Experimental Determination of Thermal Couplings in Packages Containing Multiple LEDs

Abstract

Light Emitting Diodes (LEDs) are the most commonly used light sources. Temperature strongly affects their operation. Considering that multiple devices are often placed in a single housing, thermal couplings between devices become important. This problem is illustrated here based on the example of a light source containing four LEDs in a single package. Thermal analyses are carried out based on measured transient temperature responses. The measurement results are processed employing the Network Identification by Deconvolution method. The obtained results demonstrate clearly that depending on the device mounting manner and applied cooling condition the temperature rise value induced in neighboring devices can exceed 70% of the rise in the heating diode. Consequently, thermal models of such LEDs should consider not only self-heating effects, but also thermal interactions with the other diodes.

1. Introduction

Light Emitting Diodes (LEDs) are commonly present in our everyday lives, including automotive applications or public space and household lighting. Their operation involves different physical phenomena; electrical optical or thermal ones [1]. Temperature is one of the most important factors influencing LED properties [2,3], which is confirmed in the recently developed standards [4,5]. Consequently, LED thermal models are now essential subcomponents of their multidomain models [6]. The intensive research on the modeling and design of luminaires containing LEDs led recently to the development of the Delphi4LED methodology employing the concept of so-called digital twins [7,8].

High LED junction temperature degrades devices and leads to parametric failures, such as lumen depreciation and chromaticity changes, or even catastrophic failures, e.g., die delamination and broken connections resulting from thermal stress [9,10]. Moreover, considering that lighting components are frequently manufactured as assemblies containing several LEDs soldered to a common substrate and enclosed in a single housing or package, thermal interactions between individual diodes play an important role.

Thermal couplings between electronic devices were studied already in stacked multi-chip modules [11], IGBTs, and diodes in power converters [12] as well as in LED-based luminaires [13,14,15,16], where it was observed that in such cases device temperature values strongly depend on the spatial arrangement of components and the number of dies used. A particularly interesting conclusion drawn in [13] was that the total thermal resistance of multi-chip packages tends to decrease with the number of chips due to the parallel heat dissipation, which seems to justify the application of multi-die LEDs. Previous research done by the authors presented in [17] demonstrated that when discrete LEDs are soldered 2–3 cm away one from another to the same substrate cooled by free convection without any heat sink, thermal couplings between devices might reach 50%, i.e., temperature rise values in neighboring LEDs attain half of the value measured in the heating one.

Here, thermal couplings between devices have investigated a case when four LEDs are enclosed in a single package soldered to a PCB and cooled either by natural convection or with forced water cooling applied. The following section of this paper briefly introduces the investigated LEDs. Then, the measured device dynamic temperature responses are presented and analyzed. Next, the thermal time constant spectra and cumulative structure functions are computed, which allowed for gaining a deeper understanding of thermal couplings occurring between the devices inside the package. Finally, the most important findings are summarized.

2. Investigated Devices

The devices investigated in this paper are the MCE4WT-A2-0000-JE5 LEDs belonging to the XLamp^® family manufactured by Cree. The considered LED device could operate at a forward current I_Fmax equal to 700 mA, total power of 2.8 W, and emitted luminous flux Φ_V = 100 lm at forward current I_F = 350 mA. The viewing angle of optical radiation is 110° and the thermal resistance between junction and solder point R_thj-s for white type MCE power LED is typically equal to 3 K/W [18]. Each of these devices contained in a single package four white power LEDs, as shown in the microscopic photo in Figure 1a. Owing to the reduced distance between diode dies, compared to discrete LEDs the MC-E devices provide in one package small optical sources of high-lumen output. Consequently, they can reduce the LED system complexity by reducing the number of components required [18]. The investigated diodes were soldered to 35 mm square Metal Core PCBs (MCPCBs) NT-1RB35I-MCE [19], shown in Figure 1b, made of 5052 Alloy and designed expressly for the MC-E devices by NIVISS. The MCPCBs are 2 mm thick and covered with a 60 μm thin layer of dielectric having a thermal conductivity of 2 W/(m*K).

3. Measurement Results

The main goal of the experimental research presented here was to demonstrate the significance of thermal couplings between devices contained in the same package. This problem is very important because the heat generated in neighboring devices increases LED temperature and consequently affects its operating point changing the efficiency and luminous flux.

Another goal was to determine the influence of thermal pads on device temperature. For this reason, two different LED modules were measured, one the device thermal pad properly soldered (WTP), and the thermal pad left unconnected (NTP). The examination of this case was particularly interesting from the reliability point of view because it might be helpful in the determination of the optimal heating time during the LED die to attach and solder quality assessment, hence contributing to the reduction of the overall testing time [20,21].

The LED dynamic temperature responses were measured with the transient thermal tester T3Ster^®, manufactured by Siemens, which has recently become a standard solution for the thermal characterization of electronic devices [22]. The measurements were taken for each module when one LED was used both as a heater and a sensor and the other ones served only as temperature sensors. The curves measured in the sensing diodes were almost overlapping. In particular, the time instants when the temperature starts to drop differed just by some 4 ms, and the steady state temperature value differences did not exceed 2 K in any mounting and cooling case. Furthermore, swapping diode roles also produced similar results and the differences did not exceed 0.2 K for all time instants. Thus, for the sake of chart clarity, in the remainder of this paper only the curves obtained for the diodes located on the diagonal will be presented and analyzed here, where the heating and sensing diode is denoted by DH and the sensing one by DS as it is indicated in Figure 1a.

Before the actual measurements, all the diodes were carefully calibrated on a cold plate, through which the water flow was forced, and the liquid temperature was stabilized by a thermostat. The calibration results obtained for the sensing forward current of 10 mA are shown in Figure 2. Hereafter, the charts on the left will represent results obtained for the diode DH and the ones on the right for the diode DS. To facilitate the comparison of results obtained for the devices, the scales on chart axes were kept the same in both columns. The measured values (MES) are shown in Figure 2 by the markers and the fitted calibration curves (FIT) by the solid and dashed lines. The second-order polynomials were used for the fitting to attain the determination coefficient R² values higher than 0.999. For the calibration current of 10 mA, the diode voltage decreases with temperature with the linear component ranging from −2.45 mV/K to −1.70 mV/K.

After the calibration, the devices were heated first by the diode DH with the forward current of 1.0 A until the thermal steady state was reached and then the cooling curves were registered, now forcing through the LEDs the current value used previously during their calibration. For both LED modules, the measurements were repeated in two types of cooling conditions illustrated by photos in Figure 3. For natural convection cooling (NC) the PCBs were suspended horizontally in the air in the thermally insulating clamps without any heat sink attached. With the forced convection cooling (FC) the substrate was squeezed tightly to a cold plate, again with some thermally insulating clamps, applying thermal paste. The cooling water temperature flow was forced by a thermostat and all the time maintained at the pre-set value of 20 °C.

The LED dynamic temperature responses were recorded at the instants equidistant on the logarithmic time scale, to capture all the thermal time constants present in the responses. The measurement results are presented here in Figure 4 as heating curves, which were obtained from the previously recorded cooling curves by subtracting them from the respective steady-state temperature values. The dashed lines are used here for the module with the unconnected thermal pad and the double lines correspond to the case of forced water cooling.

Looking at the figure, it could be observed that the temperature response develops in the heating diode DH instantaneously, whereas the sensing diode DS responds only after approximately 20 ms. For both diodes, the temperature rise values in the steady state are similar when the thermal pad is not connected and the module is water cooled, and when the pad is connected but the module is cooled by the natural convection, but the dynamics of thermal processes is visibly different. When the thermal pad is not connected the curves diverge after some 100 ms, but the effects of altered cooling conditions become obvious only after several seconds.

The thermal coupling between the devices is definitely not negligible and it amounts almost to 80%, comparing the steady-state temperature values. The only exception is the case when the thermal pad is used and the modules are water-cooled, then the coupling is less than 40%, hence confirming the important influence of cooling conditions. The use of a thermal pad lowers the LED temperature rise, by at least 30%, and in the case of forced water cooling when the thermal pad is soldered the temperature rise is reduced by almost four times.

Analyzing the temperature rise values, the DS diode in any considered case is almost exactly 15 K cooler than the DH one. Then, when all four LEDs are active the temperature rise value with natural convection cooling would be around 160 K even if the thermal pad is properly soldered, hence assuming that the device is operated at room temperature the datasheet value of maximally allowed junction temperature of 150 °C would be exceeded. For forced water cooling, which is rarely used in practice, the total temperature rise value would amount just to 50 K. The measurement results discussed in this section will be explored in more detail in the following section.

4. Thermal Analyses

The measured curves were processed with the T3Ster Master software implementing the Network Identification by Deconvolution (NID) method [23], which has been recently adopted as a standard thermal analysis method offering a wide range of tools allowing deeper insight into thermal phenomena occurring in electronic systems. According to the current standard [4], the real heating power was used as the input quantity, which was found as the difference between the measured electrical and optical power. The optical power was determined using the methodology described in [24], where the light intensity was measured first in a light-tight box at the specified distance directly over the device and then the emitted optical power was computed based on the angular spatial radiation pattern provided in the datasheet by the device manufacturer.

First, based on the previously analyzed device temperature responses, the thermal time constant spectra were computed by performing the deconvolution as required by the NID method. The resulting curves shown in Figure 5 were obtained by integrating in time the classic spectra mentioned in the standard, to expose the short thermal time constant components. Consequently, the units in the left axis are given in K/W. The minima in these curves theoretically indicate the change of material in a heat flow path.

Looking at the figure, it could be concluded that the thermal response in the sensing diode DS is delayed and the short thermal time constants below 100 ms are not present in the spectra. On the contrary, the spectra corresponding to the heating diode DH have significant components already in the microsecond range, with distinct maxima just above 1 ms. For both diodes, leaving the thermal pad unsoldered shifts the peak in the middle from some 0.2 s to 1.0 s, increasing at the same visibly the thermal resistance. Then, the minimum located in Figure 5a at 30 ms could indicate the solder point location in the heat flow path. The curves computed for both diodes agree very well in the range of long-time constants with their dominant peaks at 15 s for the forced water cooling and 70 s for the natural convection. Looking from the device testing perspective, these results indicate that the sufficient heating time during tests, determined as three times the values of thermal time constants indicated by the vertical arrows in Figure 5a, would be 100 ms to check the quality of the attached die and almost 20 s to test the solder joint.

Analyzing the thermal resistance values accumulated at the solder point location indicated in Figure 5a, they range from 4.3 K/W to 5.3 K/W and are noticeably higher than the thermal resistance value of 3 K/W provided in the datasheet. On the other hand, the thermal resistance of the middle peak associated with the interface to the MCPCB varies for both diodes from 3.0 K/W to 4.5 K/W, when the thermal pad is soldered, and 11.4 K/W to 13.4 K/W in the other case. The difference between the considered device mounting methods is 8.5 ÷ 8.9 K/W. When forced water cooling is applied, the spectral component associated with the rightmost peak amounts to only 0.1 ÷ 0.2 K/W in the WTP case and 1.1 ÷ 1.2 K/W in the other one. With natural convection cooling, the two large peaks visible on the right contribute 12.1 ÷ 13.6 K/W and 11.8 ÷ 12.5 K/W for the two respective device mounting manners. Always the partial thermal resistance values obtained for the sensing diode are slightly lower than for the heating one.

Applying the electro-thermal analogy, the time constant spectra could be regarded as the continuous representation of the Foster RC ladder, where the heat dissipated in the LED junction flows through the package and MCPCB towards the ambient (the thermal ground). The time constants are equal to the products of respective thermal resistances and capacitances. The current represents the heat flow and the voltages correspond to the temperature rise over the ambient. Unfortunately, the resistance and capacitance values in the Foster RC ladders cannot have any physical interpretation [25,26].

Therefore, to assign some physical interpretation to thermal resistances and capacitances, the Foster RC representation has to be converted into the mathematically equivalent Cauer RC model, which might be already physically correct because all the partial capacitances in the heat flow path are connected to the thermal ground. Then, heat flux does not reach the ambient until the thermal capacitances are charged, so the speed of heat diffusion is limited. This conversion can be carried out by executing the continued fraction algorithm [26,27].

The conversion results can be presented in the form of cumulative thermal structure functions pictured in Figure 6. These curves show accumulated thermal resistance and capacitance from the diode junction (the origin of the coordinate system) to the ambient (the steep vertical sections at the line ends). Except for the previously discussed influence of the device mounting manner and cooling conditions on the thermal resistance, which shifts the curves horizontally, it could be observed that there exist two large plateaus. The first of them, located just below 0.1 J/K, thou present in all the curves, is more pronounced, when the thermal pad is not soldered. This is due to the fact that in this case the evacuation of generated heat from the package is hampered and it spreads through its entire volume. Consequently, these flat sections could be attributed to the thermal capacitance of the devices and their packages. The other one, at around 8 J/K, appears only when the devices are cooled by natural convection, and the above-mentioned value of thermal capacitance corresponds to the MCPCB. With the forced convection cooling heat diffuses immediately to the cold plate and the heat transfer occurs only in a small part of the substrate volume.

5. Conclusions

This paper showed that, unlike in the case of individual LEDs mounted on a common PCB [17], thermal couplings among the diodes are very significant when they are placed in a single package. Therefore, thermal models of such devices have to take into account not only the heat generated in a single LED, but also in all neighboring diodes. Moreover, the measurement and thermal analysis results presented in this paper demonstrated both the importance of using thermal pads and the influence of cooling conditions on the LED operating temperature. The approach proposed in this paper allowed in all the considered cases the proper identification of thermal resistances in particular heat flow path sections and their contribution to the overall device temperature rise. Moreover, the proposed analysis method renders it possible to consider each diode, both the self-heating phenomenon and the mutual thermal couplings between each pair of LEDs. The couplings result from the fact of sharing a common package base.

The methodology of analyzing dynamic thermal responses presented in this paper may help determine the proper device testing time and accelerate the testing phase. The problem of increasing the test station throughput is very important, especially in industrial high-volume testing. Equally important was the demonstration of significant thermal couplings occurring in multi-chip LEDs. The proper analysis of thermal couplings is crucial for the prediction of junction temperature and increasing the device reliability by slowing down their degradation leading to parametric and catastrophic failures. These considerations may be particularly interesting for the designers of LED luminaire control and cooling systems.