On the thermal characterization of an RGB LED-based white light module

Applied Thermal Engineering 38 (2012) 105e116

Contents lists available at SciVerse ScienceDirect

Applied Thermal Engineering journal homepage: www.elsevier.com/locate/apthermeng

On the thermal characterization of an RGB LED-based white light module Hsien-Chie Cheng a, b, Jia-Yun Lin c, Wen-Hwa Chen c, d, * a

Department of Aerospace and Systems Engineering, Feng Chia University, Taichung 30013, Taiwan National Center for High-Performance Computing, Hsinchu 30076, Taiwan c Department of Power Mechanical Engineering, National Tsing Hua University, Hsinchu 30013, Taiwan d National Applied Research Laboratories, Taipei 10622, Taiwan b

a r t i c l e i n f o

a b s t r a c t

Article history: Received 30 August 2011 Accepted 6 January 2012 Available online 11 January 2012

The study aims at evaluating the thermal characteristics of a red-green-blue (RGB) light emitting diode (LED)-based white light module in natural convection through infrared (IR) thermography and forward voltage measurements. The light output and packaging efficiencies of these three color types of LEDs are first characterized using a spherical integrating photometer employing an integrating sphere and an optical analysis program, respectively. These two efficiency data give an estimate of the power fraction not converted into emitting light, by which the temperature distribution of the module is calculated using thermal finite element (FE) modeling. To facilitate the IR thermography measurement, the coefficient of emissivity of the transparent molding compound as the LED optical lens is explored. The results of the measurements are compared with those of the thermal FE modeling and thermocouple measurement. Moreover, the limitation of the IR thermography measurement in LED thermal characterization is addressed, and enhancement of the measurement accuracy is achieved through a proposed temperature correction procedure. Besides, the uncertainty in the forward voltage measurements is also assessed. Results show that the maximum junction temperature of the RGB LED-based white light module can reach about 100
C even at a low ambient temperature of 25
C, and the low thermal conductivity of the transparent molding compound is one of the major causes of the poor thermal performance. In addition, IR thermography tends to overrate the surface temperature of the LEDs, and the transparency of the molding compound could be the direct consequence of the overestimate. It also turns out that the IR thermography can be an effective tool for LED surface temperature measurement after modification using the proposed temperature correction procedure. At last, the uncertainty analysis reveals that the uncertainty in the measured junction temperature using the forward voltage method is about 4e8%, depending on the color of the LEDs, and can it be greatly improved by increasing the accuracy of the voltage measurement. Ó 2012 Elsevier Ltd. All rights reserved.

Keywords: Thermal performance Light-emitting diode Infrared thermography Forward voltage method Finite element modeling Thermocouple measurement Uncertainty analysis

1. Introduction Light-emitting diodes (LEDs) are a solid-state lighting technology. It has drawn remarkable attention from the lighting industry in recent years because of their many advantageous features, such as high reliability, long life time, low power consumption, small scale in size.etc. In color performance, they possess the characteristics of wide color gamut and high color saturation. The optical shape of the light emitted from LEDs can be determined by controlling the process of epitaxy and packaging. The increase of the luminous flux or luminance can be feasible by the use of high power and brightness

* Corresponding author. National Center for High-Performance Computing , Hsinchu 30076, Taiwan. Tel.: þ886 3 5742913. E-mail address: [email protected] (W.-H. Chen). 1359-4311/$ e see front matter Ó 2012 Elsevier Ltd. All rights reserved. doi:10.1016/j.applthermaleng.2012.01.014

LED chips. To date, for commercial products, the illuminating efficiency for 1 W high power LED has reached above 80 lm/W, which is also far superior to the conventional fluorescent (50 lm/W) and incandescent (20 lm/W) light sources. With the rapid improvement in luminance and efficiency, LEDs are expected to be used as an alternative for the conventional light sources, and has also potential for applications with high commercial value, such as general light source, backlight of liquid crystal display. Despite of the great improvement in the illuminating efficiency, high percentage of electrical input power in LEDs remains being converted into redundant heat. The problem becomes severe for lighting using high power and brightness LEDs, leading to high chip junction temperature and temperature gradient. High temperature and temperature gradient are essential reliability parameters for LED packages since they potentially create high thermal stresses

106

H.-C. Cheng et al. / Applied Thermal Engineering 38 (2012) 105e116

and strains, eventually degrading the packages’ fatigue life [1]. Except the mechanical reliability or integrity concerns, high junction temperature would also impact the electrical and optical performances of LEDs. For example, an exceeding temperature in the p-n junction of the LED chips may result in the lessening of their lumen efficiency and bandgap energy, and also the causing of a redshift of the peak wavelength. While the drastic upgrading of light output becomes possible, thermal performance of LED packages must be also enhanced in order to improve their electrical, optical and mechanical performances simultaneously. A better understanding of their thermal characteristics can be facilitated through effective thermal characterization. In literature, several experimental approaches are available for LED temperature measurement, including micro-Raman spectroscopy [2], electroluminescence [2], photoluminescence [3], noncontact method such as infrared (IR) thermography [4], nematic liquid crystal [5] and forward voltage measurement [6]. IR thermography is a technique that constructs images (also called thermographs) of infrared light emitted by objects resulting from their thermal condition through a thermography camera. Note that all materials beyond 0 K would emit radiation, and some of the radiation falls in the IR range. The main features of IR thermography include more brilliant colors, broader temperature range, higher accuracy, and most importantly, a noncontact measurement, which can be very useful when dealing with circumstances involving measurement of a moving, fragile and toxic object. The nondestructive technique has proved to be effective and powerful in characterizing the surface temperature of structures in many applications (see, e.g. [7,8]). It is, however, found that it could not be used to assess the internal temperature of objects, such as the junction temperature of LEDs. By contrast, the forward voltage method is feasible for LED junction temperature measurement, and has been widely used today in the area because of easy operation and better accuracy. Beside the experimental approaches, numerical simulation of the thermal performance of LED packages, such as finite element (FE) modeling, is also very popular, due to the great advance of computer technology. Many other studies related to the thermal characterization of LED packages using the numerical and experimental approaches can be also found [9e11]. Several approaches are available to produce white light using LEDs, such as the use of a blue or ultraviolet LED to excite one or more phosphors or the integration of red, green, and blue LEDs (briefly termed an RGB LED-based white light module in the work). The latter approach implies several advantageous features, including a light source with a variable color point and the highest efficiency LED-based white light. Except the key challenge of holding the

desired white point within an acceptable tolerance, the white light LED technology also face thermal challenges because of high power density and limited heat transfer capability as a result of the close placement of the LEDs. Thus, the study aims at thermal characterization of the RGB LED-based white light module in natural convection. To achieve the goal, two effective thermal characterization techniques are proposed, namely infrared (IR) thermography for surface temperature measurement and forward voltage method for junction temperature measurement. Since the effectiveness of the IR thermography measurement mainly relies on an accurate emissivity data, the emissivity coefficient of the transparent molding compound is first examined through experiment prior to the measurement. Besides, the limitation of using the IR thermography technique for LED surface temperature measurement is identified, and most importantly, a temperature correction procedure is proposed to enhance the measurement accuracy. Moreover, to grasp the accuracy of the forward voltage measurement, uncertainty analysis is performed. The validity of these thermal characterization models are demonstrated through comparison with threedimensional (3D) FE modeling (FEM) and thermocouple measurement results. Before conducting the thermal FE analysis, the illuminating efficiencies of these three color types of LEDs are first assessed using a spherical integrating photometer employing an integrating sphere and an optical analysis program, and subsequently used to estimate the fraction of the electrical input power not converted into emitting light. The power fraction, denoting the heat generation rate of the LED chips, is used in the thermal FE modeling for calculating the temperature distribution of the module. 2. Description of the arrayed LED module The RGB LED-based white light module consists of 9 piranha type LED packages, namely 3 red, 3 green, and 3 blue LEDs, which are also a pin-through-hole type package, as shown in Fig. 1(a). These LED packages are arranged in a 3x3 array, and their placement layout is shown in Fig. 1(b), where “G” represents the green LEDs, “B” the blue LEDs and “R” the red LEDs, the subscript of which, from 1 to 3, represents the ID of the LEDs for each color type. The LED packages of the same color are electrically connected in series, as shown in Fig. 1(b). Each LED package comprises the following components: molding compound, leadframe, LED chip, gold wire and die attach, as presented in Fig. 1(a). As a p-n junction is forward biased, the energy possessed by chargeecarrier pairs can be converted into either electromagnetic radiation (a radiative transition) or into heat (a nonradiative transition). The generated heat will propagate to the surrounding components of the package, and

Fig. 1. The RGB LED-based white light module III. Light and thermal measurements. (a) A piranha type LED package, (b) The layout of the LED array and its series electrical connection configuration.

H.-C. Cheng et al. / Applied Thermal Engineering 38 (2012) 105e116

subsequently to the ambient by means of heat convection and radiation. The leadframe is made of copper alloy and consequently offers excellent thermal conductance. Since the gold wire is very tiny, its heat conduction capability is neglected in the investigation. The mixture of the red, green and blue LEDs can create white light as the color components are appropriately weighted. The RGB LED-based white light illumination system has predominance in color reproduction. In the standard of National Television System Committee (NTSC), it can present more than 100% color gamut and a considerably improved brightness and contrast. However, to sustain a desired white point within acceptable tolerances is a great challenge for the white light module, mainly because of its temperature-dependent optical performance and material characteristics. Unfortunately, the close placement or spacing of LEDs in a PCB would not help improve the thermal performance of the module due to the increase of the power density and the limited heat transfer capability.

3. Light and thermal measurements 3.1. Characterization of LED efficiency through light measurement LED output efficiency is a measure of the light output compared to the total electrical input energy. It can be mainly categorized into two parts: 1) internal quantum efficiency, and 2) light extraction efficiency. The former is considered as the electrical-to-optical conversion efficiency of the light emitting device, and the latter is determined by two main parts: extraction efficiency of LED chips and packaging efficiency of LED packages. More specifically, the internal quantum efficiency is the optical power emitted from the active region divided by the electrical input power. Furthermore, the extraction efficiency of LED chips highlights their inability to emit all of the recombination photons generated by their active layer due to the total internal reflection (TIR) or the direct emitted light absorption by the substrate or the other bonded structure in the package. With the implementation of a metal reflector formed in the LED structure during fabrication, the light extraction efficiency of LED chips can be greatly enhanced [12], and the associated light extraction loss can be neglected, especially when the chip size is small, thereby reducing the light absorption. Furthermore, the packaging efficiency of LEDs is the ratio of the optical power emitted out of the package to that emitted from the LED chips embedded. The output efficiency of LED packages can be approximately expressed as the product of the internal quantum efficiency and packaging efficiency. By the relationship, the internal quantum efficiency and so the fraction of the electrical input power that is not transferred into emitting light can be predicted. To assess the output efficiency of an LED package, a spherical integrating photometer employing an integrating sphere is built for measuring the total luminous flux emitted from the LED. When a ray of light comes into the sphere from any direction, it runs into the inner wall. Subsequently, diffuse reflection (Lambertian scattering) takes place until the light becomes randomized within the sphere, which is then collected by a detector through a detector port and an aperture on the integrating sphere. The brightness and uniformity of the light measured from the integrating sphere would not be influenced by the angle of incidence, spatial distribution and polarization of the light source. The ratio of the intensity of the incident light to the intensity of the light passing through the aperture of the sphere is the same as the ratio of the surface area of the aperture to that of the inner wall of the integrating sphere. In the measurement, auxiliary light source is used to compensate for changes in the spectral efficiency of the integrating sphere due to self-absorption and reflectance properties of LEDs and their packages.

107

The experiment is set up following the standard of International Commission on Illumination [13]. The integrating sphere is a relative measuring method. Before the measurement, calibration is needed to assure the accuracy of the measurement. The standard is also used to calibrate the integrating sphere. In the standard LED method, a standard LED with a known luminous flux is used and measured under the same experimental set up as the test LED. The following relation can be employed to minimize the error of the integrating sphere measurement,

Ft ¼ a  F 

Yt  Fs ; Ys

(1)

where a is the self-absorption correction factor used to estimate the light absorbed by the LED package and chip, Ft the total luminous flux of the test LED, Fs the total luminous flux of the standard LED, Yt the reading figure from the photometer for the test LED, Ys the reading figure from the photometer for the standard LED, and F the mismatch spectral correction factor. To investigate the packaging efficiency of the LEDs, the optical analysis program LightToolÒ is applied. In this analysis, the LED light emission is also assumed to be a diffuse Lambertian source, and the luminous intensity distribution of the light source is considered to be symmetrical for modeling simplification. The material properties and the associated refraction index can be referred to [14], and the leadframe is considered as a total reflection material. 3.2. Characterization of surface temperature through IR thermography measurement The IR thermography measurement follows the StefaneBoltzmann law, that is, a black body can absorb all wavelengths of emitted light in any temperature and the radiated intensity per unit area from the surface of a black body at temperature T is proportional to the fourth power of the temperature. It is expressed in the following mathematical expression:

J ¼ ε s T 4;

(2) 8

where s is the StefaneBoltzmann constant (5.67  10 W/m K4), T the temperature, and ε the emissivity indicating the radiation of heat from a grey body. For a given material, it varies from 0 to 1. For example, the emissivity of a pure black body is equivalent to 1, illustrating the full capability of a body to radiate and also to absorb radiation. This equation shows a dramatically increasing function with temperature. The emissivity can change with the materials and wavelength, and would not be uniform even in an object. Once the emissivity of a given material that is to be measured is known, IR thermometer can detect the energy of a specific electromagnetic wave, which yields the temperature of the measured object according to the StefaneBoltzmann law. The IR thermography measurement is first performed to estimate the thermal characteristics of the white light module. The testing condition conforms to the EIA/JEDEC standard JESD51-2 [15]. According to the standard, the module should be placed in a lighttight cubic box so as to keep the internal environment from external thermal disturbances. The box is to simulate a natural convection environment, the material of which is made of black acrylic to assure as much as possible that the radiative heat detected by the IR thermometer all originates from the test sample itself. The length of the cubic box is one foot, and the emissivity of the inner wall of the box is about 1.0. To facilitate the surface temperature measurement using an IR thermometer, the measured object is placed inside the box, where a round hole with a diameter of 90 mm is created on the top surface. In addition, a thermocouple is placed at one inch below the test sample to measure the ambient temperature 2

108

H.-C. Cheng et al. / Applied Thermal Engineering 38 (2012) 105e116

inside the box. In the investigation, an IR thermometer (NEC TH3102MR) is utilized for the temperature measurement. The lens of the IR camera holds a measurable range between 50
C and 2000
C with a spatial resolution of 0.468 mm and a thermal resolution of 0.02
C. As mentioned earlier, the LED chips and also major part of the top surface area of the substrate are covered by a transparent molding compound as the LED optical lens. The effectiveness of the present IR thermography measurement strongly relies on an accurate estimate of its surface emissivity coefficient. However, the data is not readily available in manual. Thus, an effective calibration experimental procedure, which makes use of both thermocouple and IR thermography measurements, is proposed to assess the surface emissivity coefficient of the transparent molding compound. 3.3. Characterization of junction temperature through diode forward voltage measurement Linear temperature (T) dependence can be observed in the forward voltage (Vf) of a p-n junction measured at a certain current. Such dependence has been extensively justified by, such as [6], based on Shockley equation [16]. In their theoretical derivation of the mathematical relationship between Vf and T, all relevant factors affecting the temperature dependence of the forward voltage are included, namely the intrinsic carrier concentration, the bandgap energy, and the effective density of states. The derivative of the junction voltage with respect to the junction temperature defines the diode temperature sensitive parameter (TSP),


dVf k aTðT þ 2bÞ 3k N N  z ln D A  2 e dT e NC NV eðT þ bÞ

(3)

where e is the electronic charge, k Boltzmann’s constant, NC and NV effective densities of states at the conduction-band and valenceband edges, ND and NA dopant concentrations, a and b the Varshni parameters. Essentially, the expression reveals that relation between the forward voltage and the junction temperature at a low biasing current [17] or at a certain temperature range can be curvefitted linearly as,

TjF ¼ Tj0 þ

VfF  Vf0

k

(4)

where k denotes the fitting coefficient or TSP parameter, representing the derivative of the linear relationship with respect to temperature, T0j the initial junction temperature, V0f the initial forward voltage under the initial junction temperature, VFf the steady forward voltage under the measured junction temperature. The coefficient k is the derivative of the linear relationship with respect to temperature, which is commonly termed TSP parameter. The diode forward voltage method can be applied to derive the diode TSP and so junction temperature. Two calibration procedures are involved in the measurement method, namely one calibration procedure and one junction temperature measurement procedure. According to the EIA/JEDEC standard JEDEC51-1 [18] the input biased current or sense current should be small enough to avoid self-heating in the junction, and large enough that the forward voltage measurement would not be influenced by the surface

experiment. To calibrate the sense current, both current-forward voltage and current-temperature dependences are sought through experiments in natural convection under the EIA/JEDEC standard JESD51-2 [15]. For the calibration experiment, the LED test vehicle is placed in a well temperature-controlled chamber, connected to a measurement and power system. The temperature in the current-temperature dependence is defined as the gap between the LED surface temperature and the ambient temperature, which is considered as the temperature at one inch below the LED. Subsequently, the TSP parameter associated with the LED is calibrated through the following experiment procedure. The calibration is also performed in the aforementioned measurement system. Two different chamber temperatures T1 and T2, also regarded as the ambient temperature, are set by the temperature-controlled chamber, respectively. As soon as thermal equilibrium with the chamber temperature is attained, these two chamber temperatures are considered to be equivalent to the junction temperatures T1j and T2j of the LED to be measured. By supplying the calibrated low pulse current to the LED, the corresponding forward voltages V1f and V2f can be measured. The TSP parameter can be then derived by:

k ¼

Vf1  Vf2 Tj1  Tj2

The junction temperature of the LED chip is measured through the following procedure. First of all, the low pulse current calibrated above is supplied to the LED to measure the initial forward voltage. Subsequently, a high continuous current is employed to heat the LED. Once thermal equilibrium is reached, the high continuous current is switched to the low pulse current to measure the steady forward voltage. By these two forward voltage data and the initial junction temperature (i.e., the ambient temperature) together with the calibrated TSP parameter calibrated above, the junction temperature can be obtained from Eq. (5). Uncertainty analysis is also performed on the forward voltage measurements according to the standard evaluation procedure for measurement uncertainty defined in the international organization for standardization (ISO) standard (1995). The uncertainty analysis is derived based on [7,19]. If various measurements are performed and the uncertainty in each measurement could be expressed with the same chance, the uncertainty of a targeted result can be then assessed based on the uncertainties in the primary measurement. Since a different experimental approach to reach the desired result may lead to a dissimilar uncertainty, an appropriate selection of the best measurement method would be critical. The uncertainty in the measured junction temperature using the forward voltage method comes from two major sources, one from the uncertainty of the measured forward voltage and the measured initial junction temperature, and the other from the calibrated TSP curve. Furthermore, the uncertainty of the calibrated TSP curve is also originated from two major aspects: the possible variations in those acquired sample points (i.e., the temperature vs. voltage) and the curve-fitting error. The uncertainty of these sample points would obviously lead to the deviation of the slope of the TSP curve. According to Eq. (5), the uncertainty in the TSP parameter εðkÞ can be derived by,

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi    2  4     . 2    1 2 1 2 1 2 1 2 1 2 2 2 2 2  Vf  Vf Tj  Tj ; Tj  Tj þ ε Vf þ ε Vf εðkÞ ¼ ε Tj þ ε Tj

leakage current. Thus the calibration procedure involves not only a TSP coefficient calibration but also a sense current calibration

(5)

(6)

where εðTj1 Þ is the uncertainty of the controlled chamber temperature “1” (Tj1 ), εðTj2 Þ the uncertainty of the controlled

H.-C. Cheng et al. / Applied Thermal Engineering 38 (2012) 105e116

109

Table 2 The parameters used in the Ellison’s convective heat transfer model [20].

A horizontal plate facing upward A horizontal plate facing downward A vertical plate

Lch (m)

f

n

wl/2(w þ l) wl/2(w þ l) t

1.0 0.5 1.22

0.33 0.33 0.35

w: Width of the plate, l: Length, t: Thickness.

Fig. 2. 3D thermal FE model of the RGB LED-based white light module.

chamber temperature “2” (Tj2 ), εðVf1 Þ the uncertainty of the voltage measurement “1” and εðVf2 Þ the uncertainty of the voltage measurement “2”. According to Eq. (4), the uncertainty in the measured junction temperature εðTjF Þ could be also derived by,

ε



TjF



natural convection and radiation through the regions could be neglected. They are, instead, modeled by way of conduction heat transfer, where the typical air conductivity 0.03 (W/m-K) is applied. Moreover, the surface loads of the thermal FE model are described by two empirical heat transfer coefficient correlation models, namely the Ellison’s empirical model for describing the package’s surface convection effect [20], and the standard radiative coefficient correlation model (see, e.g., [21]), as shown in the following,

hc ¼ 0:83f


Tw  Ta n ; Lch

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi   2   .     k4 þ ε2 VfF þ ε2 Vf0 k2 : ¼ ε2 Tj0 þ ε2 ðkÞ  VfF  Vf0

From the above uncertainty analysis, it is found that the uncertainty in the measured junction temperature is strongly determined by the uncertainty in the TSP parameter. This suggests that an accurate calibration of the TSP parameter could significantly help improve the accuracy of the junction temperature measurement. Besides, a larger TSP parameter can also reduce the uncertainty in the measured junction temperature according to Eq. (7). 4. Thermal FE modeling A rigorous 3D thermal FE model is also developed to investigate the thermal performance of the RGB LED-based white light module in natural convection. The constructed thermal FE model for the module is shown in Fig. 2, where it consists of all the major components in the module, including 9 LED chips, die attach, leadframe, molding compound, air gap and PCB. The total number of FE nodes and elements are 54,150 and 40,635, respectively. The steady-state thermal FE analysis takes into account all the heat transfer mechanisms, namely heat transfer conduction, natural convection and radiation. Table 1 presents the components and the associated thermal conductivities of the module. The air zone between the LED package and PCB, and among these LED packages are inconsiderable; thus, heat dissipation to the ambient due to Table 1 Thermal conductivity of the LED components. Component Blue and green chip Red chip

Conductivity (W/m$K)) Junction of chip (InGaN) Substructure of chip(Al2O) Junction of chip (AlGaInP) Substructure of chip (Si)

Die attach Leadframe (Copper C1100) Molding compound PCB (x,y/z) Air

150 28 77 120 1 391.1 0.7 0.3 0.03

  2 hr ¼ gεs Tw þ Ta2 ðTw þ Ta Þ;

(8)

(7)

(9)

where Tw denotes the wall temperature, Ta the ambient temperature, Lch the characteristic length, f and n the constants, s the Stephen-Boltzmann constant, which is equal to 5:678  108 (J/ s$m2 K4), ε the surface emissivity (0 < ε < 1) and g the radiative view factor. The definition of the characteristic length Lch and the values of the constants f and n are listed in Table 2 [20]. According to [7], the feasibility of these two HT coefficient correlation models has been well confirmed for a small package like the LED module. 5. Results and discussion 5.1. LED output and packaging efficiencies A spherical integrating photometer employing an integrating sphere is applied to assess the output efficiency of these three color types of LEDs (i.e., red, green and blue). The measured total emitted luminous fluxes of these three color types of LEDs are shown in Fig. 3(a), and the associated output efficiencies are also depicted in Fig. 3(b). It is found that all these three luminous fluxes shown in Fig. 3(a) increase with an increasing input power, and tend to become stable as the input power is large. The reason can be that the increase of the input power would also raise the junction temperature of the LED chip, thereby potentially increasing ineffective recombination. Under the same driving power, the green LED can provide the maximum luminous flux output, followed by the red, and the blue has the minimum output. On the other hand, Fig. 3(b) shows that the output efficiencies of all the red, green and blue LED packages, defined as the amount of the emitted light divided by the consumption power, tend to decrease with the increase of input power. Under an identical driving power, the blue LED shows the best output efficiency, followed by the red and green. The result trend is totally opposite to that of the luminous flux-input power relation. Since a higher input power tends to yield a larger junction temperature, it is thus

110

H.-C. Cheng et al. / Applied Thermal Engineering 38 (2012) 105e116

a

b

14 Red 12

Green

30

Efficiency(%)

Flux (lm)

Red Blue Green

35

Blue

10

40

8 6 4

25 20 15 10

2

5

0

0 0

0.1

0.2

0.3

0.4

0

0.1

Input Power (W)

0.2

0.3

0.4

Input Power (W)

Fig. 3. The optical characteristics of the red, green and blue LEDs. (a) Total luminous flux vs. input power, (b) Output efficiency vs. input power.

deduced that high temperature would have a negative effect on the output efficiency of the LEDs. Under the specific driving power shown in Table 3, the output efficiency is shown in Table 4, where it is 9.21%, 26.2% and 31.24%, associated with the red, green and blue LEDs. The output efficiency of the LEDs is mainly determined by the internal quantum efficiency and light extraction efficiency. If the internal quantum efficiency is poor, more input power would be transformed into heat, thereby affecting the light output performance. Furthermore, the optical software LightToolÒ is employed to assess the packaging efficiency of these three color types of LEDs. In the optical simulation, the input driving powers associated with these three color types of LEDs are shown in Table 4, where the green LEDs comprise the highest driving power (i.e., 0.344 W), followed by the red (i.e., 0.115 W) and the blue (0.058 W). The predicted luminous fluxes shown in Table 4 are assumed as the emitting luminous fluxes of the LEDs. Moreover, the metal leadframe surface is assumed a total internal reflection surface. The resultant packaging efficiency data are shown in Table 5. From the table, one can find that the packaging efficiencies of these three color types of LEDs are very comparable, i.e., about 90%. Based on these two predicted efficiency data, the resulting internal quantum efficiency associated with the red, green and blue LEDs is estimated about 10.3%, 29.2% and 34.9%. 5.2. Surface temperature measurement by thermocouple The surface temperature of the white light module is greatly determined by several key factors, such as the placement layout and spacing of the chips [22], driving power and internal leadframe structure and materials. The input driving power is equivalent to those shown in Table 3. According to the configuration of any of the LED packages, as shown in Fig. 1(a), the LED chip is

Table 3 The driving powers for the red, green and blue LEDs.

Red LED Green LED Blue LED

Input current (mA)

Input voltage (V)

Peak wavelength (nm)

Driving power (W)

52 95 18

2.22 3.62 2.89

620 530 460 Total power

0.115 0.344 0.052 1.53

directly attached to the leadframe on the left-hand side, and also electrically connected to that of the right-hand side through a gold wire. Due to the minuscule cross-sectional area of the wire, surrounded by the low-conductivity epoxy molding compound, relatively high thermal resistance can be observed in the righthand part of the LED, thereby potentially having a lower temperature at the region, as compared to that of the left-hand. It is also because of that reason that the left and right parts of the LED tend to have a higher temperature than the top and bottom parts. The measured surface temperatures by the thermocouple (TC) are shown in Fig. 4. The data listed at the center of each LED represents the temperature at the center of the top surface of the LEDs. It reveals that because of the highest driving power, the green LEDs (G1, G2, G3) consist of a higher surface temperature than the others. In addition, the LED located at the center of the substrate, G2, would have the maximum surface temperature due to being fully encircled by the other LEDs, and G1 consists of a larger temperature than G3, probably due to being surrounded by the LEDs with higher temperature, such as B1 and R2. For the red LEDs, it is found that the surface temperature of R3 tends to be larger than that of R2 although they are all neighbored by two green and one blue LED. The reason could be that there is a heat source (i.e., B3) neighboring on the left-hand side (i.e., high thermal conductance region) of R3. By contrast, R1 holds the lowest temperature among these three red LEDs because it does not neighbor on any high power or high temperature LEDs such as the green LEDs. For the blue LEDs, B1 has a higher surface temperature than B2 even though they are all surrounded by one red and two green LEDs. This could be due to that only the right-hand side of B2 is adjacent to an LED (i.e., G2) while both sides of B1 neighbor on an LED. The corner LED B3 exhibits the minimum temperature not only among the blue LEDs, but also among all these LEDs owing to the lowest input driving power.

Table 4 Output efficiencies of the red, green and blue LEDs.

Red LED Green LED Blue LED

Driving power (W)

Luminous flux (lm)

Output power (W)

Output efficiency (%)

0.115 0.344 0.058

5.969 13.694 0.802

0.030 0.032 0.018

26.20 9.21 31.24

H.-C. Cheng et al. / Applied Thermal Engineering 38 (2012) 105e116 Table 5 Packaging efficiencies of the red, green and blue LEDs.

Red LED Green LED Blue LED

Input power (W)

Output power (W)

Packaging efficiency (%)

0.115 0.344 0.058

0.103 0.308 0.052

89.6 89.6 89.5

5.3. Surface temperature characterization by IR thermography and FE modeling The surface temperatures of the RGB LED-based white light module under both light-on (power-on) and light-off (power-off) conditions are characterized using the IR thermometer. The measured results are compared against the thermocouple measurements. The influences of the surface emissivity of the transparent molding compound are also investigated. First of all, the surface temperature of the module under a light-off condition is first examined, and the results are shown in Fig. 5, as function of the emissivity. The figure illustrates that these two measurement results become most comparable as the emissivity is about 0.90e0.94. In addition, it is found that the effects of the surface emissivity are not very considerable on the surface temperature. Furthermore, a power of 0.344 W is individually supplied into the LEDs of the module to simulate a light-on state. Since there is a very consistent trend in these measured data, only one example of the measured results is presented. Fig. 6 shows the results derived from the scenario when the R1 LED is powered on while the rest in the module are powered off. It should be noted that R1 is located on the top right corner of the module, as shown in Fig. 1(b). Both these two measurements show that the surface temperature of the module decreases with an increasing distance from R1, and in addition, the R1 LED exhibits the highest surface temperature among them. It is, however, found that there is a gap in the measured surface temperature of R1 between these two measurements. The measured temperatures by the IR thermography all exceed those of the thermocouple, and the average difference is around 10%. The significant discrepancy in the

80

Surface Temperature (oC)

70

measured temperatures between these two measurements is probably due to that some of the emitted red lights may be degraded or red-shifted into IR due to the transparent molding compound and temperature, respectively. This also suggests that there is a crucial need to perform temperature correction for the IR measurement results. On the other hand, there is a very consistent result in those light-off LEDs between these two measurements, probably because of without the emitting light effect. To evaluate the capability of the IR thermography for surface temperature measurement of a transparent material like the epoxy-based LED optical lens, a simple calibration experiment is performed. Instead of using a real LED model, a piece of plate made of an epoxy material exactly identical to the transparent molding compound is considered as the test specimen of the experiment. By using the test sample, the aforementioned optical effect can be neglected when performing the IR thermography measurement. The specimen is placed on a heating plate, which is gradually heated up from 35
C to 90
C. Following to the EIA/JEDEC standard for natural convection testing, the specimen together with the heating plate is placed in the aforementioned light-tight cubic box. Moreover, three T-type thermocouples are attached to the top surface of the specimen using an appropriate amount of silver-filled epoxy attachment. The average measured temperature is used to benchmark the results of the IR thermography measurement. Note that the volume of the silver-filled epoxy attachment should be as small as possible to prevent it from disturbing the surface temperature. By assuming the emissivity of the transparent molding compound to be 0.92, according to the above finding, the temperature difference between the thermocouple and IR thermography at each temperature increment is recorded, and the results are shown in Fig. 7, associated with the heating plate temperature. It is found from the figure that the temperature difference between these two measurements is noticeable, and tends to increase linearly with an increasing heating plate temperature. In addition, the measured temperatures by the IR thermography are all larger than those of the thermocouple. The transparency of the molding compound may be the direct consequence of the overestimate of the surface temperatures by the IR thermography due to that the detected IR radiation may be stemmed from the emission not only by the transparent molding compound but also by the LED chip and substrate, and even from the reflection by the other surfaces. The linear curvefitting result of the temperature differences plotted in Fig. 7 is given below

DT ¼ 0:161T  3:916:

60

50

40

TC measurement IR Thermography Corrected IR Thermography

30

FE modeling 20 R1

R2

R3

G1

G2

G3

B1

B2

B3

ID of LED Fig. 4. Surface temperatures of LEDs by thermocouple, IR thermography, corrected IR thermography, and FE modeling.

111

(10)

The surface temperature measurement of the RGB LED-based white light module under the input driving powers shown in Table 3 is carried out using the IR thermography. The measurement results in degree of Celsius are also presented in Fig. 4. A significant temperature deviation between the thermocouple and IR thermography measurement can be observed, where there is as much as about 12e18% for the green LEDs, and 5e11% for the other LEDs. Furthermore, Fig. 8 shows the captured IR thermal image of the RGB LED-based white light module (see Fig. 8(a)) and the calculated surface temperature distribution using the proposed FE modeling (see Fig. 8(b)). By comparing these two results, a big discrepancy is detected, especially in the surface temperature of the green LEDs (i.e., G1, G2 and G3). Furthermore, a very analogous trend in the present experimental results and the above calibration data shown in Fig. 7 can be observed, where an increasing surface temperature tends to produce a higher temperature deviation between the IR

a

H.-C. Cheng et al. / Applied Thermal Engineering 38 (2012) 105e116

30

c 30

28

28

26

26

IR Temp. TC Temp.

28

26

Temp.( oC)

Temp.( oC)

b 30

24

22

20 0.8

Temp.( oC)

112

24

22

0.84

0.88 0.92 Emissivity

0.96

20 0.8

1

22

0.84

G1

e

30

0.96

20 0.8

1

f

30

26

26

26

Temp.( oC)

28

24

24

22

0.84

0.88 0.92 Emissivity

0.96

20 0.8

1

h

0.84

0.88 0.92 Emissivity

0.96

20 0.8

1

i

30

26

26

Temp.( oC)

26

Temp.( oC)

28

22

24

22

0.84

0.88 0.92 Emissivity

B3

0.96

1

20 0.8

0.88 0.92 Emissivity

0.96

1

0.96

1

30

28

20 0.8

0.84

B2

28

24

1

24

G2

30

0.96

22

R2

g

0.88 0.92 Emissivity

30

28

20 0.8

0.84

R1

28

22

Temp.( oC)

0.88 0.92 Emissivity

B1

Tem p.( oC)

Temp.( oC)

d

24

24

22

0.84

0.88 0.92 Emissivity

0.96

1

20 0.8

0.84

R3

Fig. 5. Measured surface temperatures of the power-off LED module by thermocouple and IR thermography.

0.88 0.92 Emissivity

G3

H.-C. Cheng et al. / Applied Thermal Engineering 38 (2012) 105e116

b IR Temp. TC Temp.

36

Temp.( OC)

Temp.( OC)

40

32

40

40

36

36

32

28

28

0.8

c

Tem p.( OC)

a

0.84

0.88 0.92 Emissivity

0.96

0.8

1

0.84

0.88 0.92 Emissivity

0.96

1

0.8

36

36

36

Tem p.( OC)

40

Temp.( OC)

40

32

0.84

0.88 0.92 Emissivity

0.96

0.8

1

0.84

R2

0.88 0.92 Emissivity

0.96

0.8

1

36

36

36

Temp.( OC)

40

Tem p.( OC)

40

32

0.84

0.88 0.92 Emissivity

B3

0.96

1

1

0.8

0.96

1

32

28

28

28

0.88 0.92 Emissivity

i

40

0.8

0.84

B2

h

32

0.96

32

G2

g

1

28

28

28

0.96

f

40

32

0.88 0.92 Emissivity

R1

e

0.8

0.84

B1

d

Tem p.( OC)

32

28

G1

Temp.( OC)

113

0.84

0.88 0.92 Emissivity

R3

0.96

1

0.8

0.84

0.88 0.92 Emissivity

G3

Fig. 6. Measured surface temperatures of the LED module with a power-on R1 of 0.344 W by thermocouple and IR thermography.

114

H.-C. Cheng et al. / Applied Thermal Engineering 38 (2012) 105e116

temperature is no larger than 8% and 6%, respectively. From the above result comparisons, it is concluded that the present temperature correction procedure is vital for improving the IR thermography measurement accuracy.

Temperature Difference (IR-TC)(oC)

12

10

5.4. Junction temperature measurement by forward voltage method

8

6

Δ T = 0.161T - 3.916 4

2

0 30

50

70

Heating Plate Temperature

90

110

(oC)

Fig. 7. Temperature difference between IR thermography and thermocouple obtained from a simple calibration experiment using a lucid epoxy plate.

thermography and thermocouple measurement. Accordingly, the IR thermography temperature measurement cannot be justified if the accuracy of the measurement results is not enhanced. In this study, the IR thermography results are further modified in accordance with Eq. (10) or the data shown in Fig. 7, which is alternatively termed the corrected IR thermography. The corrected results are also shown in Fig. 4. It is found that after the correction, the maximum temperature deviation in percentage between the thermocouple and corrected IR thermography measurements is significantly reduced down to less than 6%. The power fractions not converted into lighting are characterized from the estimated internal quantum efficiencies of the LEDs. They are further imposed in the thermal FE modeling as the heat generation rates of the LED chips. In addition, the calibrated surface emissivity of the transparent molding compound, i.e., 0.92, is also applied in the thermal analysis. The calculated surface temperatures associated with the LED packages are shown in Fig. 4. It is clear to see that the FE modeling results are very comparable not only to the thermocouple (TC) but also to the corrected IR thermography data, where, in specific, their maximum difference in

Fig. 9(a) demonstrates a current-temperature curve. It is found that the LED would generate significant self-heating at a sense current above 0.1 mA. The characteristic curves of the current and forward voltage of these three color types of LEDs are shown in Fig. 9(b). As the LED chips start to generate self-heating at 0.1 mA sense current, the forward voltage becomes highly nonlinear with the input current. According to the EIA/JEDEC standard, the sense current should be selected at the inflexion point of the currentevoltage characteristic curve. In the investigation, the measured ambient temperature for the measurement is 25.2  0.1
C. Since the maximum allowable design temperature of LEDs is about 120
C [23], the input current should be well controlled to avoid exceeding the limit. Fig. 10 shows the relations between the forward voltage and LED junction temperature for these three color types of LEDs. A linear relationship between them can be observed for these three types of LEDs. In addition, as the junction temperature increases from about 30
C to 100
C, the forward voltage decreases from 2.58 V to 2.39V for the green LED, from 2.46V to 2.32V for the blue LED and from 1.66V to 1.54V for the red LED. Linear curve fits of these data are made and shown in the same figure. The corresponding TSP coefficient is about 2.714  103 mV/
C for the green LED, 1.714  103 mV/
C for the red and 2  103 mV/
C for the blue. By these voltage-temperature curves, the junction temperatures of the LEDs under the driving powers shown in Table 3 are predicted and shown in Fig. 11, together with the measured surface temperatures by the corrected IR thermography. As can be observed in Fig. 11, a higher surface temperature tends to create a higher junction temperature. Accordingly, the green LEDs would comprise a higher junction temperature. The junction temperature of the G2 LED can be as large as about 100
C at a low-temperature environment (i.e., about 25
C). It can thus be expected to have an LED junction temperature beyond the design limit (i.e., 125
C in general) as the ambient temperature is raised up to above 50
C. For better thermal-mechanical reliability and optical performance, an extensive enhancement of the thermal performance of the white light module through effective

Fig. 8. The characterized surface temperature distribution by (a) IR thermography measurement before the temperature correction and (b) FE modeling.

H.-C. Cheng et al. / Applied Thermal Engineering 38 (2012) 105e116

a

b 15

60 50

Red Blue Green

12

40

Input Current (mA)

Input Current (mA)

115

30 20 10 0

9

6

3

0 0

2

4

6

8

10

12

14

0

1

Temperature Difference (oC)

2

3

4

Forward Voltage (V)

Fig. 9. (a) The current-temperature difference relationship and (b) The current-forward voltage relationship.

thermal management is needed. By further comparing the junction temperatures with the surface temperatures measured by the corrected IR thermography, it is not surprising to find that there is a temperature gap between them. What surprises is that the temperature gap can be as significant as about 20e30
C for the green LEDs, indicating that the molding compound has very poor thermal conductivity, thereby resulting in a great temperature gradient from the LED chip to the surface of the optical lens. This also suggests that heat conduction to the PCB, through leadframe, and then to the ambient would be the main heat dissipation path. Uncertainty analysis is subsequently performed on the forward voltage measurement. By the present measurement system, the uncertainty in the voltage measurement using a power supply is 5.0 mV and that in the ambient temperature measurement using a T-type thermocouple is 0.1
C. Based on these two uncertainty data together with the voltage-temperature curve shown in Fig. 10, the uncertainty εðkÞ in the calibrated TSP parameter is around 1.0116  104 mV/
C, 1.0107  104 mV/
C and 1.0159  104 mV/
C corresponding to the red, green and blue LEDs. At the forward voltage of 2.396V, 1.604V and 2.385V,

the corresponding measured maximum junction temperature associated with the red, green and blue LEDs is 98.4
C, 64.0
C and 72.3
C, and besides, the initial ambient (junction) temperature for the present measurements is about 31.2  0.1
C. Substituting these measured data together with the corresponding measurement uncertainties into Eq. (7) yields the worst possible uncertainty in the measured junction temperature using the forward voltage method. It is about 3.7
C for the green LEDs, 4.6
C for the red and 4.1 for the blue. Further combining the standard deviation of the linear fit of the voltage-temperature data, 0.2
C, leads to the maximum possible uncertainty of about 3.9
C, 4.8
C and 4.3
C. These uncertainties are about 4e8% of the corresponding measured temperatures. According to Eq. (7), there shows an inversely proportional relationship between the uncertainty and the TSP parameter. This confirms why the red LED, consisting of the minimum TSP parameter, tends to have the largest uncertainty. Apparently, these uncertainties are quite significant, which could be mainly due to the poor accuracy in the voltage measurement using the present power supply instrument. If the uncertainty in the voltage measurement can be reduced from 5 mV to 1 mV, the corresponding maximum

100

2.6

90

2.4

Red

2.2

Voltage (V)

Surface Temperature (oC)

80

Blue 2

Green

1.8

1.6

70 60 50 40 30

Forward Voltage (Junction Temp.)

20

Corrected IR Thermography (Surface Temp.)

10

1.4

0

20

40

60

Temperature

80

100

(oC)

Fig. 10. The forward voltageetemperature relationship of these three color types of LEDs.

R1

R2

R3

G1

G2

G3

B1

B2

B3

ID of LED Fig. 11. The measured junction temperatures of the LEDs using the forward voltage measurement.

116

H.-C. Cheng et al. / Applied Thermal Engineering 38 (2012) 105e116

possible uncertainty can be greatly improved to 0.95
C, 1.12
C and 1.06
C. 6. Conclusions This paper proposes two thermal measurement approaches, namely IR thermography for surface temperature measurement and forward voltage method for junction temperature measurement, to assess the thermal performance of the white light LED module in natural convection. The measurement results are extensively benchmarked by the 3D thermal FE modeling and thermocouple measurement. The emissivity coefficient of the transparent molding compound is also characterized through a simple calibration experiment. The optical analysis shows that the luminous fluxes of these three color types of LED packages increase with an increasing input power, and tend to become stable as the input power is large. It turns out that the green LED package presents the maximum luminous flux output, followed by the red and the blue. On the other hand, the output efficiencies show a totally opposite trend, where it decreases with the increase of input power. Under the same driving power, the blue LED would have the best output efficiency, followed by the red and green. By the specific driving power shown in Table 3, the output efficiency associated with the red, green and blue LED packages is 9.21%, 26.2% and 31.24%, and the associated packaging efficiency is about 90%, leading to the corresponding internal quantum efficiencies of about 10.3%, 29.2% and 34.9%. The calibrated emissivity coefficient of the transparent molding compound is roughly in the range of 0.90e0.94. In addition, because of having the largest driving power, the green LEDs tend to have a higher surface temperature among all the LEDs. Furthermore, among these 3 green LEDs, the G2 LED has the maximum surface temperature as a result of being fully surrounded by the other LEDs and having the largest driving power. A similar result is also found in the measured junction temperature using the forward voltage method. Furthermore, the measured surface temperature of the LEDs by the IR thermography tends to be overestimated, probably due to the transparency of the molding compound. Consequently, the detected IR radiation may be originated from the emission not only by the molding compound but also by the other components such as the LED chip and substrate, and also even from the reflection by the other surfaces, thus resulting in an overestimate of the surface temperature of the LEDs. The discrepancy tends to increase linearly with an increasing temperature. However, the IR thermography technique remains a feasible tool for LED surface temperature characterization when the measurement results are corrected using the presently proposed temperature correction procedure. The forward voltage measurement reveals that the junction temperature of the green LEDs can be up to about 100
C even in a low-temperature environment (i.e., about 25
C). Moreover, there is also a great difference between the surface and junction temperatures (i.e., as much as around 30
C), indicating that the molding compound is a main thermal barrier to keep the LED chips from heat dissipation. It is believed that application of a high thermal conductivity molding compound can be quite effective for improving the thermal performance of the module. Finally, the uncertainty analysis demonstrates that the worst possible uncertainty in the measured junction temperature using the forward voltage method is about 3.9
C, 4.8
C and 4.3
C associated with the red, green and blue LEDs, suggesting that the red LEDs tend to have the largest uncertainty as a result of its minimum TSP parameter. The great uncertainties can be mainly attributed to the poor accuracy of the voltage measurement using the present power supply instrument. A power supply with better

accuracy is required in order to improve the measurement uncertainty. Acknowledgements The work is partially supported by National Science Council, Taiwan, R.O.C., under grants NSC98-2221-E-007-016-MY3 and NSC98-2221-E-035-024. The authors would also like to acknowledge Dr. Shyh-Rong Tzan and Dr. Chin-Yin Yu, Material and Chemical Research Labs, Industrial Technology Research Institute, Hsinchu, Taiwan, for providing test samples and material data, and Mr. Kung-Hsiung Wang, Dept. of Power Mechanical Engineering, National Tsing Hua University, Hsinchu, Taiwan, for helpful discussion. References [1] M. Arik, S. Weaver, Chip scale thermal management of high brightness LED packages, Proceedings of the international Society for Optical Engineering 5530 (2004) 214e223. [2] Y. Wang, H. Xu, S. Alur, A.-J. Cheng, M. Park, S. Sakhawat, A. N. Guha, O. Akpa, S. Akavaram and K. Das, “Determination of Junction Temperature of GaNbased Light Emitting Diodes by Electroluminescence and Micro-Raman Spectroscopy”, CS MANTECH Conference, May 18e21, Tampa, Florida, USA, 2009. [3] D.C. Hall, L. Goldberg, D. Mehuys, Technique for lateral temperature profiling in optoelectronic devices using a photoluminescence microprobe, Applied Physics Letters 61 (1992) 384. [4] A. Corfa, A. Gasse, S. Bernabe, and H. Ribot, “Analytical and FEM Simulations of the Thermal Spreading Effect in LED Modules and IR Thermography Validation, ” EuroSimE 2010 conference, Bordeaux, France, April 2010. [5] J. Park, M.W. Shin, C.C. Lee, Measurement of temperature profiles on visible light-emitting diodes by use of a nematic liquid crystal and an infrared laser, Optics Letters 29 (2004) 2656. [6] Y. Xi, E.F. Schubert, Junction-temperature measurement in GaN ultraviolet light-emitting diodes using diode forward voltage method, Applied Physics Letters 85 (2004) 2163. [7] W.H. Chen, H.C. Cheng, H.A. Shen, An effective Methodology for thermal characterization of electronic packaging, IEEE Transactions on Components and Packaging Technologies 26 (2003) 222e232. [8] H.C. Cheng, W.H. Chen, H.F. Cheng, Theoretical and experimental characterization of heat dissipation in a Board-Level Microelectronic Component, Applied Thermal Engineering 28 (No. 5e6) (April 2008) 575e588. [9] J. Hu, L. Yang, M.W. Shin, Thermal and mechanical analysis of high-power LEDs with Ceramic packages, IEEE Transactions on Components and Packaging Technologies 8 (No. 2) (2008) 297e303. [10] B.-J. Huang, C.-W. Tang, Thermal-electrical-luminous model of multi-chip polychromatic LED luminaire, Applied Thermal Engineering 29 (2009) 3366e3373. [11] Y.S. Chan, S.W. Ricky Lee, Spacing optimization of high power LED arrays for solid state lighting, Journal of Semiconductors 32 (No. 1) (2011) 014005-1014005-5. [12] W. Jan, T.-F. Liao, T.P. Chen, C.S. Chang, AlGaInP light-emitting diode with metal reflector structure, Proceedings of the international Society for Optical Engineering 5739 (2005) 81e85. [13] CIE Publication 127, Measurement of LEDs, second ed. HIS. Inc, 2007. [14] Á Borbély, S.G. Johnson, Performance of phosphor-coated light-emitting diode optics in ray-trace simulations, Optical Engineering 44 (111308) (2005). [15] EIA/JEDEC Standard, Integrated Circuits thermal test method environment conditions-natural convection (Still air)”, EIA/JESD51-2 (1995). [16] W. Shockley, Electrons and Holes in Semiconductors with Applications to Transistor Electronics, D. Van Nostrand Company, New York, 1950. [17] B.-J. Huang, C.-W. Tang, M.-S. Wu, System dynamics model of high-power LED luminaire, Applied Thermal Engineering 29 (No. 4) (2009) 609e616. [18] EIA/JEDEC Standard, Integrated Circuits thermal measurement methodelectrical test method (Single Semiconductor device)”, EIA/JESD51-1 (1995). [19] P. Holman, Experimental Methods for Engineers, sixth ed. McGraw-Hill, Inc., 1994. [20] G.N. Ellison, Thermal Computations for Electronic Equipment, R. E. Krieger Publishing Company, Malabar, FL, 1989. [21] G. Ridsdale, B. Joiner, J. Bigler, V.M. Torres, Thermal simulation to Analyze design features of Plastic Quad Flat package, Journal of Microcircuits and Electronic Package 19 (1996) 103e109. [22] H.C. Cheng, Y.-H. Tsai, K.-N. Chen, J. Fang, Thermal placement optimization of Multichip modules using a Sequential Metamodeling-based optimization approach, Applied Thermal Engineering 30 (No. 17e18) (2010) 2632e2642. [23] O. Kuckmann, High power LED arrays Special Requirements on packaging technology, Proceedings of the international Society for Optical Engineering 6134 (2006) 613404(1)-613404(8).