Reliability analysis and service life evaluation method of in-service grid-control power device

Abstract

The invention relates to the technical field of reliability and life of power electronics, in particular to a reliability analysis and life evaluation method for an in-service grid-control power device. When the reliability of the in-service grid-control power device is analyzed, a plurality of electrical performance indexes collected by the grid-control power device under different reliability working conditions are mainly utilized for analysis, the change trend of each electrical performance index is analyzed, whether the change trend of the electrical performance indexes can meet the mechanism of the performance change trend or not is determined, and the reliability of the in-service grid-control power device is analyzed. The method comprises the following steps of: determining passing and not passing of the AEC-Q, selecting a vehicle-mounted motor driving application scene, converting a reliability test working condition and an application scene working condition by utilizing a reliability formula, and evaluating the service life of the in-service grid-control power device by taking the AEC-Q as a degradation evaluation standard according to the degradation degree of a performance degradation curve.

Description

Technical Field

[0001] The invention belongs to the technical field of power electronics reliability life, and in particular relates to a reliability analysis and life assessment method for an in-service gate-controlled power device. Background Art

[0002] As power electronic conversion devices, in-service gate-controlled power devices have the characteristics of simple circuit, small switching loss, fast switching speed, high input impedance, and easy driving. The long-term, stable, safe and efficient operation of in-service gate-controlled power devices is the guarantee of the smooth operation of power electronic devices. With the increase of on-off and off-off times of in-service gate-controlled power devices in use, some electrical performance index parameters of electronic components will change in a certain pattern, and the power and performance of the devices will decline accordingly, thus affecting the safe operation of the entire system. In particular, damage to power devices of key electronic equipment will lead to safety accidents such as vehicle emergency stops, causing casualties and economic losses. Accurately predicting the life of in-service gate-controlled power devices will help improve the operating efficiency of power electronic devices, improve the safety of electronic devices in production and life, and reduce losses caused by failures. Reliability analysis of power electronic switches is the core issue of evaluating the life of electronic equipment, and is crucial for companies that produce electronic equipment. Reliability analysis of in-service gate-controlled power devices is an important part of life assessment. Only by analyzing and modeling accelerated test data can the failure of electronic equipment be accurately predicted. Therefore, how to conduct reliability analysis and life assessment of in-service gate-controlled power devices is the key to solving the problem of predicting failures of electronic equipment in advance.

[0003] The maintenance and replacement of existing electronic equipment are generally based on the equipment instructions and the occurrence of failures, and it is impossible to evaluate the working life in advance. How to accurately evaluate the working life of electronic equipment and provide corresponding reliability indicators is of great significance to the production efficiency and safety of electronic equipment. Summary of the invention

[0004] The purpose of the present invention is to propose a reliability analysis and life assessment method for in-service gate-controlled power devices in view of the above applications and needs.

[0005] The technical solution adopted by the present invention is:

[0006] A reliability analysis and life assessment method for an in-service gate-controlled power device comprises the following steps:

[0007] Step 1: Perform various reliability tests on the gate-controlled power device to obtain operating condition data and electrical performance index parameters, and obtain the relationship between the operating condition data and the electrical performance index parameters;

[0008] Step 2: Based on the obtained relationship, the operating condition data is used as input and the electrical performance index parameters are used as output to train the neural network, and the BP neural network training method is used to train and establish a neural network model;

[0009] Step 3: According to the AEC-Q standard, constrain the actual operating condition data of the gate-controlled power device to be tested;

[0010] Step 4: Convert the reliability test working condition data and the actual working condition data after constraints according to the reliability formula to obtain the acceleration factor;

[0011] Step 5: Input the operating condition data of the in-service gate-controlled power device to be tested into the neural network model to obtain the predicted electrical performance index parameters, reduce the time scale according to the converted acceleration factor, fit the performance degradation curve with the reduced time and the predicted electrical performance index parameters, and use AEC-Q as the degradation evaluation standard to evaluate the life of the in-service gate-controlled power device to be tested;

[0012] Step 6: Perform reliability analysis on the in-service gate-controlled power device to be tested, and verify whether the performance degradation curve conforms to the performance degradation law of the in-service gate-controlled power device to be tested;

[0013] Further, the reliability test in step 1 includes a high temperature and high humidity reverse bias test, a highly accelerated stress test, a high temperature gate bias test, a high temperature reverse bias test, a high temperature storage test, and a temperature cycle test;

[0014] The defined working condition data include junction temperature T_j (℃), storage temperature T_stg (℃), test humidity RH (%RH), drain-source voltage V_ds (V), gate-source voltage V_gs (V) and test time Time (hrs);

[0015] The corresponding electrical performance index parameters include the threshold voltage V_TH250μA (V) when the leakage current is 250μA, the gate forward leakage current I_GSSF (A), the gate reverse leakage current I_GSSR (A), the drain cut-off current I_DSS50V (A) when the drain-source voltage is 50V, the drain cut-off current I_DSS80V (A) when the drain-source voltage is 80V, the drain cut-off current I_DSS100V (A) when the drain-source voltage is 100V, the drain-to-source voltage V_BVDS250μA (V) when the leakage current is 250μA, the drain-to-source voltage V_BVDS1mA (V) when the leakage current is 1mA, the on-resistance R_DON (Ω), and the forward conduction voltage drop V_FSD (V) of the diode;

[0016] The relationship between operating condition data and electrical performance index parameters is expressed as:

[0017] f(X1,X2,...,X6)=(Y1,Y2,...,Y10)

[0018] Among them, X1, X2, …, X6 represent the operating condition data T_j (℃), T_stg (℃), RH (%RH), V_ds (V), V_gs (V), and Time (hrs), respectively; Y1, Y2, …, Y10 represent the electrical performance index parameters V_TH250μA (V), I_GSSF (A), I_GSSR (A), I_DSS50V (A), I_DSS80V (A), I_DSS100V (A), V_BVDS250μA (V), V_BVDS1mA (V), R_DON (Ω), and V_FSD (V), respectively.

[0019] Furthermore, the neural network model established in step 2 is expressed as:

[0020] net (k) =W (k) Y (k-1) +b (k)

[0021] Y (k) =f (k) (net (k) )=[Y1 (k) ,Y2 (k) ,...,Y mk (k) ] T

[0022] Where k is the number of layers; net (k) is the net input of the kth layer; W (k) is the weight matrix of the kth layer; Y (k-1) is the output of the k-1th layer; b (k) is the bias vector of the kth layer; Y (k) is the output vector of the kth layer; f (k) is the activation function of the kth layer; Y i (k) is the output of the i-th neuron in the k-th layer; mk is the number of neurons in the current layer; T is the transpose.

[0023] Furthermore, in step 3, the actual operating condition data are constrained, specifically: T_j(℃)∈(-55,175), T_stg(℃)∈(-55,175), RH(%RH)∈(0,100), V_ds(V)∈(0,100), V_gs(V)∈(0,20), Time(hrs)∈(10000,200000).

[0024] Furthermore, the acceleration factors obtained in step 4 include temperature acceleration factor, humidity acceleration factor, and voltage acceleration factor:

[0025] Temperature acceleration factor AF T It is expressed as:

[0026]

[0027] Among them, a1 and b1 are unknown coefficients, satisfying a1+b1=1, T j 、T stg are the junction temperature and storage temperature under the reliability test conditions, T use is the temperature under actual use conditions, T use1 is the actual storage temperature, Ea = 0.7 eV is the activation energy, k = 8.617 × 10-5 eV / K is the Boltzmann constant;

[0028] Humidity Acceleration Factor AF RH It is expressed as:

[0029]

[0030] RH stress is the humidity under reliability test conditions, RH use is the humidity under actual use conditions, n=3~4 is the humidity acceleration index;

[0031] Voltage acceleration factor AF V It is expressed as:

[0032] AF V =a2·exp(bvds·(V ds,stress -V ds,operating ))+b2·exp(bvgs·(V gs,stress -V gs,operating ))

[0033] Where a2 and b2 are unknown coefficients, satisfying a2+b2=1, V ds,stress 、V gs,stress are the drain voltage and gate voltage under reliability test conditions, V ds,operating 、V gs,operating are the drain voltage and gate voltage under actual use conditions, bvds=0.05-0.1 is the acceleration coefficient related to V_ds(V), and bvgs=0.02-0.05 is the acceleration coefficient related to V_gs(V);

[0034] The final overall speed-up factor is:

[0035] AF total =AF T +AF RH+AF V

[0036] AF total is the overall acceleration factor.

[0037] Furthermore, the time scale reduction method in step 5 is:

[0038]

[0039] Where Time is the time predicted under actual working conditions, Time ex is the time under the converted reliability condition.

[0040] Furthermore, in step 6, a reliability analysis is performed on the in-service gate-controlled power device to be tested to verify whether the performance degradation curve conforms to the performance degradation law of the in-service gate-controlled power device to be tested. If the performance degradation law of the reliability analysis is consistent with the performance degradation law predicted by the neural network, it can indicate that the training data is accurate and the effect of life assessment is good; if the performance degradation law of the reliability analysis is inconsistent with the performance degradation law predicted by the neural network, it can indicate that the neural network training data is inaccurate, and it is necessary to re-collect the reliability test data to perform steps 1 to 6 until the performance degradation law of the reliability analysis is consistent with the performance degradation law predicted by the neural network, and further perform a graded evaluation on the performance degradation curve.

[0041] According to the AEC-Q evaluation standard, a degradation mechanism table is developed, as shown in Table 1, to determine the classification and evaluate the life of the in-service gate-controlled power device.

[0042] Table 1 Performance parameters and life evaluation

[0043]

[0044] Among them, PB means that the device is in excellent health when the performance parameters change within 5%, PS means that the device is in good health when the performance parameters change within 5% to 10%, ZO means that the device is in medium health when the performance parameters change within 10% to 15%, NS means that the device is in poor health when the performance parameters change within 15% to 20%, and NB means that the device is in extremely poor health when the performance parameters change within 20% to 25%. If the change exceeds 15%, the power device must be replaced.

[0045] The beneficial effects of the present invention are as follows: the present invention uses a neural network model to predict and convert into an electrical performance index parameter degradation curve to perform accurate life assessment, providing data support for the risk analysis of power device failure. The existing reliability test life assessment prediction method lacks accurate quantitative indicators and cannot use data to intuitively display the reliability and life of the corresponding period; the present invention uses a reliability test method to perform life assessment on its in-service gate-controlled power devices, thereby predicting the occurrence of power electronic equipment failures. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A flow chart of a reliability analysis and life assessment method for an in-service gate-controlled power device provided by an embodiment of the present invention;

[0047] Figure 2 A performance degradation curve of an in-service gate-controlled power device taking V_TH250μA (V) as an example is provided for the embodiment of the present invention. DETAILED DESCRIPTION

[0048] The technical solution of the present invention is described in detail below with reference to the accompanying drawings.

[0049] like Figure 1As shown, the reliability analysis and life assessment method of in-service gate-controlled power devices provided by the embodiment of the present invention includes the following steps: first, the performance of 6*77 in-service gate-controlled power device samples is preliminarily tested and analyzed, and the samples are subjected to high temperature and high humidity reverse bias test (H3TRB), highly accelerated stress test (HAST), high temperature gate bias test (HTGB), high temperature reverse bias test (HTRB), high temperature storage test (HTS), and temperature cycle test (TC) reliability tests. Furthermore, the reliability test condition data T_j(℃), T_stg(℃), RH(%RH), V_ds(V), V_gs(V), Time(hrs) and electrical performance index parameter data V_TH250μA(V), I_GSSF(A), I_GSSR(A), I_DSS50V(A), I_DSS80V(A), I_DSS100V(A), V_BVDS250μA(V), V_BVDS1mA(V), R_DON(Ω), V_FSD(V) were collected and collated. Furthermore, a neural network model was established based on the collected data. Furthermore, according to the AEC-Q standard, the operating conditions are constrained as T_j(℃)∈(-55,175), T_stg(℃)∈(-55,175), RH(%RH)∈(0,100), V_ds(V)∈(0,100), V_gs(V)∈(0,20), and Time(hrs)∈(10000,200000). Furthermore, the Arrhenius, Eyring, and Peck models are used to convert the reliability test conditions to the actual conditions. Furthermore, the performance degradation curve is generated. The collected data is further used to draw a variability diagram (used to show the distribution and fluctuation of data to help understand the degree of variation in the data set or between variables) to estimate the trend of changes in electrical performance index parameters, and then a reasonable explanation for the changes in electrical performance index parameters is given from the semiconductor physics level (electron concentration, carrier mobility, drift) and the test process level (5M+E) to verify whether the performance degradation curve conforms to the performance degradation law of the in-service gate-controlled power device to be tested. After the performance degradation curve conforms to the performance degradation law of the in-service gate-controlled power device, the life of the in-service gate-controlled power device is evaluated based on the degree of degradation of the converted performance degradation curve and AEC-Q as the degradation evaluation standard.

[0050] Example:

[0051] This example performs reliability analysis and life evaluation on the CRSG022N10N4Q automotive-grade SGT MOS. A reliability test was carried out on CRSG022N10N4Q, and then the data were sorted and summarized to form a mapping relationship of (X1, X2, …, X6) → (Y1, Y2, …, Y10), that is, a mapping relationship of (T_j (℃), T_stg (℃), RH (%RH), V_ds (V), V_gs (V), Time (hrs)) → (V_TH250μA (V), I_GSSF (A), I_GSSR (A), I_DSS50V (A), I_DSS80V (A), I_DSS100V (A), V_BVDS250μA (V), V_BVDS1mA (V), R_DON (Ω), V_FSD (V)); further, a neural network model was constructed based on these data, which is equivalent to a "black box". When the working conditions are input, the output electrical performance index parameters can be predicted. Furthermore, according to the AEC-Q standard, the operating conditions are constrained within the range of the neural network modeling parameters, and the operating conditions are constrained to be T_j(℃)∈(-55,175), T_stg(℃)∈(-55,175), RH(%RH)∈(0,100), V_ds(V)∈(0,100), V_gs(V)∈(0,20), and Time(hrs)∈(10000,200000); further, the initial values ​​of the reliability conversion parameters are roughly estimated, and the intermediate values ​​are selected within the range for calculation. The Arrhenius model selects a1=0.7, b1=0.3, and T j =150, T use =80, T use1 =80, where junction temperature is the main factor, AF is calculated according to formula 4 T =45.07, Eyring model selects RH stress =80, RH use =50, n=3, AF is calculated according to formula 5 RH =4.91, Peck model selects a2=0.7, b2=0.3, V ds,stress =100, V gs,stress =20, V ds,operating =48, V gs,operating =12, bvds=0.08, bvgs=0.03, where the drain voltage is the main factor, AF is calculated according to formula 6 V =45.23, and then add up the various acceleration factors to get the overall conversion acceleration factor AF total =95.21; When the input working condition is (80℃, 80℃, 50%RH, 48V, 12V, 5000hrs), if Figure 2As shown, the performance index degradation curve of V_TH250μA (V) is obtained as an example. Further, the reliability analysis of the gate-controlled power device to be tested is analyzed to analyze whether the change trend conforms to the semiconductor physical level and the test process level to give a reasonable explanation for the change of the electrical index parameters; further, when the performance degradation curve conforms to the performance degradation law of the gate-controlled power device to be tested, the evaluation mechanism of failure when the performance index parameter changes by more than 15% is used to determine the horizontal coordinate time of the 15% change point corresponding to the performance degradation curve. The vertical starting point is 2.8, and the curve is an upward trend. When V_TH250μA (V) reaches 3.22, the corresponding horizontal coordinate is 33800 hours, that is, 3.86 years of failure, so as to predict and evaluate the device life.

Claims

1. A reliability analysis and life assessment method for an in-service gate-controlled power device, characterized in that: The following steps are involved: Step 1: Perform various reliability tests on the gate-controlled power device to obtain operating condition data and electrical performance index parameters, and obtain the relationship between the operating condition data and the electrical performance index parameters; Step 2: Based on the obtained relationship, the operating condition data is used as input and the electrical performance index parameters are used as output to train the neural network, and the BP neural network training method is used to train and establish a neural network model; Step 3: According to the AEC-Q standard, constrain the actual operating condition data of the gate-controlled power device to be tested; Step 4: Convert the reliability test working condition data and the actual working condition data after constraints according to the reliability formula to obtain the acceleration factor; Step 5: Input the operating condition data of the in-service gate-controlled power device to be tested into the neural network model to obtain the predicted electrical performance index parameters, reduce the time scale according to the converted acceleration factor, fit the performance degradation curve with the reduced time and the predicted electrical performance index parameters, and use AEC-Q as the degradation evaluation standard to evaluate the life of the in-service gate-controlled power device to be tested; Step 6: Perform reliability analysis on the in-service gate-controlled power device to be tested to verify whether the performance degradation curve conforms to the performance degradation law of the in-service gate-controlled power device to be tested.

2. The reliability analysis and life assessment method of an in-service gate-controlled power device according to claim 1, characterized in that: The reliability test described in step 1 includes a high temperature and high humidity reverse bias test, a highly accelerated stress test, a high temperature gate bias test, a high temperature reverse bias test, a high temperature storage test, and a temperature cycle test; The defined working condition data include junction temperature T_j (℃), storage temperature T_stg (℃), test humidity RH (%RH), drain-source voltage V_ds (V), gate-source voltage V_gs (V) and test time Time (hrs); The corresponding electrical performance index parameters include the threshold voltage V_TH250μA (V) when the leakage current is 250μA, the gate forward leakage current I_GSSF (A), the gate reverse leakage current I_GSSR (A), the drain cut-off current I_DSS50V (A) when the drain-source voltage is 50V, the drain cut-off current I_DSS80V (A) when the drain-source voltage is 80V, the drain cut-off current I_DSS100V (A) when the drain-source voltage is 100V, the drain-to-source voltage V_BVDS250μA (V) when the leakage current is 250μA, the drain-to-source voltage V_BVDS1mA (V) when the leakage current is 1mA, the on-resistance R_DON (Ω), and the forward conduction voltage drop V_FSD (V) of the diode; The relationship between operating condition data and electrical performance index parameters is expressed as: f(X1,X2,...,X6)=(Y1,Y2,...,Y10) Among them, X1, X2, …, X6 represent the operating condition data T_j (℃), T_stg (℃), RH (%RH), V_ds (V), V_gs (V), and Time (hrs), respectively; Y1, Y2, …, Y10 represent the electrical performance index parameters V_TH250μA (V), I_GSSF (A), I_GSSR (A), I_DSS50V (A), I_DSS80V (A), I_DSS100V (A), V_BVDS250μA (V), V_BVDS1mA (V), R_DON (Ω), and V_FSD (V), respectively.

3. The reliability analysis and life assessment method of an in-service gate-controlled power device according to claim 2, characterized in that: The neural network model established in step 2 is expressed as: net (k) =W (k) Y (k-1) +b (k) Y (k) =f (k) (net (k) )=[Y1 (k) ,Y2 (k) ,...,Y mk (k) ] T Where k is the number of layers; net (k) is the net input of the kth layer; W (k) is the weight matrix of the kth layer; Y (k-1) is the output of the k-1th layer; b (k) is the bias vector of the kth layer; Y (k) is the output vector of the kth layer; f (k) is the activation function of the kth layer; Y i (k) is the output of the i-th neuron in the k-th layer; mk is the number of neurons in the current layer; T is the transpose.

4. The reliability analysis and life assessment method of an in-service gate-controlled power device according to claim 3, characterized in that: In step 3, the actual operating condition data are constrained, specifically: T_j(℃)∈(-55,175), T_stg(℃)∈(-55,175), RH(%RH)∈(0,100), V_ds(V)∈(0,100), V_gs(V)∈(0,20), Time(hrs)∈(10000,200000).

5. The reliability analysis and life assessment method of an in-service gate-controlled power device according to claim 4, characterized in that: The acceleration factors obtained in step 4 include temperature acceleration factor, humidity acceleration factor, and voltage acceleration factor: Temperature acceleration factor AF T It is expressed as: Among them, a1 and b1 are unknown coefficients, satisfying a1+b1=1, T j , T stg are the junction temperature and storage temperature under the reliability test conditions, T use is the temperature under actual use conditions, T use1 is the actual storage temperature, Ea = 0.7 eV is the activation energy, k = 8.617 × 10-5 eV / K is the Boltzmann constant; Humidity Acceleration Factor AF RH It is expressed as: RH stress is the humidity under reliability test conditions, RH use is the humidity under actual use conditions, n=3~4 is the humidity acceleration index; Voltage acceleration factor AF V It is expressed as: AF V =a2·exp(bvds·(V ds,stress -V ds,operating ))+b2·exp(bvgs·(V gs,stress -V gs,operating )) Where a2 and b2 are unknown coefficients, satisfying a2+b2=1, V ds,stress 、V gs,stress are the drain voltage and gate voltage under reliability test conditions, V ds,operating 、V gs,operating are the drain voltage and gate voltage under actual use conditions, bvds=0.05-0.1 is the acceleration coefficient related to V_ds(V), and bvgs=0.02-0.05 is the acceleration coefficient related to V_gs(V); The final overall speed-up factor is: OF total =OFF T +AF RH +AF V AF total is the overall acceleration factor.

6. The reliability analysis and life assessment method of an in-service gate-controlled power device according to claim 5, characterized in that: The time scale reduction method in step 5 is: Where Time is the time predicted under actual working conditions, Time ex is the time under the converted reliability condition.

7. The reliability analysis and life assessment method of an in-service gate-controlled power device according to claim 6, characterized in that: In step 6, a reliability analysis is performed on the in-service gate-controlled power device to be tested to verify whether the performance degradation curve conforms to the performance degradation law of the in-service gate-controlled power device to be tested. If the performance degradation law of the reliability analysis is consistent with the performance degradation law predicted by the neural network, it can indicate that the training data is accurate and the life assessment effect is good; if the performance degradation law of the reliability analysis is inconsistent with the performance degradation law predicted by the neural network, it can indicate that the neural network training data is inaccurate, and it is necessary to re-collect the reliability test data to perform steps 1 to 6.

Images