Budapest University of Technology and Economics Department of Electron Devices Diagnostics of LED-based streetlighting luminaires by means of thermal transient method Gábor Marosy, Zoltán Kovács, András Poppe The 16 th THERMINIC Workshop, 6-8 October 2010, Barcelona, Spain
Introduction Similar trends in SSL like Moore s Law in microelectronics Continuous decrease of price of light [$/lm] More application fields, streetlighting is among the early birds Low voltage levels controllability 2
Introduction LED based light sources still more expensive than traditional ones, but they provide some benefits Longer life-times expected if junction temperatures are kept low 50% liminous flux 80% limous flux Controllability (in case of DC driving) DC power distribution networks for indoor applications? Easy dimming (e.g. by PWM) Possibility of integration with communications networks Possibility of built-in intelligence, including diagnostic functions Streetligting: present safety standards allow low voltage only inside an LED luminaire DC driven LEDs with driver/control circuitry 3
Introduction: vision of intelligent streetlighting Control, diagnostics, sensing and communications functions can be built in. 4
What is the real issue? Keep T j low! In case of LEDs, heat-loss is mainly by conduction Power LED heat loss: ~85-60%, mostly by conduction - Heat-conduction is the initial part of the dissipation The luminaire is used as a heatsink, like in case of retrofit LED lamps There are different thermal interfaces present in the heat-flow path light: ~15..40% active layer to submount classical die attach (TIM1) solder/glue to MCPCB classical TIM to heat-sink (TIM2) Thermal transient testing proved to be a good testing tool for TIMs 5
The luminaire acts as a heat-sink Quality of the TIM between the LEDs and the luminaire is critical LED package models identified by T3Ster+TERALED, simulation by FloTHERM CAD model by courtesy of HungaroLux Ltd. (Budapest, Hungary) 6
Cth [Ws/K] Pulse thermal resistance [K/W] Data from thermal transient tests of LEDs Power h(t) Junction temperature a(t) time time Package dynamic model 100 Structure function T3Ster Master: cumulative structure function(s) CREE_MCE_AL_2_25_MY_F1_T25_I0350 - Ch. 0 3.5 3 Pulse thermal resistance diagram T3Ster Master: Pulse Rth Diagram CREE_MCE_AL_2_25_MY_F1_T25_I0350-0.50 CREE_MCE_AL_2_25_MY_F1_T25_I0350-0.25 CREE_MCE_AL_2_25_MY_F1_T25_I0350-0.10 CREE_MCE_AL_2_25_MY_F1_T25_I0350-0.05 10 2.5 2 1 1.5 0.1 1 0.01 0.5 0.001 1e-5 1e-4 0.001 0.01 0.1 1 10 100 0.5 1 1.5 2 2.5 3 3.5 4 Rth [K/W] The h(t) step-wise change in heating is applied at the junction (abrupt switching) The a(t) temperature response at the junction is being measured (unit-step response function) while linearity is assumed All available information is extracted from a(t) using sophisticated mathematical procedures structure functions ideal to characterize TIM compact dynamic thermal models pulsed thermal resistance / complex locus (frequency domain representation) 7 Time [s]
The goal is: assess quality of TIMs in the field How feasible is it? Case studies of different assemblies for streetligting luminaires analyzed by thermal transient measurements Temperature dependent change of TIM observed (Poppe et al, SEMI-THERM 10) Long term stability studies have been performed to see what one may expect Changes of classical TIM observed frequently (Poppe et al, SPIE SSL 10) In case of some LEDs DA degradation and degradation of other thermal interfaces (such as glue/solder between the LED package and MCPCB was observed) Which change can be measured in a luminaire? At PWM-based dimming Short time windows only, e.g. 10..100ms Would allow DA testing, but requires time and V F measurement resolution of a lab equipment Daytime, planned on/off sequences No testing time limitations With smaller time resolution classical TIM issues can be detected Timing requirements: for time constant measurement between 0.1 and 10 is necessary (Székely, THERMINIC 08) realistic with cheap circuitry Problems: no option for K-factor calibration / computation 8
Case study for the KÖZLED project 3 different assemblies with high-end 10W white LEDs FR4 PCB, TIM between the heat-slug and the Cu block FR4 PCB, heat-slug soldered to the Cu block MCPCB-s made of Al and Cu, heat-slug soldered CAD images by courtesy of OptimalOptik Ltd. 9
Results for 10W white LEDs Measured at 700 ma and 85 o C Thermal impedance of 3 samples, power corrected with P opt FSF52 AL TG2500 R thjc real 2 K/W 10
Results for 10W white LEDs Measured at 700 ma and 85 o C Structure functions of 3 samples, power corrected with P opt TG2500 AL FSF52 R thjc real 2 K/W R thjc is identified in a way similar to the transient double interface method, standardized by the JEDEC JC15 committee 11
Results for 10W white LEDs Measured at 700 ma & between 15 o C and 85 o C Structure functions of sample AL-2, power corrected with P opt LED package: no variation Al MCPCB &TIMs: temperature dependence As temperature changed, TIM quality changed 12
LM80 Tests @ Univ. of Pannonia, Hungary LM80 test chamber with all the LEDs assembled All measurements are done in-situ to eliminate any R th change which is NOT due to ageing In-situ thermal transient measurement In-situ light output measurement 13
Results from LM80 test of different LEDs In cooperation with University of Pannonia, Veszprém (Hungary), within the KöZLED project 8 different kinds of LEDs from 4 vendors, so far 4000h burning time 110% 105% 100% 95% 90% 85% 4.Osram Vendor LUWV5AM O 350mA Relative luminous flux 0 200 400 600 800 1000 1200 Time [h] 41 42 43 44 45 46 aver. Ligh output drop likely due to increased R th caused by TIM degradation, not by LED degradation No change inside the LED package Structure functions taken at 0h, 500h, 1000h 14
T J [ C] C th [Ws/K] Results from LM80 tests: with accurate dv F /dt Results when accurate temperature sensitivity of the forward voltage is considered in thermal transient testing: T3Ster Master: Smoothed response T3Ster Master: cumulative structure function(s) 15.1 0.9 16 14 12 10 8 6 4 2 Vendor O, sample #44, 0h Vendor O, sample #44, 500h Vendor O, sample #44, 2000h Vendor O, sample #44, 3000h With real values of dv F /dt 0 1e-6 1e-5 1e-4 0.001 0.01 0.1 1 10 0.001 Problems: dv F /dt sensitivity also changes in time We have no means to calibrate t [s] 100 10 1 0.1 0.01 1e-4 1e-5 Vendor O, sample #44, 0h Vendor O, sample #44, 500h Vendor O, sample #44, 2000h Vendor O, sample #44, 3000h With real values of dv F /dt 0 2 4 6 8 10 12 LED package: small variation after 500h R th [K/W] TIM: +0.06 R thja 15
Sensitivity [mv/k] Results from LM80 tests: changing dv F /dt Results for the changes of the temperature sensitivity of the forward voltage for samples of vendor O: Vendor O, samples 41-46: dv F /dt Sensitivity Change -2.4-2.3-2.2-2.1-2 -1.9-1.8-1.7-1.6 O_41 O_42 O_43 O_44 O_45 O_46 0 500 1000 1500 2000 2500 3000 t [hour] E.g. for sample #44 4.3% change in the temperature sensitivity was observed This change compensates the effect of TIM ageing 16
Cth [Ws/K] T J [ C] Results from LM80 tests: without real dv F /dt Considering a constant sensitivity value (-2mV/K) the aging of TIM is screened: 16 14 T3Ster Master: Smoothed response Vendor O, sample #44, 0h Vendor O, sample #44, 500h Vendor O, sample #44, 2000h Vendor O, sample #44, 3000h 1000 100 10 T3Ster Master: cumulative structure function(s) GoldenD_44_0h - Ch. 0 44_54_64_500h - Ch. 0 14-24-34-44_2000h - Ch. 3 14_24_34_44_3000h - Ch. 3 12 dv F /dt = -2mV/K 1 10 8 6 4 2 0.1 0.01 0.001 1e-4 0 1e-6 1e-5 1e-4 0.001 0.01 0.1 1 10 t [s] We may expect to see drastic changes only! 1e-5 0 2 4 6 8 10 12 14 16 Rth [K/W] With an assumed constant sensitivity even structure functions appear to be identical, though in a quick diagnostic system in the field probably there will be no possibility to implement structure function calculations. 17
C th [Ws/K] Sensitivity [mv/k] Drastic ageing observed at another vendor Results with real sensitivities: -1.7-1.6-1.5-1.4-1.3-1.2-1.1-1 T3Ster Master: cumulative structure function(s) 5.6 4.7 0.6 Vendor HK, samples 61-65: dv F /dt Sensitivity Change -0.9 0 500 1000 1500 2000 2500 3000 HK_61 HK_62 HK_63 HK_64 HK_65 t [hour] 100 10 1 0.1 0.01 0.001 1e-4 Vendor HK, sample #61, 0h Vendor HK, sample #61, 500h Vendor HK, sample #61, 2000h Vendor HK, sample #61, 3000h With real values of dv F /dt 1e-5 0 2 4 6 8 R th [K/W] 18
Cth [Ws/K] T J [ C] Drastic ageing observed at another vendor Results considering an assumed constant sensitivity only: T3Ster Master: Smoothed response 6 5 4 Vendor HK, sample #61, 0h Vendor HK, sample #61, 500h Vendor HK, sample #61, 2000h Vendor HK, sample #61, 3000h dv F /dt = -2mV/K Delamination of the LED/MCPCB interface as well as TIM ageing can be observed even from the measured raw forward voltage transients. 3 2 1 1000 100 10 T3Ster Master: cumulative structure function(s) HK_61_0h - Ch. 0 41_51_61_500h - Ch. 2 51-61-91-101_2000h - Ch. 1 51_61_71_3000h - Ch. 1 0 1e-6 1e-5 1e-4 0.001 0.01 0.1 1 10 t [s] 1 30 s measurement time is sufficient for this test, sufficient time resolution can be 0.1ms or even 1ms Real life conditions are even harsher: vibration humidity temperature cycling 0.1 0.01 0.001 1e-4 0 1 2 3 4 5 6 Rth [K/W] 19
Conclusions We made an analysis of possible structural integrity issues in the junction to ambient heat-flow paths of LEDs to be used in streetlighting luminaires. Preliminary results of long term stability studies of LEDs suggest that in harsh environment we should expect degradation of the different thermal interfaces. If the degradation exceeds a given limit (resulting thermal resistance change bigger then variation of temperature sensitivity of the forward woltage), with built-in thermal transient measurement based diagnosis early warnings can be provided and maintenance of luminaires can be scheduled to avoid fatal breakdown. Requirements for such testing circuitry are not too high, ~1ms time resolution seems to be sufficient ti detect LED/MCPCB or TIM2 delamination/ageing No sophisticated data processing is required for such tests, it seems, raw forward voltage transients are sufficient 20
Acknowledgements This work was partially supported by the KÖZLED TECH_08-A4/2-2008-0168 project of the Hungarian National Technology Research and Development Office. Further support of the Hungarian Government through the TÁMOP-4.2.1/B-09/1/KMR-2010-0002 project at the Budapest University of Technology and Economics is also acknowledged. We are grateful for the help of KÖZLED partners: University of Pannonia (Veszprém, Hungary) for sharing their LM80 test facility and test results with us. OptimalOptic Ltd. for LED samples with different thermal management solutions and their CAD images Hungarolux Ltd. for CAD images of their luminaires Help of Mentor Graphics MicReD Division is also acknowledged for some CFD simulations as well as helping us perform some of our measurements 21