proportion. Exemplarily, the effective occurrence proportion is 50%, and filter out the associated performance parameters that reach the effective occurrence proportion as key performance parameters.

[0062] By classifying and summarizing the associated performance parameters in each historical operation record and counting the proportion of each associated performance parameter appearing in all records as described above, performance parameters that frequently appear and have significant relevance can be filtered out as key performance parameters. This method avoids problems such as insufficient data representativeness, contingency, noise interference, and statistical deviation caused by determining key performance parameters solely relying on the associated performance parameters in a single historical operation record, and improves the reliability and accuracy of key performance parameter selection.

[0063] It should be understood that the present invention identifies associated performance parameters by analyzing the cross-correlation between performance parameters, environmental parameters, and electrical parameters in historical operation records, and determines key performance parameters accordingly. This is because when a certain performance parameter does not show significant changes when the operating state (including environmental parameters and electrical parameters) changes, it indicates that the performance parameter is not sensitive to these changes, has small fluctuations, and the corresponding degradation rate is also low. In this case, the significance of studying it as a key performance parameter is limited because its degradation information is not sufficient to provide valuable prediction data; on the contrary, if a certain performance parameter shows significant changes when the operating state changes, it indicates that the performance parameter is more sensitive to changes in environmental and electrical parameters, and its degradation rate is higher and more significant. In this case, studying its degradation trend has high practical significance and can provide important reference for subsequent life prediction and maintenance strategies.

[0064] The method of the present invention for determining key performance parameters by using historical operation data to identify inflection points and calculate cross-correlation coefficients through time series analysis and statistical methods is a data-driven method. Compared with pure theoretical derivation or hypothesis, the key performance parameters determined by this method can more accurately reflect the actual working conditions.

[0065] Step 3: Construct a virtual operation model based on the material characteristics and device structure of the field-effect transistor. The input of the virtual operation model is environmental parameters and power parameters, and the output is performance parameters.

[0066] It should be noted that the virtual operation model is a simulation tool based on physical principles and mathematical models, which can simulate the behavior of actual devices under different working conditions. The virtual operation model is constructed based on the material characteristics and device structure of the field-effect transistor to ensure that the model can be built based on the real physical principles of the field-effect transistor, so as to effectively simulate the performance of the field-effect transistor under various environmental and power conditions, and provide a prediction model that is highly close to the actual working conditions for subsequent life prediction.

[0067] It should be noted that during the construction of the virtual operation model, the application circuit of the field-effect transistor needs to be taken into account to ensure that the model can accurately simulate the behavior and performance of the device in the actual application circuit.

[0068] Step4: Input the environmental parameters and electrical parameter time series data in the historical operation record into the virtual operation model, and verify the effectiveness of the model by comparing the performance parameters output by the model with the corresponding performance parameters in the historical operation record.

[0069] In the way that the above scheme can be realized, the model effectiveness verification is as follows: Compare the output performance parameters obtained when the environmental parameters and power parameter time series data in each historical operation record are input into the virtual operation model with the performance parameters of the corresponding historical operation record, and calculate the simulation error amount.

[0070] As an example of the above implementation method, the mean squared error can be used for the simulation error amount.

[0071] As another example, the mean absolute error can be used for the simulation error amount.

[0072] Statistically analyze the simulation error amounts of each historical operation record to obtain the mean simulation error and the standard deviation of the simulation error.

[0073] It can be understood that the mean simulation error reflects the overall error level of the model, and the standard deviation of the simulation error reflects the degree of dispersion of the errors, which can be used to evaluate the stability of the model.

[0074] Compare the mean simulation error and the standard deviation of the simulation error with the set acceptable critical values respectively. If both the mean simulation error and the standard deviation of the simulation error do not exceed the set acceptable critical values, and if both the error mean and the standard deviation of the model are within the acceptable range, then the model is considered to have high effectiveness and reliability. At this time, the model is verified to be effective; otherwise, the model is verified to be invalid, and it is necessary to consider adjusting the model parameters to improve the accuracy of the model.

[0075] The acceptable critical values corresponding to the above-mentioned mean simulation error and standard deviation of simulation error are both initially set according to requirements. Exemplarily, the acceptable critical value of the mean simulation error is 0.01, and the acceptable critical value of the standard deviation of simulation error is 0.05, which are used to determine whether the model meets the accuracy requirements.

[0076] In the improved implementation of the above solution, Step4 also includes adjusting the accuracy of the virtual running model when the verification model is invalid. The specific adjustment is as follows: In the comparison of the mean simulation error and the standard deviation of simulation error with the set acceptable critical values respectively, if only the mean simulation error exceeds the acceptable critical value, it indicates that the model has a systematic bias. At this time, the complexity of the virtual running model can be increased to improve its fitting ability.

[0077] The specific implementation measures for increasing complexity are as follows: 1. Increase the number of layers or nodes: Appropriately increase the depth and width of the model to enhance the model's expressive ability.

[0078] 2. Introduce more features: Optimize the input features, add new features or improve the existing feature construction methods.

[0079] If only the standard deviation of simulation error exceeds the acceptable critical value, it indicates that the stability of the model is poor. Then, regularization is introduced into the virtual running model to prevent overfitting and improve the generalization ability of the model.

[0080] The specific implementation measures for introducing regularization are as follows: L1 / L2 regularization: Add L1 or L2 regularization terms to the loss function to constrain the magnitude of the model parameters.

[0081] If both the mean simulation error and the standard deviation of simulation error exceed the acceptable critical values, it indicates that the model has both systematic bias and lack of stability. Then, increase the complexity of the virtual running model and introduce regularization.

[0082] It should be added that after adjusting the accuracy of the virtual running model, continue with the validity verification until the verification model is valid and then stop the adjustment.

[0083] Before using the selected key performance parameters to predict the lifespan, the present invention constructs a virtual running model of the field effect transistor and uses historical data to verify the validity of the virtual running model. Only when the verification is valid, the model is used for lifespan prediction, thus avoiding prediction deviations caused by inaccurate models and significantly improving the accuracy and credibility of the prediction. When the verification is invalid, by identifying the specific reasons why the model error does not meet the acceptable critical values, the model accuracy is adjusted accordingly to optimize the model performance and provide reliable support for subsequent lifespan prediction.

[0084] Step 5. When the verification model is effective, use historical data in the virtual operation model to simulate the dynamic changes of environmental parameters and electrical parameters, and record the degradation trend of key performance parameters.

[0085] The specific implementation process of the above steps is as follows: Extract all environmental parameters and electrical parameters from the historical operation records, and calculate the maximum and minimum values of each parameter respectively as its fluctuation range.

[0086] Obtain the corresponding sensitive factors from the environmental fluctuation range and electrical fluctuation range according to the sensitive factors corresponding to each key performance parameter, and form a simulation group with these sensitive factors as independent variables, other non-sensitive factors as fixed quantities, and key performance parameters as dependent variables.

[0087] Under the above implementation example, assume that the key performance parameters are threshold voltage, on-resistance, and transconductance, and the sensitive factors of the threshold voltage are voltage and temperature, the sensitive factor of the on-resistance is voltage, and the sensitive factors of the transconductance are voltage and temperature. The simulation group formed at this time is shown in Table 2.

[0088] Table 2: Example of simulation group

[0089]

[0090] Define the minimum and maximum values within the fluctuation range corresponding to the sensitive factors in the simulation group combined with the safety margin as the boundaries of the simulation interval, and use the normal value within its fluctuation range as the simulation value for the fixed quantity of the simulation group.

[0091] Applied to the example in Table 2, assume that the fluctuation range of voltage in simulation group 1 is , where , respectively represent the minimum and maximum values within the corresponding fluctuation range of voltage. The safety margin is , then the boundaries of the simulation interval are , .

[0092] It should be understood that the main purpose of setting the safety margin is to ensure the safety and reliability of the simulation results and cover potential extreme working conditions. Specifically, the safety margin can provide an additional buffer space within the simulation interval to prevent model failure or prediction deviation caused by actual operating conditions exceeding the expected range. Increasing or decreasing a certain percentage of the safety margin based on the minimum and maximum values in the historical data can cover those extreme working conditions that may occur but are not recorded, thereby improving the robustness of the model. In short, by introducing the safety margin, the adaptability of the model to uncertainty and variability factors can be enhanced, ensuring that it can provide reliable prediction results under various working conditions. Exemplarily, a 5% safety margin can be set.

[0093] It should also be understood that for a fixed quantity, the normal value is selected as the simulation value because the fixed quantity has a relatively small impact on the key performance parameters. Using the normal value can, on the one hand, better represent the central tendency of the data, avoid deviations caused by individual extreme data points, and help improve the stability of the model; on the other hand, it can reflect the normal performance under actual working conditions, enabling the model prediction to be based on actual normal data.

[0094] In specific implementation, the normal value can be obtained according to the historical data of the fixed quantity, and its probability distribution graph (such as a histogram) can be drawn. If the data is normally distributed, the mean value can be selected as the normal value; if the data has skewness or outliers, the median value can be selected as the normal value.

[0095] A step size is set within the simulation interval for multiple simulation experiments, so that the sensitive factor gradually increases at the set step size within its simulation interval. The selection of the step size should comprehensively consider the computing resources and accuracy requirements. An overly small step size may lead to too high a computing cost, while an overly large step size may miss important degradation trends.

[0096] It should be added that the prerequisite for using historical data to simulate the dynamic changes of environmental parameters and electrical parameters is that the historical data should cover various typical working conditions and extreme situations. If the data only contains records under certain specific conditions, it may lead to incomplete simulation results. In addition, the time span of the data should be long enough to cover the performance of the equipment under different seasons and operating conditions.

[0097] During the dynamic simulation process of each simulation group, record the output results of the key performance parameters, and draw a degradation simulation curve based on these results.

[0098] The above-mentioned degradation simulation curve shows the trend of the key performance parameter changing with the sensitive factor.

[0099] For the degradation simulation curve of the key performance parameter corresponding to each simulation group, select a mathematical model to fit it, and use the fitted model to extrapolate the key performance parameter value at future time points as the future trend value of the key performance parameter value.

[0100] Specifically, when selecting a mathematical model for fitting, it is necessary to conduct data analysis and preliminary observation on the degradation simulation curve. Common degradation modes include linear, exponential, power function, polynomial, etc.

[0101] Linear degradation: If the change of the key performance parameter with time or the sensitive factor shows a linear relationship, a linear regression model (such as ) can be considered.

[0102] Exponential degradation: If the change rate of the key performance parameter increases or decreases with time, an exponential model (such as ) may be suitable.

[0103] Power function degradation: In some cases, the change of key performance parameters conforms to the power-law relationship, and a power function model (such as ) can be used.

[0104] Polynomial degradation: When the data shows complex non-linear changes, a polynomial regression model ( ) can be considered. Among them, represents sensitive factors, represents key performance parameters, , represents the parameters to be estimated.

[0105] Once a suitable mathematical model is selected and the fitting is completed, the model can be used to extrapolate the key performance parameter values at future time points. The basic principle of extrapolation is to substitute the future sensitive factors into the equation of the fitting model and calculate the corresponding key performance parameter values.

[0106] The specific extrapolation steps are as follows: Determine the specific values of sensitive factors at future time points according to requirements.

[0107] Substitute the sensitive factor values at different future time points into the equation of the fitting model and calculate the corresponding predicted values of key performance parameters.

[0108] Suppose an exponential degradation model is fitted through a degradation simulation curve to describe the change of the threshold voltage of a field-effect transistor with temperature. The specific fitting model is , where

[0109] is the temperature.

[0110] The threshold voltage prediction under this exponential degradation model is shown in Table 3.

[0111]

[0112] It can be seen from the table that as the temperature rises, the threshold voltage of the field-effect transistor shows a downward trend, reflecting the performance degradation phenomenon of the field-effect transistor at higher temperatures.

[0113] It should be added that in order to accurately extrapolate the key performance parameter values at future time points using the fitting model, the sensitive factor values at future time must be obtained first. The accuracy of these sensitive factor values directly determines the reliability of the extrapolation results. An effective method to obtain the sensitive factor values at future time is to analyze historical data, identify the trends, periodicity and seasonal changes in it, and establish an appropriate regression model to describe the relationship between sensitive factors and time.

[0114] Step 6. Predict the remaining life of the field effect transistor according to the degradation trend and generate a reliability assessment report.

[0115] Specifically, the remaining life prediction in the above solution is as follows: Set the failure threshold of the key performance parameters according to the technical specifications of the field effect transistor.

[0116] Use the fitting model to solve the time point when the key performance parameter reaches the failure threshold, which is used as the life aging time point corresponding to this parameter.

[0117] The above-mentioned life aging time point refers to the critical time point when the key performance parameter reaches failure, which reflects the termination of the aging of the key performance parameter.

[0118] Furthermore, the remaining life of the key performance parameter is obtained by combining the life aging time point of the key performance parameter with the current time.

[0119] For each life aging time point corresponding to the key performance parameter, calculate the confidence interval of the remaining life by combining the statistical distribution model.

[0120] In the preferred implementation of the above solution, the steps for calculating the confidence region of the remaining life are as follows: a. Select a suitable statistical distribution model (such as normal distribution, Weibull distribution) to describe the probability distribution of the remaining life.

[0121] b. Estimate the parameters (such as mean, standard deviation) of the selected distribution model based on historical data.

[0122] c. Calculate the confidence interval of the remaining life using the estimated parameters.

[0123] In the example of the above implementation, assume that the Weibull distribution is selected to describe the probability distribution of the remaining life, and its probability density function is , where represents time, is the scale parameter, is the shape parameter, represents the natural constant.

[0124] Use historical data to estimate and using the maximum likelihood estimation method, and calculate the confidence interval of the remaining life accordingly.

[0125] It should be noted that the calculation method of the confidence interval is well known in the industry and is part of the prior art. Therefore, the specific details will not be elaborated in the present invention.

[0126] It should be understood that the lifetime aging time point for identifying key performance parameters does give a specific prediction value, but relying solely on this single time point has certain limitations. Calculating the confidence interval of the remaining life has the following important functions: 1. Quantifying uncertainty: There is a certain degree of uncertainty in the data and models of actual systems, which may come from various factors such as measurement errors, inaccurate model assumptions, and environmental changes. By calculating the confidence interval, this uncertainty can be quantified. Understanding the range of uncertainty helps better conduct risk assessment and management. For example, if the confidence interval is wide, it indicates that the uncertainty of the prediction result is large, and more cautious maintenance plans need to be formulated.

[0127] 2. Improving prediction reliability: The prediction based on a single lifetime aging time point may be affected by outliers or extreme situations, while the confidence interval provides a range, making the prediction more robust. Even if some data points deviate, the overall prediction result still has a high credibility.

[0128] In the further implementation of the plan, the accelerated degradation time is predicted as follows: Draw a degradation trend curve based on the future trend values of the key performance parameter values.

[0129] Calculate the degradation rates at different time points on the drawn degradation trend curve of the key performance parameter.

[0130] By comparing the degradation rates at different time points, identify the time point with a significant increase in the degradation rate as the accelerated degradation time point.

[0131] It should be noted that the time point with a significant increase in the degradation rate is the time point when the device enters the accelerated degradation stage from the stable degradation stage, indicating that the device performance begins to decline rapidly. Identifying this time point can be used as a warning signal to prompt the maintenance personnel that maintenance will be required soon to avoid sudden failure.

[0132] In the further implementation of the plan, the life assessment report includes the following content: S1. The lifetime aging time points of the field-effect transistor under each key performance parameter, as well as the corresponding confidence intervals of the remaining life and the accelerated degradation time points.

[0133] S2. The arrangement results of the key performance parameters, and the specific sorting is as follows: S21. Assign weights to the key performance parameters based on the width of the confidence intervals of the remaining life of the field-effect transistor under each key performance parameter.

[0134] The specific weight assignment formula is , where represents the weight value of the th key performance parameter, represents the number of the key performance parameter, , Represents the width of the confidence interval of the remaining life of the field-effect transistor under the key performance parameters, and represents the longest confidence interval width among the widths of the confidence intervals of the remaining lives of all key performance parameters.

[0135] The principle of the above weight assignment formula is that the wider the confidence interval width, the greater the uncertainty of the prediction result, that is, our confidence in the authenticity of the prediction result is lower. In the formula, reflects the proportion of the width of the confidence interval of the remaining life of each key performance parameter. The larger this proportion, the greater the uncertainty of the prediction result, and the corresponding weight should be smaller. Therefore, the weight value obtained by subtracting this proportion from the numerical value 1 can better reflect the true importance and priority of each key performance parameter.

[0136] S22. Combine the life aging time points of each key performance parameter with its weight for importance calculation, and arrange them in descending order of importance.

[0137] In the implementation of the above solution, since the closer the life aging time point is, the greater the importance, it is necessary to appropriately process the life aging time point in the setting of importance calculation. For example, perform negative exponential processing on the life aging time point. Exemplarily, , represents the life aging time point, and after processing, perform importance scoring with the assigned weight value.

[0138] By sorting the key performance parameters as described above, it is possible to quickly identify which key performance parameters will reach the failure threshold in a short time. This helps to prioritize attention to those parameters that are about to fail, so as to take timely maintenance or replacement measures.

[0139] It should be emphasized that the present invention mainly performs single-factor prediction on the life of the field-effect transistor under different key performance parameters, can independently analyze the influence of each key performance parameter on the life, reduces the complexity of the model, makes the calculation more convenient, and can obtain the life prediction results of each key performance parameter in a short time, facilitating quick decision-making. More importantly, it can formulate a maintenance plan specifically, prioritize the treatment of those key performance parameters close to failure, and reduce unnecessary maintenance costs. Of course, multi-factor prediction can be introduced to provide a more comprehensive evaluation perspective. Specifically, the prediction model can be flexibly adjusted according to different application scenarios and requirements to provide customized solutions to meet diverse needs.

[0140] Embodiment 2

[0141] The present invention provides a reliability prediction device for a field-effect transistor, which device includes a reliability prediction method for a field-effect transistor according to the present invention.

[0142] Example 3

[0143] Refer to Figure 3 As shown, the present invention provides a reliability prediction system for a field effect transistor, including the following modules: Historical operation data retrieval module: Retrieve the historical operation records of the field effect transistor from the operation database, including the time series data of environmental parameters, electrical parameters, and performance parameters.

[0144] Key performance parameter identification module, connected to the historical operation data retrieval module, for determining the correlation between changes in environmental parameters or electrical parameters and changes in performance parameters through cross-correlation analysis of the historical operation records, and screening out the key performance parameters.

[0145] Virtual operation model construction module: Construct a virtual operation model based on the material characteristics and device structure of the field effect transistor. The input of the virtual operation model is environmental parameters and electrical parameters, and the output is performance parameters;

[0146] Model validity verification module, connected to the virtual operation model construction module, for inputting the time series data of environmental parameters and electrical parameters in the historical operation records into the virtual operation model, and verifying the validity of the model by comparing the performance parameters output by the model with the corresponding performance parameters in the historical operation records.

[0147] Degradation dynamic simulation module, respectively connected to the model validity verification module and the key performance parameter identification module, for simulating the dynamic changes of environmental parameters and electrical parameters in the virtual operation model using historical data when the model is verified to be valid, and recording the degradation trend of the key performance parameters;

[0148] Remaining life prediction module, connected to the degradation dynamic simulation module, for predicting the remaining life and accelerated degradation time of the field effect transistor according to the degradation trend, and generating a life assessment report.

[0149] The above content is only an example and explanation of the structure of the present invention. Those skilled in the art of the present technology can make various modifications or supplements to the described specific embodiments or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined by the present invention, they should all fall within the protection scope of the present invention.

Claims

1. A reliability prediction method for a field effect transistor, characterized in that It includes the following steps: Step 1: Retrieve the historical operation records of the field-effect transistor from the operating database, including the time-series data of environmental parameters, electrical parameters, and performance parameters; Step 2: Determine the correlation between the changes in environmental parameters or electrical parameters and the changes in performance parameters through cross-correlation analysis of the historical operation records, and screen out the key performance parameters; Step 3: Construct a virtual operation model based on the material characteristics and device structure of the field-effect transistor. The input of the virtual operation model is environmental parameters and electrical parameters, and the output is performance parameters; Step 4: Input the time-series data of environmental parameters and electrical parameters in the historical operation records into the virtual operation model, and verify the effectiveness of the model by comparing the performance parameters output by the model with the corresponding performance parameters in the historical operation records; Step 5: When the model is verified to be effective, use historical data in the virtual operation model to simulate the dynamic changes of environmental parameters and electrical parameters, and record the degradation trend of the key performance parameters; The specific process of Step 5 is as follows: Take the sensitive factors corresponding to each key performance parameter as independent variables, other non-sensitive factors as fixed quantities, and the key performance parameter as the dependent variable to form a simulation group; Define the minimum and maximum values within the fluctuation range of the sensitive factors in the simulation group combined with the safety margin as the boundaries of the simulation interval; Set a step size within the simulation interval so that the sensitive factors gradually increase at the set step size within their simulation intervals; During the dynamic simulation process of each simulation group, record the output results of the key performance parameters, and draw a degradation simulation curve based on these results; Select a mathematical model to fit the degradation simulation curve corresponding to the key performance parameter of each simulation group, and use the fitted model to extrapolate the key performance parameter values at future time points as the future trend values of the key performance parameter values; Step 6: Predict the remaining life and accelerated degradation time of the field-effect transistor according to the degradation trend, and generate a life assessment report; The specific process of Step 6 is as follows: The process of predicting the remaining life of the field-effect transistor is: Set the failure threshold of the key performance parameter according to the technical specifications of the field-effect transistor; Use the fitted model to solve the time point when the key performance parameter reaches the failure threshold as the life aging time point corresponding to this parameter; Calculate the confidence interval of the remaining life in combination with the statistical distribution model for each life aging time point corresponding to the key performance parameter; The process of predicting the accelerated degradation time is: Draw a degradation trend curve based on the future trend values of the key performance parameter values; Calculate the degradation rate at different time points on the drawn degradation trend curve of the key performance parameter; By comparing the degradation rates at different time points, identify the time point with a significant increase in the degradation rate as the accelerated degradation time point.

2. The reliability prediction method of a field effect transistor according to claim 1, characterized in that: The content of Step 2 is as follows: Plot the environmental parameter change curve, electrical parameter change curve, and performance parameter change curve in the coordinate system with time as the horizontal axis and environmental parameters, electrical parameters, and performance parameters as the vertical axis for the time-series data of environmental parameters, electrical parameters, and performance parameters in each historical operation record respectively; Identify the inflection points on the environmental parameter change curve, power parameter change curve, and performance parameter change curve in each historical operation record, and form an environmental parameter inflection point set, a power parameter inflection point set, and a performance parameter inflection point set with the timestamps corresponding to the identified inflection points; Calculate the cross-correlation coefficients between each performance parameter and environmental parameters, power parameters in the historical operation record by using the cross-correlation function for the environmental parameter inflection point set and power parameter inflection point set corresponding to each historical operation record respectively and the performance parameter inflection point set; Compare the cross-correlation coefficients between each performance parameter and environmental parameters, power parameters in each historical operation record with the set correlation coefficient threshold, and screen out the performance parameters with cross-correlation coefficients reaching the correlation coefficient threshold as associated performance parameters, and record the relevant environmental parameters or power parameters as sensitive factors; Classify the associated performance parameters in each historical operation record, count the occurrence proportion of the associated performance parameters, and compare it with the configured effective occurrence proportion to screen out the associated performance parameters reaching the effective occurrence proportion as key performance parameters.

3. The reliability prediction method of a field effect transistor according to claim 1, wherein: The process of verifying the effectiveness of the model by comparing the performance parameters output by the model with the corresponding performance parameters in the historical operation record is as follows: Compare the output performance parameters obtained when the time series data of environmental parameters and power parameters in each historical operation record are used as the input of the virtual operation model with the performance parameters of the corresponding historical operation record, and calculate the simulation error amount; Statistically analyze the simulation error amounts of each historical operation record to obtain the average simulation error and the standard deviation of the simulation error; Compare the average simulation error and the standard deviation of the simulation error with the set acceptable critical values respectively. If both the average simulation error and the standard deviation of the simulation error do not exceed the set acceptable critical values, the model is verified to be effective, otherwise the model is verified to be invalid.

4. The reliability prediction method of a field effect transistor according to claim 3, characterized in that: Step 4 also includes adjusting the accuracy of the virtual operation model when the model is verified to be invalid, and the specific process is as follows: In the comparison of the average simulation error and the standard deviation of the simulation error with the set acceptable critical values respectively, if only the average simulation error exceeds the acceptable critical value, increase the complexity of the virtual operation model; If only the standard deviation of the simulation error exceeds the acceptable critical value, introduce regularization in the virtual operation model; If both the average simulation error and the standard deviation of the simulation error exceed the acceptable critical value, increase the complexity of the virtual operation model and introduce regularization.

5. The reliability prediction method of a field effect transistor according to claim 1, characterized in that: The formation process of the simulation group is as follows: Extract all environmental parameters and electrical parameters from the historical operation record, and calculate the maximum and minimum values of each parameter respectively as its fluctuation range; Obtain the corresponding sensitive factors from the environmental fluctuation range and electrical fluctuation range according to the sensitive factors corresponding to each key performance parameter, and form a simulation group with these sensitive factors as independent variables, other non-sensitive factors as fixed quantities, and key performance parameters as dependent variables. For the fixed quantities of the simulation group, use the normal values within their fluctuation ranges as simulation values.

6. The reliability prediction method of a field effect transistor according to claim 1, characterized in that: The life assessment report includes the following contents: S1. The life aging time points of the field effect transistor under each key performance parameter, as well as the corresponding remaining life confidence intervals and accelerated degradation time points; S2. Arrangement results of key performance parameters, and the specific sorting is as follows: S21. Assign weights to key performance parameters based on the width of the confidence interval of the remaining life of the field effect transistor under each key performance parameter; S22. Calculate the importance by combining the life aging time points of each key performance parameter with its weight, and arrange them in descending order of importance.

7. A reliability prediction device for a field effect transistor, characterized in that: The device includes a reliability prediction method for a field effect transistor according to any one of claims 1-6.

8. A reliability prediction system for a field effect transistor, which is used to perform the steps in a reliability prediction method for a field effect transistor according to any one of claims 1-6, characterized in that, It includes the following modules: Historical operation data retrieval module: Retrieve the historical operation records of the field effect transistor from the operation database, including the time series data of environmental parameters, power parameters, and performance parameters; Key performance parameter identification module: Determine the correlation between the changes in environmental parameters or power parameters and the changes in performance parameters through cross-correlation analysis of the historical operation records, and screen out the key performance parameters; Virtual operation model construction module: Construct a virtual operation model based on the material characteristics and device structure of the field effect transistor. The input of the virtual operation model is environmental parameters and power parameters, and the output is performance parameters; Model validity verification module: Input the time series data of environmental parameters and electrical parameters in the historical operation records into the virtual operation model, and verify the validity of the model by comparing the performance parameters output by the model with the corresponding performance parameters in the historical operation records; Degradation dynamic simulation module: When the model is verified to be valid, use historical data to simulate the dynamic changes of environmental parameters and electrical parameters in the virtual operation model, and record the degradation trend of key performance parameters; Remaining life prediction module: Predict the remaining life and accelerated degradation time of the field effect transistor according to the degradation trend, and generate a life assessment report.

Images