Thermal Management in Color Variable Multi-Chip LED Modules Theo Treurniet and Vicky Lammens Philips Lighting P.O.Box 80200, 5600 JM Eindhoven, The Netherlands [email protected] Abstract One of the main advantages of LED technology is the possibility to build efficient color variable lighting systems. In order to apply these systems in illumination applications high power, color variable LED modules are developed. These modules contain different LED colors, e.g. red, green and blue dice, which can be combined to generate colored and white light. For optimal color mixing, the dice of the different colors are mounted close to each other, resulting in an increased thermal load of the LED dice in the module. One of the key issues in color variable LED applications is to maintain a defined color point over life and over all flux levels. The relevant optical properties of LEDs, like luminous flux and wavelength, depend on the junction temperature of the LED. One of the ways to maintain a stable color point is to compensate for these temperature dependencies. In order to be able to perform this compensation, one has to know the junction temperature of the different dice under all loads and the relation between the luminous flux and the wavelength and this junction temperature. In this paper we present a thermal design method of a multi-chip LED module that is able to handle an increasing thermal load up to 20 Watt. Further, we present a compact model to estimate the junction temperature of the different dice at an arbitrary load. This model is used in the color control system to calculate the junction temperatures under load and adjust the loads of the different dice accordingly to maintain a defined color point. The model is validated with transient tester measurements. Application of this model results in a considerable improvement in color stability. Keywords LEDs, multi-chip, color variation, compact model Nomenclature θ temperature, ºC ∆θ temperature difference between junction and sensor, ºC P power, W R thermal resistance, ºC/W k thermal coefficient of the LED forward voltage, mV/ºC CTE coefficient of thermal expansion, 1/K The subscripts R, G, B and A indicate the power or temperature of the red, green, blue or amber die. 1. Introduction In view of recent increases in LED performance, LEDs are expected to penetrate general lighting applications in the near future. However, even at the current and future higher efficiency levels, a considerable amount of the electrical energy will be converted in heat. At the present level of 40 1-4244-0154-2/06/$20.00 ©2006 IEEE

173

lumen/W for white light generation, in the order of 15% of energy is converted into visible light. The remainder 85% is converted into heat. With fluorescent lamps an efficacy of 100 lumen/W can be achieved. This level is considered achievable also for high power LEDs in the near future. At the efficacy level of 100 lumen/W, still 70% of the energy will be converted into heat. In table 1, an overview of different frequently used light sources is given, together with the amount of power needed to replace that source with LEDs. This table shows that with the current LED performance, in the order of 10 – 100 W heat is generated when a conventional light source is replaced with LEDs. With a performance of 100 lm/W, these loses are reduced considerably. However, especially in cases where a small light source is required for on optimal optical performance, so called etendue critical applications, a large amount of heat is generated in a small area. This means that even in case of more efficient sources, the thermal issues in LED light sources will be considerable. The reliability of electronic components, including LEDs, decreases at high operating temperatures. The current generation of high power LED can be operated at junction temperature of 135 °C or even higher. However not only the reliability but also the performance decreases with increasing temperature. In the figures 1 and 2, the relative light output is plotted as a function of the junction temperature for the two main material systems, used in LEDs. Off these two material systems, the InGaN system for blue, green and white LEDs and the AlInGaP system for red and amber LEDs, the latter one is the most sensitive to temperature variation. In this paper, we consider multi-chip LED modules. These modules contain multiple LED chips of different colors and can be used to generate color variable light as describe by C. Hoelen et.al1. The color point of the light is determined by the contribution of the different dice. Due to the temperature dependence of the die performance, the resulting color point also depends on the die temperature. This dependence is enhanced by the fact that not only the luminous flux, but also the wavelength of the LEDs is temperature dependant. This means that a control system is needed in order to maintain a stable color point, as described by P. Deurenberg et.al.2 For some of the control systems, an estimate of the junction temperature is needed. This paper describes the thermal design of a color variable LED module, which is designed for optimal thermal performance. Further a method is described to estimate the junction temperature of the different dice. This method shows similarities with a compact model approach3 and uses a temperature sensor that is mounted in the module as a reference temperature. This model is validated with transient thermal tester measurements. 22nd IEEE SEMI-THERM Symposium

Light source

Application

Efficacy (lm/W)

Flux (lm)

LED power @ 40 lm/W Input 85% Loss (W) (W) 60W GLS Home 12 720 18 15.3 100W GLS Home 12 1200 30 25.5 11W CFL-I Home 65 720 18 15.3 50W Halogen Shop 18 900 22.5 19.1 90W Halogen Shop 18 1620 40 34 58W TL Office 90 5220 130.5 110.5 Table 1: An number of widely used light sources and the amount of power needed to replace these sources with LED based sources, for LEDs with an efficacy of 40 and 100 lm/W. Input: electrical input in the system. Loss: thermal loss, not converted into light.

Figure 1: Relative light output as a function of the junction temperature of the InGaN material system. (From Lumileds Luxeon Emitter Technical Datasheet DS25)

consume in the order of 0.8 W. The blue and green die have a slightly higher forward voltage and consume in the order of 1.2 W. The resulting maximum power, consumed by the module is about 4 W. Note that these powers can differ from module to module, since the forward voltages might differ from chip to chip. Further, the 4 W is a maximum; depending on the color point the maximum power will be lower. For optical reasons, the dice are mounted close together. This allows smaller optics for beam shaping and improves the color mixing. However, the thermal load on the module is further increased. At die level the flux density is in the order of 1 W/mm2. When we assume the system to be cooled by natural convection a flux level at the heat sink surface of about 200 W/m2 can be reached. This means that the heat should be spread about a factor of 5000 in the system. Therefore, heat spreading should be the main thermal function in a high power, etendue critical LED module. Due to the high flux density, close to the dice, combined with the required electrical insulation of the substrate, the dice are mounted on a Aluminum Nitride submount. This submount on is mounted on an AlSiC heatspreader, which combines good thermal properties with a CTE close to the CTE of AlN. This reduces the risk of mechanical failure of the module. Using a CTE matched material avoids cracking of the submount as well as delamination of the interconnect between the submount and the heatspreader. Conductive glue is used; both as 1st level interconnect between the die and the submount and as 2nd level interconnect between the submount and the heatspreader. In figure 3, a cross section of the module is sketched. LED power @ 100 lm/W Input 70% Loss (W) (W) 7.2 5 12 8.4 7.2 5 9 6.3 16.2 11.3 52.2 36.5

lens bondwire interconnect

thermal sensor AlN AlSiC heatsink/luminaire

Figure 3: Sketch of the cross section of the module

Figure 2: Relative light output as a function of the junction temperature for the AlInGaP material system. (From Lumileds Luxeon Emitter Technical Datasheet DS25) 2. Thermal design In this paper, we consider the thermal performance of a color variable LED module that contains 4 chips, 1 red, 1 green, 1 blue and 1 amber die. These dice can be driven at a current of 350 mA. At that current, the red and amber die Treurniet, Thermal management in Color Variable …

In order to measure the temperature close to the dice, a thermal sensor is also mounted on the submount. In this case a NTC resistor is used as sensor. The submount is covered with a COC lens that is filled with a Silicone based encapsulant material. Since the thermal conductivity of the lens material and the surrounding air is 3 orders lower than the substrate materials, the thermal resistance of the path via the optics is large and the amount of heat that is transferred via the lens can be neglected.

22nd IEEE SEMI-THERM Symposium

3. Model to estimate the junction temperature As described in the introduction, an estimate of the junction temperature is needed in order correct the color shift of the module, due to heating of the dice. In order to do that, we need a relation between the measured temperature of the sensor and the heating of the dice, due to the heat, generated at junction level. Under the assumption that the thermal conductivity is constant, conduction equation to describe the relation between temperature and power are linear. Therefore, the relation between the thermal power of the dice and the temperature of the different dice can be describe with the following linear system,

⎛θ R ⎞ ⎛ RRR ⎜ ⎟ ⎜ ⎜θ G ⎟ ⎜ M ⎜ θ ⎟ = θ sens + ⎜ M ⎜ B⎟ ⎜ ⎜θ ⎟ ⎜R ⎝ A⎠ ⎝ AR

RRG

L

RGG RBB K

K

RRA ⎞⎛ PR ⎞ ⎟⎜ ⎟ M ⎟⎜ PG ⎟ , (1) M ⎟⎜ PB ⎟ ⎟⎜ ⎟ R AA ⎟⎠⎜⎝ PA ⎟⎠

where θR, θG, θB, θA and θsens are the temperatures of respectively the red, green, blue and amber junction and the temperature of the sensor. PR till PA are the thermal powers of the different dice. This description has the shape of a thermal resistance network, where the thermal resistance matrix R gives the relation between the power and the different junction temperatures, relative to the measured sensor temperature. A schematic representation of the network is given in figure 4.

sensor Figure 4: Schematic representation of the interaction between the different dice on the common submount This description also shows similarities with a compact model approach3, which describes the thermal behavior of an electronic device by means of a resistance network. The aim of a compact model is de describe the thermal behavior of the internal nodes correct, independent of the boundary condition that is applied to the boundary nodes of the network. The similarity with this approach is that a thermal resistance network describes the thermal behavior. The difference is that instead of one or more boundary nodes, a reference node is Treurniet, Thermal management in Color Variable …

used. This node represents the measured temperature of the integrated thermal sensor. We can define the temperature difference ∆θ relative to the sensor temperature as

⎛ ∆θ R ⎞ ⎛ θ R ⎞ ⎜ ⎟ ⎜ ⎟ ⎜ ∆θ G ⎟ ⎜θ G ⎟ ⎜ ∆θ ⎟ = ⎜ θ ⎟ − θ sens ⎜ B⎟ ⎜ B⎟ ⎜ ∆θ ⎟ ⎜ θ ⎟ ⎝ A⎠ ⎝ A⎠

(2)

When we assume the different sources in the network to be point sources, the thermal resistance matrix will by symmetric. However, in practice this is not the case. Figure 4 shows that the dice of the different color differ in size. Red and amber are the same size. Green and blue are also the same size, but a bit larger than red and amber. This means that the matrix does not have to be symmetric, since the junction temperature is defined as the average temperature over the junction. E.g. the average heating of the green die by the amber die might be different than the reverse, the heating of the amber die by the green one, due to the differences in size of the green and the amber dice. The interaction between the red and amber and between the blue and green dice is expected to be symmetric. The diagonal elements of the matrix describe the selfheating of the dice relative to the sensor temperature. Powering one LED and measuring the temperature of all the dice and the sensor can determine the coefficients of R. When this procedure is performed for all 4 colors, all coefficients can be determined, due to the fact that these 4 measurements are independent of each other. 4. Measurement method and results These measurements are performed with a Micred thermal transient tester (T3Ster)4. It is known from the literature5,6 that transient tester measurements on LEDs are not straightforward. Known issues are a dependence of the thermal resistance from junction to ambient of the driving current. This is an effect of an electrical resistance in series with the die and the fact that the optical efficiency depends on the current density and the temperature. These effects are not taken into account in this analysis. Further, the power that is mentioned is electrical input powers, not corrected for power that is converted into visible light. Since all measurements and calibrations are performed at the same temperature and driving currents, the error in this analysis is limited. However for improved accuracy, these effects should be included. The sensor that is integrated in the module is a NTC sensor. The resistance of the NTC is determined with an Agilent multimeter via a 2-points measurement. Since the NTC has a typical resistance of 10 kΩ at 60 °C, the resistance of the wires will be small and taken into account with the calibration. The module is mounted on a thermostat. This thermostat consists of a 40x40x10 mm aluminum sample holder that is either heated or cooled with a peltier element. The system is controlled with a Keithly 2510 TEC controller. The pt-100 sensor that is used to control the temperature is mounted in 22nd IEEE SEMI-THERM Symposium

the center of the aluminum sample holder. This temperature is used as reference for the calibration. During calibration, the forward voltage of the different dice is measured at a sensor current of 4 mA. Further, the NTC resistance is determined at the same temperatures, 20, 40, 60 and 80 °C. After calibration, the samples are measured four times, where for each measurement a different channel was driven at 350 mA. This is the nominal current for the dice. The sample holder temperature is maintained at 60 °C, which is a realistic housing temperature in an application. 9 samples where measured in order to study the spread between different samples. The spread between the different samples is small. The minimum and maximum values of the diagonal element of the thermal resistance matrix differ in the order of 10% of the average, which can easily be explained by issues in the processing of the different modules. The results of one sample are summarized below. Table 2 shows the thermal coefficient of the forward voltage for the different color channels. Note that there is a big difference between the different colors. However, the values are quite consistent over the different samples. However, for an accurate measurement, this value has to be determined via calibration for all the dice in the system. k @ 4 mA (mV/°C) Red Green Blue Amber 1.545 3.210 1.960 1.816 Table 2: Thermal coefficient of forward voltage at the measurement current for the different color channels Table 3 shows the input power and measured temperature differences for the different channels. The temperature difference of the channel that is driven at 350 mA is indicated with a bold number and are larger, due to self-heating of the dice. P @ 350 mA (W)

⎛ 21.05 ⎜ ⎜ − 0.63 R=⎜ 0.15 ⎜ ⎜ 0.55 ⎝

0.67 5.81 1.00 2.16

1.64 0.49 ⎞ ⎟ 1.01 0.24 ⎟ °C/W 4.92 − 0.85 ⎟ ⎟ 0.84 23.27 ⎟⎠

From this matrix, we can conclude a few things. The selfheating of the red and amber dice is much higher, compared to the green and blue dice. This can be explained by the internal structure of the red and amber chips, which have the active layer on top of the dice, while the active layer of the green and blue dice is locate in the lower part of the dice, just above the die attach. The interaction between the amber and red dice, indicated with the red circle, is quite symmetric, as is the interaction between the green and blue dice, indicated with the blue circle. This is expected as they are of the same size and positioned symmetric with respect to the NTC sensor. The interaction between the dice of different type is not symmetric. When we assume the power of the LEDs to be distributed equally over the dice with 1W per die, the temperature difference relative to the NTC is:

⎛ ∆θ R ⎞ ⎛ 23.8 ⎞ ⎜ ⎟ ⎜ ⎟ ⎜ ∆θ G ⎟ ⎜ 6.4 ⎟ ⎜ ∆θ ⎟ = ⎜ 5.2 ⎟ °C. ⎜ B⎟ ⎜ ⎟ ⎜ ∆θ ⎟ ⎜ 26.8 ⎟ ⎠ ⎝ A⎠ ⎝ When we neglect the interaction between the different dice and only take into the diagonal element of the resistance matrix, which represent the self-heating of the dice, the temperature difference relative to the NTC is:

⎛ ∆θ R ⎞ ⎛ 21.0 ⎞ ⎜ ⎟ ⎜ ⎟ ⎜ ∆θ G ⎟ ⎜ 5.8 ⎟ ⎜ ∆θ ⎟ = ⎜ 4.9 ⎟ °C. ⎜ B⎟ ⎜ ⎟ ⎜ ∆θ ⎟ ⎜ 23.3 ⎟ ⎠ ⎝ A⎠ ⎝

∆T junction – NTC This means that neglecting the ∆θ (°C) interaction between the different dice and Red Green Blue Amber only considering the self-heating relative to the NTC temperature underestimates the Red @ 350 mA 0.761 0.12 0.42 16.02 -0.48 temperature 0.3 – 3.6 °C, which is in the Green @ 350 mA 1.206 0.81 1.21 2.61 7.01 order of 13% of the total temperature Blue @ 350 mA 1.126 1.84 1.14 0.94 5.54 difference. Amber @ 350 mA 0.825 0.40 0.20 -0.70 19.20 This simplified model might be Table 3: Power step and temperature difference between acceptable in certain circumstances, since an error of 5 °C is NTC and junction. The bold numbers indicate the temperature considered acceptable for the color control algorithm to difference, due to self-heating of the dice. maintain the color point within specification. However, when we increase the power levels, this simplified model will not The negative numbers show that the junction temperature be acceptable anymore. of the green and blue dice, sometimes are lower than the In the analysis above, all temperatures are determined sensor temperature. relative to the sensor temperature. Due to the thermal design From these measurements we can construct the thermal of the module, this temperature only rose a few degrees above resistance matrix: the temperature of the sample holder. This means that the absolute temperature of the red and amber junctions stayed Treurniet, Thermal management in Color Variable …

22nd IEEE SEMI-THERM Symposium

below the 90 °C and less than 30 °C above the sample holder. For green and blue, the junction temperature was even lower, below 70 °C and less than 10 °C above the sample holder, which is an acceptable thermal performance for this module. 5. Conclusions A method to estimate the junction temperature of the dice in a multi-chip color variable LED module is presented. This method uses a model, similar to a compact model, to determine the temperature relative to a reference temperature. Since this reference temperature is measured in the module with a NTC, it is possible to determine the junction temperature of the dice in the module. This junction temperature can be used in a control system to stabilize the color point of the module. References 1. C. Hoelen et.al., Multi-chip color variable LED spot modules, Proc. SPIE, 2005 2. P. Deurenberg et.al., Achieving color point stability in RGB multi-chip modules using various color correction methods, Proc. SPIE, 2005 3. C.J.M. Lasance, H. Vinke, H. Rosten and K.L. Weiner, A novel approach for the thermal characterization of electronic parts, Proc. of the 11th SEMITHERM Symposium, 1995 4. www.micred.com/t3ster.html 5. G. Farkas, Q. van Voorst Vader, A. Poppe, Gy. Bognár: Thermal investigation of high power optical devices by transient testing, Proc. of the 9th THERMINIC, Aix-enProvence, France, 2003 6. Gábor Farkas et.al, Electric and Thermal Transient Effects in High Power Optical Devices, Proc. of the 20th SEMITHERM Symposium, 2004

Treurniet, Thermal management in Color Variable …

22nd IEEE SEMI-THERM Symposium