Thermal management in flexible substrates for LEDs Tapaninen Olli, Ollila Jyrki, Juntunen Eveliina and Keränen Kimmo VTT Technical Research Centre of Finland, Kaitoväylä 1, 90570 Oulu, Finland [email protected], [email protected], [email protected], [email protected] Abstract Evolution of lumens per watt efficacy has enabled exponential growth in light-emitting diode (LED) lighting applications. Small size of the LED has given the opportunity to integrate light sources in luminaires of many shapes and sizes. Flexible printed circuit boards enable low cost and easily integrable modules. However thermal management becomes important as the heat load on the board increases. In this paper we compare different thermal via techniques with thermal slugs as heat management structures. We present their challenges and benefits with measurements and simulations. Keywords: Light emitting diode, thermal vias, flexible substrates Introduction

Test Structure Design

The rapid development of efficient high brightness light emitting diodes (LEDs) has fueled the currently ongoing LED lighting revolution [1, 2, 3, 4]. Applications for general lighting need more and more high power components to provide desired luminosity levels. Small size of the LEDs provides an opportunity to design luminaires of innovative shapes and sizes. Flexible substrates enable these designs [5] when regular rigid printed circuit boards (PCBs) are not an option. Manufacturing of printed flexible substrates can also reduce material costs and manufacturing time when produced in large volumes in roll to roll process compared with standard PCBs. Because the LED does not emit heat radiation like the traditional incandescent bulb, the heat generated by the LED must be conducted away through the substrate [2]. This introduces a lot of thermal stress to the flexible structure and the luminaire housing. The challenge of LED -board and -module manufacturers is to provide a good thermal path from component to the luminaire [4, 6]. Thermal issues are significant in flexible substrates because the materials used are typically polymers of low thermal conductivity. Materials with higher conductivity exist but they are typically expensive and difficult to purchase. In this paper we explore thermal management techniques such as thermal vias, thermal slugs and heat spreading layers typically applied to ceramic substrates [2, 7, 8, 9] and regular PCBs [10, 11]. More recently different metal core circuit board manufacturing methods to thermally connect the LED directly to the core metal have also been introduced with enhanced thermal performance [12, 13, 14, 15, 16] . We apply these techniques to flexible substrates and present the challenges and benefits.

The test structures were designed to reflect 6 different thermal management techniques. The types of structures are listed on Table 1 and shown in Figure 1. The amount of vias was chosen to provide three different surface areas for heat conduction. The via-configurations occupy 10 % and 21 % of the area of the thermal pad of the Oslon LED as shown in Table 1. The no-via option and the slug-option act as a reference point to 0 % and 100 % coverage of the area of the thermal pad respectively. The layouts of the via structures are shown in Figure 1. Table 1: Tested via and slug combinations.

Identifier

Description

Via Fraction

No vias

No thermal vias. Plain sheet.

Small vias

10 vias of 200 µm diameter

10 %

Many Small vias

18 vias of 200 µm diameter

21 %

Large vias

4 vias of 320 µm diameter

10 %

Many Large vias

7 vias of 320 µm diameter

21 %

Slug

Copper thermal slug. 0.75 mm x 2.3 mm

100 %

-

Figure 1: Sizes and layouts of the via-structures. From left to right: Large vias, Many Large vias, Small vias, Many Small vias. Unit is µm.

Test module manufacturing Conductors of silver ink on top of a 125 µm PET-film were screen-printed using a 325mesh, 13-µm emulsion screen. The ink was dried at 150 °C for 20 minutes. Thermal vias were laser drilled on bottom side of the PET-film so that they did not pierce through the printed circuit on the other side.

Figure 3: Filled thermal vias of the Large via -option. Small indentation is apparent.

The thermal slug consisted of 100µm thick copper foil. Laser was used to drill a rectangular hole to the PET-sheet with dimensions of 700 µm x 2250 µm. The hole did not pierce the printed circuit on the other side of the substrate. The slug was attached to the bottom of the hole by using thermally and electrically conductive Ag-epoxy (EpoTek H20E). The slug was then screen-printed over with silver ink. Nine LEDs (Osram Oslon LD CQAR) were die-bonded to each module using thermally and electrically conductive epoxy. The curing time was 60 minutes at temperature of 100 °C. A finished no vias -module is presented in Figure 4.

Figure 2: Drilled empty thermal vias of the Large via -option.

Thermal vias were filled by screenprinting using the same silver ink. Filled-through holes were dried for 20 minutes at 150 °C in a ventilated oven. The printed thermal vias are shown in Figure 2. The via structures did not fill up totally in any of the variations. There were small indentations on the via area as shown in Figure 3. This was due to limited amount of ink per area on the printing process. Good thermal connection to the cold plate was ensured by applying thermal paste carefully to the back of the sheet so that the via indentations were also filled. To fill the vias properly with silver ink, another printing run would be required.

Figure 4: Finished module with no vias.

Thermal simulations Steady state thermal simulations were performed with Comsol Multiphysics software. The model was constructed to resemble the measurements closely. The Oslon LED was constructed of 430 µm thick alumina base with 100 µm thick thermal pad and electrical leads. The sapphire chip of the LED was 1400 µm square with 150 µm thickness. The 9 LEDs were placed in 3 x 3 matrix on conductors of silver ink on top of a 125 µm PET-film. The printed conductors were 13 µm thick silver ink with thermal conductivity of 18

W/mK. This module was then placed on a 10 mm thick aluminum block which acted as cold plate. The bottom surface of the aluminum block was set to 25 ºC. The simulation structure is presented in Figure 5.

Table 3: Properties of the materials used in the simulation model. Thickness , µm

Thermal conductivity , W/mK

Silicon

150

20

AlO

150

34

LED pads

Aluminum

100

201

LED dome

Silicone

~1000

0.2

Thermal paste

HTCPpaste

50

2.5

Module substrate

PET-sheet

125

0.2

Object

Chip LED ceramic case

Figure 5: The simulated structure.

One dimensional heat flow was assumed for thin interfaces such as the adhesive connection of the LEDs. These interfaces defined as mathematical surfaces with specific thickness and thermal conductivity. Their parameters are listed on Table 2. A detailed cut-out of the structure identifying the surfaces as dashed lines is shown in Figure 6. Also materials and their thermal conductivities are listed on Table 3. Table 2: Properties of contact resistance surfaces.

Surface

Material

Chip contact surface

Silver paste

LEDpackage contact surface

Adhesive, EpoTek H20E

Thickness, µm

15

Thermal conductivity, W/mK 20

Material

Radiation and convection was taken into account also by mathematical surfaces. All outer surfaces were associated with Raleigh-Jeans radiation law, where the emission coefficient was 0.8. The same surfaces were also associated with convection defined by Fourier's law dQ = h × A × (T dt

T

),

where T is the temperature of the surface, Tamb is the temperature of the ambient air and h is the heat transfer coefficient -5 W/m2 K and A is the area of the surface. Measurement Procedure

10

18

Figure 6: Materials and mathematical surfaces of the simulation model.

Temperature measurements were performed with a thermal transient tester (T3Ster, Mentor Graphics Corp.). T3Ster uses the temperature dependence of the LED forward voltage for thermal characterization of device packages. The individual LEDs of the module were calibrated in temperature controlled chamber by recording the forward voltage of the LEDs as a function of temperature using sensor current of 5 mA. The dependence is linear in the current range. The mean temperature coefficient of nine LEDs was -1.53 mV/ºC with standard deviation of 0.08 mV/ºC. Since the deviation was small, the same coefficient was used for all LEDs instead of measuring their own. After the calibration, the LED sheets were placed on a 25 ºC water cooled cold plate. Thermal paste was used to ensure good thermal contact. The LEDs were then driven using 650 mA current for

100 s until they were thermally stabilized. The electrical power of an individual LED was 2 W. After the heating period, the LEDs were switched to sensor current of 5 mA and the voltage change was recorded for 100 s until the LEDs were cooled back to ambient temperature. The good cooling capacity of the cold plate ensured that the stabilization times were short. Results and Discussion Results of the thermal simulations and measurements are presented in Table 4. Large variations were introduced to measurements due to manual application of the thermal paste. To alleviate the issue, each LED was measured three times and the module was slid around the cold plate surface to get good thermal contact. The result with lowest junction temperature was included in the analysis. The statistical spread of the results by thermal management technique is presented in a box plot in Figure 7. Temperature plot of the cut out of the center LED is presented in Figure 8. The temperatures of the LEDs had a considerable amount of variation among the same type of thermal management structures. This variation is show in boxplot in Figure 7. One cause for this variation is probably related to the adhesive epoxy used to attach the LEDs. A manually operated dispenser was used to administer the adhesive, so the amount can vary from LED to LED. Dispenser with automatic dosage control used in R2R automatic bonder, such as EVO 2200, eliminates adhesive dosage variation. Another cause for the variation could be the thermal paste used to make contact to the coldplate. The paste might not fill the indentations on the vias properly and some air-bubbles could remain. Table 4: Results measurements.

Identifier

No vias

Simulated temperatures are about 20 ºC higher than measured median temperatures in all cases except the thermal slug. This suggests that the thermal model of the LED is correct but the thermal conductivity used for the PET-sheet at the critical area is probably too low. We made verification simulations which showed that modifying the conductivity of the PET-sheet from 0.2 W/mK to 0.3 W/mK produced a 27 ºC decrease in temperature. The simulations and measurement agree on the relative order of the different techniques though, thermal slug being the best and no vias the worst. The thermal slug appears to be most effective heat management structure. The result was expected because the slug has the largest surface fraction of highly conducting material. Also it was made of copper, which is far better heat conductor than silver ink used in the vias. The challenge of this method is manufacturing. The slug requires more processing steps than the vias and can therefore increase the production cost and time. As seen in Table 4 the slug offers only 10 ºC cooler temperatures than many small vias.

Figure 7: Box plot of the measured temperatures grouped by thermal management technique type.

of thermal simulations and

Via fraction

Simulated average Measured average Standard deviation, temperature of the temperature of the [ºC] LEDs, [ºC] LEDs, [ºC]

-

117

97

10

Small vias

10 %

99

75

4

Many Small vias

21 %

89

69

3

Large vias

10 %

104

83

6

Many Large vias

21 %

94

74

8

Slug

100 %

59

59

3

prospects. Physica Status Solidi A, Vol. 208, No. 1, pp. 17–29. [2] Liu, Z., Liu, S., Wang, K., and Luo X. 2009. Status and prospects for phosphor-based white LED packaging. Frontiers Optoelectronics China, Vol. 2, No. 2, pp. 119–140, DOI 10.1007/s12200-0090011-2.

Figure 8: Temperature cut-plot of the LED at the center of the 3x3 matrix.

The best via solution was the design with many small vias. Many large vias are not equally effective, although the structures have the same viafraction. Possible causes are the differences in via location and filling. Notably, the small via option with 10% via fraction is only 1 ºC hotter than the many large vias with 20% via fraction. This implies the same conclusion as in our previous study with rigid substrates [17] we also discovered the same trait that the smaller vias could be filled more easily than the larger vias. Drilling more vias results in longer processing time, so it might be cost effective to choose the large via option, although thermal performance degrades. Another factor to consider is that silver ink is expensive material and larger via fraction would cause more ink to be used. This shows that the optimal amount of vias and their sizes is a compromise between manufacturing speed, cost and adequate thermal performance. Conclusion A series of 6 flexible LED-modules with 9 LEDs each were simulated and fabricated with different thermal management structures. Measurements of the junction temperatures of the LED modules agree with the simulations on the order of the techniques. Thermal slug was deemed to be the best solution with lowest junction temperature. The best via solution was the many small vias which were measured to be 10 ºC hotter. Also it was discovered that small vias outperformed the large vias although they had the same via fraction. Acknowledgements The authors would like to thank co-workers at VTT for their invaluable efforts and support. The research leading to these results has received funding from the European Union Seventh Framework Programme [FP7/2007-2013] under grant agreement n°619556, project LASSIE-FP7. References [1] Haitz, R. and Tsao, J. Y. 2011. Solid-state lighting: ‘The case’ 10 years after and future

[3] Krames, M. R., Shchekin, O. B., Mueller-Mach, R., Mueller, O.G., Zhou, L., Harbers, G., and Craford, M. G. 2007. Status and Future of HighPower Light-Emitting, Diodes for Solid-State Lighting, Journal of display technology, Vol. 3, No. 2, pp. 160 – 175. [4] Crawford, M. H. 2009. LEDs for Solid-State Lighting: Performance Challenges and Recent Advances. IEEE journal of selected topics in quantum electronics, Vol. 15, No. 4, pp. 1028 – 1040. [5] Keränen, K., Mäkinen J.- T., Heikkinen, M., Hiltunen, M.,Koponen, M., Lahti, M.Sunnari, A., Rönkä, K. 2012. Hot Laminated Multilayer Polymer Illumination Structure Based on Embedded LED Chips. IEEE Transactions on Components, Packaging and Manufacturing Technology. Vol. 2.,No. 12: 1965 -1972. [6] Lasance, C. J. M. and Poppe, A. 2014. Thermal Management for LED Applications. Springer Science+Business Media, NewYork. doi:10.1007/978-1-4614-5091-7 [7] Lin, S. C., Huang, R.-F., and Chiu, C. H. 2009. Investigation of Thermally Conductive Ceramic Substrates for High-Power LED Application. In: Microsystems, Packaging, Assembly and Circuits Technology Conference (IMPACT), Taipei, Taiwan, 21 - 23 Oct. 2009, pp. 589-592. [8] Yang, L., Jang, S., Hwang, W., and Shin, M. 2007. Thermal analysis of high power GaN-based LEDs with ceramic package. Thermochimica Acta, Vol. 455, No. 1-2, pp. 95–99. [9] Sim, J.-K., Ashoka, K., Ra, Y.-H., Im, H.-C., Baek, B.-J., and Lee C.-R. 2012. Characteristic enhancement of white LED lamp using low temperature co-fired ceramic-chip on board package. Current Applied Physics, Vol. 12, No. 2, pp. 494-498. [10] Shin, H. W., Lee, H. S., and Jung, S. B. 2011. Analysis on Thermal Resistance of LED Module with Various Thermal Vias. In: 18th IEEE International Symposium on the Physical and Failure Analysis of Integrated Circuits (IPFA), Incheon, Korea, 4 - 7 July 2011.

[11] Kim, Y.-W., Kim, J.-P., Kim, J.-B., Kim, M.S., Sim, J.-M., Song, S.-B., and Hwang, N. 2009. Thermal analysis of a package substrate with filling viaholes for COB LED manufacturing. Journal of the Korean Physical Society, Vol. 54, No. 5, pp. 1873–1878. [12] E. Juntunen, A. Sitomaniemi, O. Tapaninen, R. Persons, M. Challingsworth and V. Heikkinen. Thermal Performance Comparison of Thick-Film Insulated Aluminum Substrates With Metal Core PCBs for High-Power LED Modules,. IEEE Transactions on Components, Packaging and Manufacturing Technology, Dec. 2012, Vol. 2, No. 12: 1957-1964. [13] E. Juntunen, O. Tapaninen, A. Sitomaniemi, M. Jämsä, M. Karppinen and P. Karioja. CopperCore MCPCB with Thermal Vias for High-Power COB LED Modules. IEEE Transactions on Power Electronics, March 2014, Vol. 29, No. 3: 14101417. [14] Heo, Y. J., Kim, H. T., Nahm, S., Kim, J., Yoon, Y. J., and Kim, J. 2012. Ceramic-metal package for high power LED lighting. Frontiers of Optoelectronics, Vol. 5, No. 2, pp. 133–137. [15] Heo, Y. J., Kim, H. T., Kim, K. J., Nahm, S., Yoon, Y. J., and Kim, J. 2013. Enhanced heat transfer by room temperature deposition of AlN film on aluminum for a light emitting diode package. Applied Thermal Engineering, Vol. 50, No, 1. pp. 799-804. [16] Lee, M.-H., Lee, T. J., Lee, H. J., and Kim, Y.J. 2010. Design and fabrication of metal PCB based on the patterned anodizing for improving thermal dissipation of LED lighting. In: Microsystems Packaging Assembly and Circuits Technology Conference (IMPACT), Taipei, Taiwan, 20 – 22. Oct. 2010, pp. 1–4. [17] Heikkinen V., Juntunen E., Kautio K., Kemppainen A., Korhonen P., Ollila J. and Sitomaniemi A. 2007. High-brightness RGB LED Modules Based on Alumina Substrate. In: Proceedings of 16th European Microelectronics and Packaging Conference and Exhibition 2007 (EMPC 2007), pp. 622-688.