Open Access
Add to BibSonomy
Abstract
We propose a non-contact measurement method for determining two-dimensional (2D) temperature distribution of light-emitting diodes (LEDs). This method is based on both micro-hyperspectral imaging technology and reflected light method, owning merits of both high efficiency and high spatial resolution. Blue and green bare LEDs are used as LED under test, while red and near-infrared LEDs provide incident light to avoid spectral overlapping so as to reduce measurement error. During data processing, the convolution linear filtering algorithm is employed to improve the measurement accuracy. This proposed method is compared with the micro-thermocouple and infrared thermal imaging, with their respective comparison results in fairly good agreements. For spatial resolution of 2D temperature distribution, this method increases at least one order of magnitude compared with the thermal imaging method.
© 2019 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
1. Introduction
With the improvement in performance of light-emitting diodes (LEDs), they are increasingly used as healthy general lights, communication signal, indicator lights, and display backlights [1–3]. Temperature is a significant parameter in evaluating the performance of LEDs, exerting significant influence on emission wavelength, luminous efficacy, quantum efficiency, phosphor efficiency, and reliability of LEDs [4–9]. Therefore, temperature measurement becomes one of key issues needed to be solved during the development of semiconductor lighting technology. At present, commonly used testing methods for LED’s temperature, i.e., p-n junction temperature or surface temperature, mainly include forward-voltage method [10, 11], pulse-current method [12], reverse-current method [13], bidirectional thermal resistance model method [14], peak-wavelength method [15], infrared thermal-imaging (TI) method [16–18], thermo-reflectance method [19–21], reflected-light method [22], and so on. Most of these methods can only measure average temperatures. Two dimensional (2D) temperature distribution can offer more detailed and valuable information for extensive evaluation of LEDs’ thermal properties as well as their reliabilities, and thus, the non-contact detection of 2D temperature distribution becomes an important task. Among above mentioned methods, the forward-voltage method [10, 11], pulse-current method [12], reverse-current method [13], and bidirectional thermal resistance model method [14], are all related with the electrical signal, being unable to obtain the 2D temperature distribution of LEDs. The peak-wavelength method [15] is based on the characteristics that the peak wavelength shifts towards longer wavelength when the temperature increases. However, since the change of peak wavelength is too small, it is difficult to achieve sufficiently accurate testing results. The TI method [16–18] is able to measure 2D temperature distribution but limited by its spatial resolution. Additionally, its accuracy is strongly dependent on the emissivity of materials which is hard to be calibrated point-to-point. The charge coupled device (CCD) based thermo-reflectance method [19–21] can also be used to measure 2D temperature distribution. However, CCD detector can not detect the emission spectrum of LEDs, which is vital for determining the incident light so as to avoid the band-gap modulation as revealed in our previous work [22]. Shih et al. have proposed a simulation method to obtain the temperature distribution of white LEDs with different packages [23]. However, this simulated model is too idealistic and lacks a detailed discussion of additional scattering losses.
Our previous reflected light method by spectrometer can only test one point at a time [22]. By combining the reflected light method and the micro-hyperspectral imaging technology, we expand this method, denoted as micro-hyperspectral imaging based reflected light (μ-HIRL) method, to the 2D temperature distribution measurement of LED under test (LUT). Hyperspectral imager consists of one thousand more spectrometers, containing both spectral and image information. Thus, μ-HIRL method owns merits of both high efficiency and high resolution for 2D temperature measurement of light emitting devices. Besides, spectrometer has higher dynamic range than CCD, hence, is more suitable for measurement of light emitting devices.
2. Theory
The μ-HIRL method relies on the principle that the reflectivity (R) of LUT changes with temperature. Therefore, the relationship between the reflectivity and the temperature of a given material is linear in a first approximation [24]
In our actual experiment, R is equivalent to the relative reflected intensity of incident light (L), and the dependence can be re-written as
where T0 denotes the heat sink temperature, Ts the surface temperature, and the reflected intensity corresponding to the temperature at Ts and T0, respectively, KL the temperature sensitive parameter (TSP), and the variation of reflected intensity of incident light.Prior to testing the 2D temperature distribution of LUT, we need to calibrate the 2D TSP coefficients, , where i and j correspond to the locations on the surface of LUT. A KL matrix of LUT is derived from linear fitting of the reflected hyperspectral data cube as a function of heat sink temperature, according Eq. (2). However, the elements of KL matrix might fluctuate the measurement accuracy considering the influence of non-uniformity of temperature controlling and the deviation of linear fitting as well as the uneven chip surface. Therefore, to exclude these influences, it is necessary to perform a pre-processing of KL matrix by a convolution linear filtering, as shown in Eq. (3),
where KL is the matrix of original TSP of LUT, is the convolution kernel as shown in Eq. (4). Here, we simply take the square matrix h if we let m=n=5, and the value of all elements equals 1/25. In order to obtain the identical resolution between original image and processed image, during the convolution processing, we expand the data on the boundary of image by treating the values outside the matrix as the values of mirror positions in the image. This treatment will minimize the boundary transition of the results as much as possible. After completing the calibration of K matrix, the LUT is lit to operate at specific electrical currents. Then, we substitute the measured reflected light intensity at different currents into Eq. (2) to calculate the 2D temperature distribution on the LUT surface, as described in Eq. (5).3. Experiments
The experimental setup for measuring the 2D temperature distribution of LEDs is schematically shown in Fig. 1, including the micro-hyperspectral imager with a CCD detector, the optical microscope with filters, the optical fiber, the incident light, the LUT under test, electrical source meters, and temperature-controlling devices. Two commercial LEDs (628 nm red and 702 nm near-infrared (NIR) LEDs) are providing the incident light, both with the same chip area of 1 mm×1 mm. The 453 nm blue and 525 nm green bare LEDs are used as LUTs. The reason of the selection of bare LED chips in this work lies in the convenient comparison between our proposed method with micro-thermocouple (μ-TC) and TI methods. As shown in Table 1 and Fig. 2, two principles for selecting incident light shall be followed. Principle I, the wavelength of incident light should be longer than that of LUT in order to avoid the extra excitation of photo luminescent emission from LUT. Principle II, the wavelength of incident light shall not be too close to that of LUT, in case that band-gap modulation will occur. It is better to be far enough away from that of LUT, for completely separating the reflected light from the emission of LUT. An electrical source meter (Keithley 2400) supplies direct current (DC) for incident light source (600 mA for red incident light, 300 mA for NIR incident light) whose heat sink temperature is controlled at 25C by a temperature-controlling device (Whtalent TLTP-TEC). Meanwhile, another electrical source meter (Keithley 2611) supplies DC for LUT, the heat sink temperature of which is controlled by another temperature-controlling device (Keithley 2510). In the calibration, the heat sink temperature of LUT under the unlit state is adjusted from 30 C to 50 C with an interval of 10 C. During the measurement, the LUT is lit to operate at electrical currents ranging from 100 mA to 500 mA with an interval of 100 mA.
Table 1. The peak wavelength of incident LED and LUT
To verify the accuracy of this proposed method, the average temperature calculated by this method is compared with that of μ-TC, and the measured temperature distribution is also compared with that of TI by an infrared thermographer (Research-N2). To test the influence of light absorption on μ-TC, we focused most of the emitting light of LED by lens on the μ-TC, the temperature rise of μ-TC is less than 2 C. Further considering the contact area of the μ-TC is only about one thousandth of the LED chip area, most of the emitting light escape to the ambience. Therefore, in this case, the light absorption impact on μ-TC is ignorable. As for the TI method, it is unable to obtain the absolute temperature of each pixel due to the incapability of calibrating emissivity of LED chip pixel by pixel, only the comparison of relative temperature trends is more practical.
Fig. 3 (a) The TSP of blue LUT under red incident light and 2D temperature distribution under different currents of (b) 100 mA, (c) 200 mA, (d) 300 mA, (e) 400 mA, and (f) 500 mA, respectively.
Fig. 4 (a) The TSP of blue LUT under NIR incident light and 2D temperature distribution under different currents of (b) 100 mA, (c) 200 mA, (d) 300 mA, (e) 400 mA, and (f) 500 mA, respectively.
Fig. 5 (a) The topography of blue LUT. The temperature distribution of Line S01 on the blue LUT at different currents under the incident of (b) red light and (c) NIR light, respectively.
Fig. 6 For five currents, the comparison of averaged temperature of blue LUT by μ-HIRL method with that by μ-TC method under two incident lights. The error bars represent MSE in results of μ-HIRL method from those of μ-TC method.
4. Results and discussions
4.1. Results of blue LUT
Figures 3(b)–3(f) show 2D temperature distribution of blue LUT at five different currents while the 628 nm red LED is used as incident light. Figures 4(b)–4(f) show 2D temperature distribution of the same blue LUT working at same currents, only with the incident light changed as a 702 nm NIR LED. The result reveals that the surface temperature distribution is more uniform in the case of NIR LED as incident light. We conjecture that it is due to less spectral interference and overlapping in the case of the longer NIR light as incident light. In addition, it is noticeable that as the current increases, the temperature distribution on the chip surface becomes more uniform. The temperature distribution of Line S01, which does not pass through electrodes, as indicated in Fig. 5(a) under the incident of red light and NIR light are shown in Figs. 5(b) and 5(c). Their results are slightly different.
In order to verify the accuracy of our proposed method, we compare the measured averaged temperature of our method with that of the μ-TC method, together with their mean square errors (MSE) presented in Fig. 6. This comparison is carried out by selecting the area (5 pixel× 5 pixel) of LUT near the position of μ-TC probe. The results in Fig. 6 indicate a fairly good agreement between the μ-TC method and our proposed method. The MSE in the case of NIR incident light appear to be lower (especially at 300 mA), implying more accurate temperature can be obtained with NIR incident light. The measurement averaged error for red and NIR light is only 2.2% and 1.8% respectively.
Fig. 7 Two lines on the blue LUT of (a) μ-HIRL method, (c) TI method. Normalized temperature (TN) distribution of blue LUT, measured by μ-HIRL method and TI method, respectively. (b) Line L01, (d) Line L02. Error bars represent the MSE in results of μ-HIRL method from those of TI method.
In addition, we also compare the measured temperature distribution of μ-HIRL method with the TI method in Fig. 7, where two representative lines, denoted as L01 and L02, are drawn to normalize and plot the temperature distribution along horizontal and vertical directions, respectively. In comparison, we performed similar data processing for error analysis. Their trends are coincident with each other, but μ-HIRL offers more details. The valleys along the temperature distribution in Figs. 7(b) and 7(d) might indicate the electrodes of LUT. However, the pixel-to-pixel size of μ-HIRL method and TI method is about 3 μm and 30 μm, respectively, indicating μ-HIRL method one order of magnitude in spatial resolution higher than TI method.
Fig. 8 (a) The TSP of green LUT under NIR incident light and 2D temperature distribution under different currents of (b) 100 mA, (c) 200 mA, (d) 300 mA, (e) 400 mA, and (f) 500 mA, respectively.
Fig. 9 (a) The topography of green LUT. (b) Temperature distribution of Line S02 on the green LUT under the NIR incident light. The circles indicate some dark dots on the right side of green LUT.
Fig. 10 For five currents, the comparison of averaged temperature of green LUT by μ-HIRL method with that by μ-TC method only under NIR incident lights. The error bars represent MSE in results of μ-HIRL method from those of μ-TC method.
4.2. Results of green LUT
Likewise, we make the same analysis on the green LUT. In order to avoid the band-gap modulation as much as possible, we only select the NIR light as the incident light. For green LUT, the 2D temperature distributions at five various currents are shown in Figs. 8(b)–8(f). From the topography of green LUT (Fig. 9(a)), it is clearly found that there exist some dark dots on the right side (marked with circle), possibly leading to poor signal-to-noise ratio of reflected light and poor fitting of K matrix, which might explain the anomalous low temperature region as shown on the right side of Figs. 8(c)–8(f). The temperature distributions of Line S02 at various currents are plotted in Fig. 9(b). We also compare the μ-HIRL method with μ-TC method in measured average temperature (Fig. 10) with an averaged error of 2.0%, and compare the μ-HIRL method with the TI method in temperature distribution (Fig. 11). In Fig. 11, we take two horizontal lines, L03 and L04, passing through the electrodes for analysis, showing the identical varying trends and locations of peaks and valleys between two methods.
Fig. 11 Two lines on the green LUT of (a) μ-HIRL method, (c) TI method. TN distribution of green LUT, measured by μ-HIRL method and TI method, respectively. (b) Line L03, (d) Line L04. Error bars represent the MSE in results of μ-HIRL method from those of TI method.
5. Conclusion
The proposed μ-HIRL method is a non-contact measurement method that combines the micro-hyperspectral imaging technique and the reflected light method to measure the 2D temperature distribution of bare LED chip without interrupting its working status. This μ-HIRL method solves the shortcomings of conventional thermo-reflectance method and TI method. Comparative experiments by TI method and μ-TC method have achieved good agreements and validated μ-HIRL that owns merits of both high efficiency and high spatial resolution, at least one order of magnitude higher than that of TI method. We suggest that the wavelength of LUT and that of incident light shall be kept far away from each other to avoid possible measurement errors caused by light interference between LUT and incident light. This method is still applicable for packaged LEDs as long as the microscope can focus on the surface of chip and the package is transparent for the incident light. But for phosphor coated LEDs, the rough surface might limit the application of this method in terms of the shallow depth of field of microscope, especially at high magnification.
Funding
National Natural Science Foundation of China (11604285, 51605404); Major Science and Technology Project of Fujian Province (2018H6022); Natural Science Foundation of Fujian Province (2018J01103); Technological Innovation Project of Economic and Information Commission of Fujian Province; Science and Technology Project of Xiamen (3502Z20173016); State Quality Inspection Administration Science and Technology Projects (2017QK140).
Acknowledgments
The authors wish to thank anonymous reviewers for their valuable suggestions.
References
1. E. F. Schubert and J. K. Kim, “Solid-state light sources getting smart,” Science 308, 1274–1278 (2005). [CrossRef] [PubMed]
2. Y. Zhang, J. Wang, W. Zhang, S. Chen, and L. Chen, “LED-based visible light communication for color image and audio transmission utilizing orbital angular momentum superposition modes,” Opt. Express 26, 17300–17311 (2018). [CrossRef] [PubMed]
3. C.-M. Kang, D.-J. Kong, J.-P. Shim, S. Kim, S.-B. Choi, J.-Y. Lee, J.-H. Min, D.-J. Seo, S.-Y. Choi, and D.-S. Lee, “Fabrication of a vertically-stacked passive-matrix micro-LED array structure for a dual color display,” Opt. Express 25, 2489–2495 (2017). [CrossRef]
4. X. Luo, R. Hu, S. Liu, and K. Wang, “Heat and fluid flow in high-power LED packaging and applications,” Prog. Energy Combust. Sci. 56, 1–32 (2016). [CrossRef]
5. H. Baumgartner, A. Vaskuri, P. Kärhä, and E. Ikonen, “Temperature invariant energy value in LED spectra,” Appl. Phys. Lett. 109, 231103 (2016). [CrossRef]
6. S. Chhajed, Y. Xi, Y.-L. Li, T. Gessmann, and E. F. Schubert, “Influence of junction temperature on chromaticity and color-rendering properties of trichromatic white-light sources based on light-emitting diodes,” J. Appl. Phys. 97, 054506 (2005). [CrossRef]
7. N. Narendran and Y. Gu, “Life of LED-based white light sources,” J. Disp. Technol. 1, 167–171 (2005). [CrossRef]
8. Z.-Q. Guo, T.-M. Shih, Z.-B. Peng, H.-H. Qiu, Y.-J. Lu, Y.-L. Gao, L.-H. Zhu, J.-H. Zheng, and Z. Chen, “On a relationship among optical power, current density, and junction temperature for InGaN-based light-emitting diodes,” AIP Adv. 7, 015307 (2017). [CrossRef]
9. J. H. Kim and M. W. Shin, “Thermal behavior of remote phosphor in light-emitting diode packages,” IEEE Electron Device Lett. 36, 832–834 (2015). [CrossRef]
10. Y. Xi and E. F. Schubert, “Junction-temperature measurement in GaN ultraviolet light-emitting diodes using diode forward voltage method,” Appl. Phys. Lett. 85, 2163–2165 (2004). [CrossRef]
11. S. Lin, T. Shih, Y. Lu, Y. Gao, L. Zhu, G. Chen, B. Wu, Z. Guo, J. Zhang, X. Fan, R. R. Chang, and Z. Chen, “Determining junction temperature in InGaN light-emitting diodes using low forward currents,” IEEE Trans. Electron Devices 60, 3775–3779 (2013). [CrossRef]
12. H.-Y. Ryu, K.-H. Ha, J.-H. Chae, O.-H. Nam, and Y.-J. Park, “Measurement of junction temperature in GaN-based laser diodes using voltage-temperature characteristics,” Appl. Phys. Lett. 87, 093506 (2005). [CrossRef]
13. B. Wu, S. Lin, T.-M. Shih, Y. Gao, Y. Lu, L. Zhu, G. Chen, and Z. Chen, “Junction-temperature determination in InGaN light-emitting diodes using reverse current method,” IEEE Trans. Electron Devices 60, 241–245 (2013). [CrossRef]
14. Y. Ma, R. Hu, X. Yu, W. Shu, and X. Luo, “A modified bidirectional thermal resistance model for junction and phosphor temperature estimation in phosphor-converted light-emitting diodes,” Int. J. Heat Mass Transf. 106, 1–6 (2017). [CrossRef]
15. K. Chen and N. Narendran, “Estimating the average junction temperature of AlGaInP LED arrays by spectral analysis,” Microelectron. Reliab. 53, 701–705 (2013). [CrossRef]
16. T. Kobayashi, T. Katagiri, and Y. Matsuura, “Multi-element hollow-core anti-resonant fiber for infrared thermal imaging,” Opt. Express 24, 26565–26574 (2016). [CrossRef] [PubMed]
17. A. Kozlowska, M. Latoszek, J. W. Tomm, F. Weik, T. Elsaesser, B. Spellenberg, and M. Bassler, “Analysis of thermal images from diode lasers: temperature profiling and reliability screening,” Appl. Phys. Lett. 86, 203503(2005). [CrossRef]
18. O. J. Joung and Y. H. Kim, “Application of an IR thermographic device for the detection of a simulated defect in a pipe,” Sensors 6, 1199–1208 (2006). [CrossRef]
19. M. Farzaneh, K. Maize, D. Luerssen, J. A. Summers, P. M. Mayer, P. E. Raad, K. P. Pipe, A. Shakouri, R. J. Ram, and J. A. Hudgings, “CCD-based thermoreflectance microscopy: principles and applications,” J. Phys. D: Appl. Phys. 42, 143001 (2009). [CrossRef]
20. J. A. Summers, M. Farzaneh, R. J. Ram, and J. A. Hudgings, “Thermal and optical characterization of photonic integrated circuits by thermoreflectance microscopy,” IEEE J. Quantum Electron. 46, 3–10 (2010). [CrossRef]
21. B. Vermeersch, J.-H. Bahk, J. Christofferson, and A. Shakouri, “Thermoreflectance imaging of sub 100 ns pulsed cooling in high-speed thermoelectric microcoolers,” J. Appl. Phys. 113, 104502 (2013). [CrossRef]
22. Y. Xiao, T.-Z. Wu, S.-J. Dang, Y.-L. Gao, Y. Lin, L.-H. Zhu, Z.-Q. Guo, Y.-J. Lu, and Z. Chen, “Determining junction temperature of LEDs by the relative reflected intensity of the incident exciting light,” IEEE Trans. Electron Devices 64, 2257–2260 (2017). [CrossRef]
23. B.-J. Shih, S.-C. Chiou, Y.-H. Hsieh, C.-C. Sun, T.-H. Yang, S.-Y. Chen, and T.-Y. Chung, “Study of temperature distributions in pc-WLEDs with different phosphor packages,” Opt. Express 23, 33861–33869 (2015). [CrossRef]
24. G. Tessier, S. Pavageau, B. Charlot, C. Filloy, D. Fournier, B. Cretin, S. Dilhaire, S. Gomes, N. Trannoy, P. Vairac, and S. Volz, “Quantitative thermoreflectance imaging: calibration method and validation on a dedicated integrated circuit,” IEEE Trans. Compon. Packag. Technol. 30, 604–608 (2007). [CrossRef]
