Life and Reliability Evaluation of High-Power LED under Thermal Environment

Transcription

1 J Jpn Ind Manage Assoc 67, , 2016 Original Paper Life and Reliability Evaluation of High-Power LED under Thermal Environment Yao HSU 1, Wen-Fang WU 2, Ching-Cheng ZOU 3 Abstract: Finite element analysis (FEA) is frequently employed by researchers to investigate the mechanics and/or thermal behaviors of high-power LEDs. However, after performing FEA, only a few people continue to discuss the life and reliability of the LED. In this study, a relationship between the junction temperature and the life of a LED is first established based on real test data. FEA is then employed to find the junction temperature of the LED. By using the numerical results of FEA as the input for the relationship mentioned above, one can predict the life of the LED. However, the life is a fixed value under a certain condition, which cannot truly reflect the discrete characteristic in real life testing. Furthermore, it cannot provide extra information such as the reliability and the failure rate of the LED. To the end, this study further accounts for uncertainties coming from parameters such as convection coefficient and photoelectric conversion efficiency, and regards them as random variables. The Monte-Carlo method is used to simulate samples of these random variables when performing FEA and predicting life of the LED. Scattered lives indicating a random sample out of the studied LED are then obtained even under the same temperature and environmental condition. By using the probability plot and statistical analysis, one can find the life distribution and reliability-related quantities of the LED. Key words: High-Power LED, Junction Temperature, Parameter Uncertainty, Life Distribution, Reliability 1 INTRODUCTION Due to excellent luminescent capabilities, long life, low power consumption and wide range of applications, the high-power light emitting diode (LED) has been considered to be the future star of the lighting industry. Although the total amount of heat generated by the LED is not high, the heat flux per unit volume is quite high and should be paid much attention. If the heat generated by the LED fails to be dissipated efficiently into ambient environment, it leads to an increase in junction temperature between components that might damage the components, and consequently shorten the life expectancy of the LED and reduce their reliability. Therefore, the thermal management has always been an important issue in the design of LEDs. Many researches about LEDs have been conducted. Narendran and Guc [1] carried out an experiment on one 1 Department of Business and Entrepreneurial Management, Kainan University, Taiwan 2 Department of Mechanical Engineering and Institute of Industrial Engineering, National Taiwan University, Taiwan 3 Department of Mechanical Engineering, National Taiwan University, Taiwan Received: December 15, 2014 Accepted: December 22, 2015 lot of LEDs to observe their thermal behaviors and lives when subjected to different ambient temperatures. It was found that the luminous efficiency decreases exponentially with the use time. Arik et al. [2] conducted a thermal analysis with ANSYS finite element software for two kinds of LED chips made of SiC and sapphire, and compared their results. Chi et al. [3] conducted an experiment to measure the junction temperature and luminous intensity of LEDs under long duration use, and investigated how the geometric dimensions and material properties of LED packages affected the junction temperature and luminous intensity by simulation. Trevisanello et al. [4] performed an accelerated life testing for LEDs and found that large input current would produce a large amount of heat in LEDs and raise their the junction temperature which eventually lead to damage. Su et al. [5] analyzed the junction temperature and temperature of the heat sink of a high-power LED by finite element method and by experiment, and made a comparison of the results. The comparison showed a good agreement with each other. The results also suggested that the junction temperature is critical to performance of high-power LED modules. In addition, some researchers [6-7] claimed that a rise of the Vol.67 No.2E (2016) 181

2 junction temperature adversely impact much on the reliability of a high-power LED. From above literature reviews, it is noted that most of the studies focus on how the junction temperature of a LED influences its function and life, research work concerning the reliability of LEDs is rarely seen. In particular, the past studies were mostly limited to find a fixed value for the life of a LED package when adopting the simulation approach, which cannot truly reflect the scattered values phenomenon in life testing in practice. Furthermore, they cannot provide valuable information such as mean time to failure (MTTF) and the failure rate of the LED. With this in mind, this study considers some uncertainties coming from parameters such as convection coefficient and photoelectric conversion efficiency and regards them as random variables, which can be simulated by the Monte-Carlo method. The sample data is then applied to the FEA for evaluating life of the LED package. By using the probability plot, the probability life distribution as well as information such as reliability and failure rate of the LED package can be obtained. (5) Silicone sub-mount: area of mm 2, thickness of 0.13 mm. (6) Copper slug: top radius of 1 mm, bottom radius of 3 mm, thickness of 2.5 mm. (7) Silicone encapsulation: inner radius of 2.6 mm, outer radius of 2.85 mm, thickness of mm. (8) Plastic lens: radius of 2.6 mm. (9) Outer package: top width of 1.4 mm, bottom width of 0.5 mm, thickness of 2.25 mm. (10) Adhesive layer: radius of 3 mm, thickness of 0.1 mm. (11) Aluminum heat sink: area of mm 2, thickness of 1.5 mm. The model was meshed with 61,178 nodes and 22,331 elements, as shown in Fig FINITE ELEMENT ANALYSIS The present study adopted finite element software ANSYS to analyze the thermal behavior of the LED, especially the junction temperature distribution between components. When taking the uncertainties on parameters into consideration, the probability design system (PDS) module provided by ANSYS is further employed together with ANSYS. Statistical procedures will be used to process the random life data, and some reliability concepts and techniques will be applied as well. Fig. 1 One-quarter schematic of HP LED. 2.1 Model The simplified schematic diagram of the LED studied in this paper is illustrated in Fig. 1 [5, 8]. To save the simulation time, the model was constructed quartersymmetrically with appropriate boundary conditions settings reflecting the symmetrical characteristics. The geometric dimensions of each component are stated as follows: (1) GaN chip: area of 1 1 mm 2, thickness of 0.01 mm. (2) Phosphor: area of 1 1 mm 2, thickness of 0.34 mm. (3) Sapphire: area of 1 1 mm 2, thickness of 0.09 mm. (4) Die attach: area of 1 1 mm 2, thickness of mm. Fig. 2 Meshes of finite element model. 2.2 Material Property For thermal analysis, the thermal conductivity is the only material constant needed to be set and input into the ANSYS. The values of each component are tabulated in Table J Jpn Ind Manage Assoc

3 Table 1 Coefficient of thermal conductivity [8,10-12]. Component GaN chip Phosphor Sapphire Die attach Silicone sub-mount Copper slug Silicone encapsulation Plastic lens Outer package Adhesive layer Aluminum heat sink Thermal conductivity (W/m C) Boundary and Loadings No settings are needed to impose on the symmetrical surface when conducting thermal analysis. However, for the outer surface which contacts with air, heat flux and heat convection coefficient must be set upon. In the present study, the heat flux was set to be zero and the heat convection coefficient was imposed as 10W/m 2 C [12-14]. The emissivity of radiation of 0.9 and 0.7 were set on the package and the aluminum heat sink, respectively [5]. In addition, since the ambient temperature is very critical to the thermal behavior of LEDs, three kinds of ambient temperatures, 45 C, 55 C and 65 C, were considered to investigate their influence on the LED. Moreover, based on the practice at present, the photoelectric conversion efficiency of the LED is about 20%~30% [8, 13-14], determined as 25% in this study, which indicates 75% of the electric power energy will be dissipated as heat. In the present study, the electric power energy was set to be 3 watts and then 3 watts 75% was applied as the heat source on the GaN chip. 3 LIFE PREDICTION RULE In the finite element analysis, the temperature field inside the LED can eventually be obtained. However, the lifetime of the LED cannot be known unless having some certain life prediction rule. In the present study, the junction temperature was chosen to predict the lifetime of the LEDs since CREE, Inc. has published the testing data regarding the junction temperatures and lives of its produced high-power LEDs under varying ambient temperatures (as shown in Fig. 3). Based on the shapes of the curves, which similarly behave like an exponential function, we constructed the exponentially curve-fitted equations relating the life and junction temperature. They are expressed as follows: Ambient temperature 45 C B45T L A45e = (1) Ambient temperature 55 C B55T L A55e = (2) Ambient temperature 65 C B65T L A65e = (3) where L is the LED life, T is the junction temperature. By curve fitting technique, the values of the coefficients of the three equations estimated by mathematical software are A = 386, and B 45 = ; A = 370, and B 55 = ; and A 65 = 360, 000 and B 65 = , respectively. Once having the simulated junction temperature of the LED from finite element analysis, introducing it into the equation, the lifetime of the LED can then be predicted. Fig. 3 Testing data of mean life vs. junction temperature [9]. 4 SIMULATION RESULTS 4.1 Deterministic Life One of the most important results of thermal transferring analysis is temperature contour on the model. Figure 4 is the simulated temperature contour of the LED subject to the ambient temperature of 45 C. From Fig. 4, it can be seen that the highest temperature is located in the die attach component, which suggests the thermally weakest region in the whole LED. The distribution of the temperature on the LED is qualitatively and quantitatively consistent with those published by other studies [2, 5, 9] which validates the finite element model and analysis. The cases for ambient temperatures of 55 C and 65 C have temperature contours similar to that for ambient temperature of 45 C. The calculated junction temperatures are C, C and C under the conditions of ambient temperatures of 45 C, 55 C and 65 C, respectively. With the help of Eq.(1)~Eq.(3), the Vol.67 No.2E (2016) 183

4 corresponding life of LEDs with respect to each ambient temperature can then be obtained. They are 45,297 hrs, 37,406 hrs and 30,390 hrs, respectively. Fig. 4 Temperature contour of LED when ambient temperature is 45 C. 4.2 Random Consideration As stated in many articles, the degree of heat convection has an influence on the junction temperature of the LED, and therefore will affect their life expectancy. Moreover, there are inevitable measuring errors while operating the integrating sphere to determine the luminous efficiency of the LED, which leads to variability on photoelectric conversion efficiency. To this end, both of the heat convection coefficient and the photoelectric conversion efficiency were regarded as random variables in the present study. They were assumed to follow the normal distribution. The mean value of each normal distribution is the nominal heat convection coefficient or the nominal photoelectric conversion efficiency. The standard deviation of each normal distribution was determined by setting the coefficient of variation (C.O.V.) to be 5% (as shown in Table 2). The Monte-Carlo method was adopted to simulate the sample data for these parameters with a sample size of 200. The 200 parameter set were then applied into the ANSYS PDS and finite element analysis was performed 200 times. By using the analytical procedure stated in the previous section, 200 lives were then obtained. They range from 42,050 to 50,800 hrs when the ambient temperature is 45 C. Similarly, lives ranging from 34,700~42,000 hrs and 28,150~34,200 hrs were obtined for cases of ambient temperatures of 55 and 65 C, respectively. this study, four probability density functions (normal, lognormal, two-parameter and three-parameter Weibull) were selected. The graphical results, with the case of ambient temperature 45 C, are shown in Fig. 5, which suggests that normal, lognormal and three-parameter Weibull distributions fit the data better than two-parameter Weibull. To further assure the fitting and determine which probability density function can fit the life data best, the Anderson-Darling test was adopted to statistically test the goodness-of-fit. The testing results show that, at the significance level of 0.05, normal, lognormal and threeparameter Weibull can all pass the goodness-of-fit testing and three-parameter Weibull distribution has the smallest test statistics of the Anderson-Darling test. In addtion, three-parameter Weibull has the highest correlation coefficient (0.998) among the other distributions. Therefore, the life data fit the three-parameter Weibull function best. The results also hold for the cases of ambient temperatures 55 C and 65 C. Once the probability density function of the life data is determined, its corresponding reliability function and failure rate function can be obtained easily. Fig. 6 illustrates the corresponding reliability and failure rate functions of the LED subject to ambient temperature of 45 C with respect to three-parameter Weibull probability density function. Reliability and failure rate of the LED subject to three ambient temperatures conditions are compared and depicted as shown in Fig. 7 and Fig. 8, respectively. It is observed that the higher the ambient temperature the LED subject to, the higher the failure rate and the lower the reliability at the same operating time. Table 2 Variation of parameters with C.O.V. of 5% [5, 9-11]. Parameter Mean S.D. Photoelectric conversion efficiency Heat convection coefficient with (W/m 2 ) Reliability Analysis Probability plot analysis is frequently used to preliminarily understand what kinds of probability density functions they are and how well they can fit the data. In Fig. 5 Probability plot of normal, lognormal and Weibull when ambient temperature is 45 C. 184 J Jpn Ind Manage Assoc

5 junction temperature and the mean value or the standard deviation can be constructed as expressed in Eq. (4) and Eq. (5). Ae BT j µ 5% = (4) = Ce DT j σ (5) 5% Fig. 6 Probability density, reliability and failure rate functions of three-parameter Weibull when ambient temperature is 45 C. Reliability Percent Ti me (hr) Fig. 7 Comparison of reliability for different ambient temperatures. Rate Time (hr) Fig. 8 Comparison of failure rate for different ambient temperatures. In addition, during the goodness-of-fit analysis to threeparameter Weibull distribution, the corresponding mean and standard deviation of the LED life can be obtained. When the ambient temperature is 45 C, 55 C and 65 C, the mean lives are 45,693 hrs, 37,737 hrs and 30,665 hrs, respectively, and 1,760 hrs, 1,472 hrs and 1,224 hrs are for the standard deviation. With these data and their associated junction temperatures, relationship between the where T j is the junction temperature, µ is the mean of 5% the LED life when the C.O.V. of the heat convection coefficient and the photoelectric conversion efficiency are 5%, and σ is the standard deviation of the LED life 5% when the C.O.V. of the heat convection coefficient and the photoelectric conversion efficiency are 5%. The coefficients determined by least squares method 6 are A = 10, B = , C = 31, 150 and D = With the help of above two equations, one can quickly get the mean and standard deviation of the life of the LED when considering the 5% uncertainty of the heat convection coefficient and the photoelectric conversion once if the junction temperature is known. 5 CONCLUSIONS The finite element analysis is employed frequently to predict the physical or thermal behaviors of LED packages. The numerical results from FEA are usually deterministic. However, real experimental data are not deterministic, but scattered to certain degrees. The discrepancy may be attributed to that uncertainties in real situations are not taken into consideration during FEA. To better simulate real situations, the present study considers uncertainties of the heat convection coefficient and the photoelectric conversion efficiency when employing the finite element method to quantitatively analyze the reliability of LEDs. Under assumptions made in the present paper, several conclusions can be drawn. (1) When the ambient temperatures are 45 C, 55 C and 65 C, the lives of the HP LED considered herein without taking parameter variability into account are 45,297 hrs, 37,406 hrs and 30,390 hrs, respectively. The life decreases as the ambient temperature increases. The orders of these values are consistent with those proposed in other literature, which validates our finite element model. (2) By considering uncertainties with C.O.V. of 5%, it is shown that the fittest distribution to life data of the LED is the three-parameter Weibull. The position Vol.67 No.2E (2016) 185

6 parameters of the three-parameter Weibull distribution are 41,380 hrs, 34,132 hrs and 30,666 hrs when subject to the ambient temperatures of 45 C, 55 C and 65 C, respectively which means that the LED will not fail when the operating time is less than these values. (3) It is found that the failure rate of LEDs increases over time, which indicates the wear-out phase according to the bathtub curve in reliability engineering. This phenomenon suggests the aging of the LED package when subject to the thermal loading imposed in this study. (4) The methodology and analytical procedure described in this study have been proved for their validation and efficiency, and could be used as a reference for LED designers. ACKNOWLEDGMENTS This work was supported by the National Science Council of Taiwan, R.O.C. under Grant No. NSC E MY2. The authors appreciate this financial support. REFERENCES [1] Narendran, N. and Gu, Y.: Life of LED-Based White Light Sources, IEEE/OSA J. of Display Technol., Vol. 1, No. 1, pp (2005) [2] Arik, M., Becker, C., Weaver, S. and Petroski, J.: Thermal Management of LEDs: Package to System, Proceedings of the 3rd International Conference on Solid State Lighting, pp (2004) [3] Chi, W. H., Chou, T.L., Han, C. N., Yang, S. Y. and Chiang, K. N.: Analysis of Thermal and Luminous Performance of MR-16 LED Lighting Module, IEEE Trans. Components Packag. Technol., Vol. 33, No. 4, pp (2010) [4] Trevisanello, L., Meneghini, M., Mura, G., Vanzi, M., Pavesi, M., Meneghesso, G. and Zanoni, E.: Accelerated Life Test of High Brightness Light Emitting Diodes, IEEE Trans. Device and Mater. Reliab., Vol. 8, No. 2, pp (2008) [5] Su, Y. F., Yang, S. Y., Chi, W. H. and Chiang, K. N.: Light Degradation Prediction of High Power Light Emitting Diode Lighting Modules, EuroSimE (2010) [6] Bera, S. C., Singh, R. V. and Garg, V. K.: Temperature Behavior and Compensation of Light- Emitting Diode, IEEE Photonics Technol. Lett., Vol. 17, No. 11, pp (2005) [7] Kim, L., Choi, J. H., Jang, S. H. and Shin, M. W.: Thermal Analysis of LED Array System with Heat Pipe, Thermochim. Acta, Vol. 455, No. 1-2, pp (2007) [8] Hou, L. X.: Temperature and Thermal Stress Distributions of High Power White Light Emitting Diodes, National Sun Yat-Sen University, Master Thesis (2011) [9] [10] Hsu, Y. C.: Failure Mechanisms Associated with Lens Shape of High-Power LED Modules in Aging Test, IEEE Trans. Electron Devices, Vol. 55, No. 2, pp (2008) [11] Wang, J., Tsai, C. C., Cheng, W. C., Chen, M. H., Chung, C. H. and Cheng, W. H.: High Thermal Stability of Phosphor Converted White-Light Emitting Diodes Employing Ce:YAG Doped Glass, IEEE J. Quantum Electron., Vol. 17, pp (2011) [12] Liu, D. and Yang, D. G.: Reliability Study on High Power LED with Chip on Board, th International Conference on Electronic Packaging Technology and High Density Packaging, Shanghai, China, Aug. pp (2011) [13] Hou, F. and Yang, D.: Thermal Transient Analysis of LED Array System with In-line Pin Fin Heat sink, 12th Thermal, Mechanical and Multiphysics and Experiments in Microelectronics and Microsystems (2011) [14] Tang, H., Yang, D. G., Zhang, G. Q., Hou, F., Cai, M. and Cui, Z.: Multi-physics Simulation and Reliability Analysis for LED Luminaires under Step Stress Accelerated Degradation Test, EuraSimE (2012) 186 J Jpn Ind Manage Assoc