- Email: [email protected]
Evaluation of thermal transient characterization methodologies for high-power LED applications
Microelectronics Journal 44 (2013) 1005–1010
Contents lists available at ScienceDirect
Microelectronics Journal journal homepage: www.elsevier.com/locate/mejo
Evaluation of thermal transient characterization methodologies for high-power LED applications Stefan Müller a,n, Thomas Zahner a, Frank Singer a, Gabriele Schrag b, Gerhard Wachutka b a b
OSRAM Opto Semiconductors GmbH, Regensburg, Germany Institute for Physics of Electrotechnology, Munich University of Technology, Germany
art ic l e i nf o
a b s t r a c t
Article history: Received 30 November 2011 Received in revised form 8 July 2013 Accepted 21 August 2013 Available online 26 September 2013
In the past, thermal characterization methodologies for LED packages have mainly been derived from already existing solutions of the microelectronics industry. Within this paper, several issues regarding the determination of the junction-to-case thermal resistance RthJC for LED packages are addressed. The new JESD51-14 standard is taken into consideration and especially the so called “point of separation” of the underlying dual-interface method is investigated. Experiments and finite element simulations were carried out in order to investigate the environmental influences on this crucial point. The investigations reveal that the point of separation changes depending on the thermal boundary condition at the case of the LED module, viz the quality of the package attach. & 2013 Elsevier Ltd. All rights reserved.
Keywords: Junction-to-case thermal resistance LED package JESD51-14 standard Point of separation
ΔT f ¼ Z ef ð1ηÞ1 th P el P opt
1. Introduction
Z real th ¼
For LED packages, the measurement of the thermal impedance and resistance follows a widely accepted determination of the thermal transient which is specified in the JESD51 standard [1]. Furthermore, the transient dual-interface (TDI) method, proposed by Schweitzer [2], has currently been standardized [3]. This method is used in order to extract the junction-to-case thermal resistance RthJC of a single device from two thermal impedance measurements. In the first of these two measurements, a thermal interface material (TIM) is placed between the device and the heat sink, and in the second measurement this interface material is removed. The new JESD51-14 standard [3] suggests to use the point at which the thermal impedance graphs of both measuring setups separate as an approximation for RthJC.
Eq. (1) represents the original definition of the thermal impedance as specified by the JESD51 standard. Here, Zth is termed as effective thermal impedance because in reality, LED modules convert a significant amount of electrical power into optical power as well. Thus, the value of Zth has to be corrected accordingly, which can be expressed in terms of the LED efficiency η (see Eq. (2)). Hence, using Eq. (1) instead of Eq. (2) would result in a reduced amplitude of the thermal impedance and therefore in a reduced value for RthJC. Unfortunately, this is not standardized. Moreover, as it had already been mentioned by Schweitzer et al. [4], one significant problem with respect to thermal characterization is the notion of the thermal resistance itself. The thermal resistance is defined as the change in temperature between two equipotential surfaces of the same cross-section, divided by the amount of heat which flows through those surfaces per unit time. However, in reality the situation is quite different (see Fig. 1). Both junction and case level, show non-homogenous temperature profiles and first investigations taking this situation into account had already been performed by Schweitzer [5]. For this study, various thermal resistances were defined according to the temperature profiles introduced in Fig. 1.
2. Issues in determining RthJC for LED applications Since LEDs convert a significant amount of electrical power into optical power, ambiguities already arise when the thermal impedance is to be defined. f Z ef ¼ th
n
ΔT P el
Corresponding author. E-mail address: [email protected] (S. Müller).
0026-2692/$ - see front matter & 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.mejo.2013.08.014
ð1Þ
RMM RAA
T J;MAX T C;MAX P th
T J;AVG T C;AVG P th
ð2Þ
ð3Þ ð4Þ
1006
S. Müller et al. / Microelectronics Journal 44 (2013) 1005–1010
2.5
TJ,MAX
TJ,AVG
Temperature
Zth [K/W]
2.0 1.5 1.0 Hot Spot Average
0.5 TC,MAX
0.0 1E-6
TC,AVG TAMB
1E-2 Time [s]
1E+0
Fig. 2. Simulated thermal impedance graphs extracted at junction hot-spot (solid line) and averaged over all five LED chips (dashed line).
Lateral Dimension Junction
1E-4
Case
2.5 LED
2.0
Aluminum PCB Fig. 1. Illustration: Junction and case temperature profiles for a one-chip LED package. Characteristic temperatures are indicated by crosses and dashed lines. The inhomogeneity of the profiles is clearly observable.
Zth [K/W]
Ceramic
Simulation Experiment
1.5 1.0 0.5
T J;MAX T C;AVG RMA P th
ð5Þ
T J;AVG T C;MAX P th
ð6Þ
RAAMB
1E+0
Fig. 3. Comparison of thermal transient data obtained from experiment and simulation.
T J;MAX T AMB P th
ð7Þ
T J;AVG T AMB P th
ð8Þ
RMAMB
1E-4 1E-2 Time [s]
These thermal resistances are investigated in more detail in Section 4 in order to see if they depend on the thermal boundary conditions (die attach, package, cooling) and, thus, on the changing temperature profile inside the case. Moreover, this directly leads to another problem in RthJC determination. For multi-chip LED packages like headlamp modules, the LEDs are connected in series in order to improve the homogeneity of the light output. According to the JESD51 standard, the temperature is measured with respect to the change in forward voltage of the device. When the thermal performance of headlamp modules is determined in this way, it is important to be aware that an average temperature is measured and not the hotspot temperature which is, however, the crucial value for device failure. The notion of an average temperature originates from the fact that the forward voltages of the individual LEDs do not change simultaneously but are instead affected by the temperature profile along the series connection. The thermal impedance graphs of Fig. 2 had been derived from finite element analysis. The thermal response of the headlamp modules to an overall thermal power of 1 W was calculated and, both, the normalized hot-spot and the normalized averaged thermal transient are depicted in the graph. For stationary operation, the difference can be quite significant, e.g., for our simulation it was approximately 10%. Furthermore, when the same simulation model as shown in Fig. 2 is used and compared to results obtained from transient
[a.u.]
RAM
0.0 1E-6
1E+0
1E+3 1E+6 # Data Points
1E+9
Fig. 4. Illustration: Parasitic electrical transient (green line) superimposed on the actual thermal signal (red line) results in the real measurement signal (blue line) and artifacts in the Rth extraction. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
measurements, the thermal impedance graphs differ as shown below (Fig. 3). The discrepancy between experiment and simulation is either caused by wrong assumptions for the simulation model (e.g., a heat capacity which is assumed too low) or it occurs due to measurement parasitics as indicated in Fig. 4. Simulations revealed that the heat capacity of the pn-junction would have to be increased tenfold in order to compensate for the shift. Since this is far from physical reality, the mentioned discrepancy is expected to be caused by parasitic capacitances in the measurement setup (Fig. 4). Hence, the difference between
S. Müller et al. / Microelectronics Journal 44 (2013) 1005–1010
experiment and simulation rather represents an offset than a logarithmic time shift. In the meanwhile, this finding has also been incorporated in the new JESD51-14 standard via a square root extrapolation of the measurement data at small times.
1007
DUT
DUT
3. Experimental investigations In order to investigate the influence of the environment on the point of separation itself, a 5-chip headlamp module was characterized first by using the transient dual-interface methodology. Additionally, in a third measurement, the module was attached to the heat sink using highly thermal conductive glue (see Fig. 5). As it can be seen from Figs. 5 and 6, the point at which the graphs separate changes when the module attach is improved or deteriorated. This artifact was further investigated by building up a new heat sink setup in order to create a more ideal device cooling. This cooling solution was chosen due to its easy manufacturability and its large cooling capacity [6]. The goal of this kind of heat sink solution was to provide an optimized heat extraction from the device. For a perfect heat sink, no point of separation would be necessary to determine RthJC and the stationary value of the thermal impedance graph would correspond to the real junction-to-case thermal resistance of the device. However, this is not possible since the performance of current high-end cooling solutions is limited to heat transfer coefficients up to approximately 105 W/m2 K [7]. Therefore, the difference
Inlet
Outlet
Fig. 7. Schematic cross-section of the new jet impingement heat sink setup.
4
Zth [K/W]
3
Fig. 8. First prototype of the new heat sink setup. Shown are two 5-chip headlamp modules mounted right above the two jets (not visible).
w/o TIM w/ TIM Glued
3
2
0 1E-6
1E-4
1E-2 Time [s]
1E+0
1E+2
Zth [K/W]
1 2
w/o TIM w/ TIM Jet
1
Fig. 5. Thermal impedance graphs for a headlamp module attached to the heat sink without thermal grease (w/o TIM), with thermal grease (with TIM) and using highly thermal conductive glue.
0 1E-6 2.5
Zth [K/W]
1E-2 Time [s]
1E+0
Fig. 9. Thermal impedance graphs for a headlamp module attached to the heat sink without thermal grease (w/o TIM) and with thermal grease (with TIM). The dotted line corresponds to the thermal transient impedance derived by using the new jet impingement cooling setup.
2.3 2.1 1.9 w/o TIM w/ TIM Glued
1.7 1.5 2E-2
1E-4
2E-1 Time [s]
2E+0
Fig. 6. Magnified points of separation. Shift of the separation point is observable for the module which is glued to the heat sink.
between a point of separation derived from measurements using jet impingement cooling and the ideal RthJC from simulation was of further interest. The results obtained by using the new heat sink setup (Figs. 7 and 8) are shown below (Figs. 9 and 10). However, a perfect module attach can never be realized by an experimental setup. But it is possible to simulate this situation by e.g., Finite Element Analysis. By using this approach it is possible to analyze the correspondence of the point of separation and the real junction-to-case thermal resistance (Fig. 11).
1008
S. Müller et al. / Microelectronics Journal 44 (2013) 1005–1010
4. Results From Simulation
2.5
Zth [K/W]
2.3
w/o TIM w/ TIM Jet
2.1 1.9 1.7 1.5 2E-2
2E-1 Time [s]
2E+0
Fig. 10. Magnified point of separation. Shift in point of separation again present if stationary value of jet impingement cooling is not taken as RthJC.
In order to investigate and quantify the effects of the environment on the point of separation in more detail, simulation models of two geometries had been constructed. The ideal model was created in a way that it represented the real headlamp package as detailed as possible and an ambient temperature boundary condition was applied to the case of the package (Fig. 11, left image). In the second model (Fig. 11, right image) a copper plate and a heat sink were added in addition since this represents the real measurement setup as it is used at OSRAM OS. Fig. 12 shows a more detailed description of the model. For all simulation models, die attach (Solder), adhesive, thermal interface material (TIM) and the thermal grease layers were modeled as lumped thermal resistive layers with individual thermal conductivities in [W/m K]. LED chips were assumed to
Fig. 11. Simulation models created for spreading point analysis. On the left, the ideal model is shown, on the right, the ideal model is attached to a heat sink as it is the case for a real measurement setup.
5 Dies
Solder
Adhesive
MCPCB
Cu Plate
Thermal Grease
Ceramic submount
TIM
Heat Sink
Fig. 12. Domains of the extended simulation model incorporating the heat sink setup. Thin thermal resistive layers had been applied to Solder, Adhesive, TIM and Thermal Grease.
S. Müller et al. / Microelectronics Journal 44 (2013) 1005–1010
nðk∇T Þ ¼ hðT amb T Þ
ð9Þ
These investigations were extended by a larger batch of stationary simulations. Referring back to the thermal resistance definitions in chapter II, the following graph could be deduced. Here, RMM, RAA, RMA and RAM represent the thermal resistance of the device itself, i.e., in a way RthJC, whereas RMAMB and RAAMB rather correspond to RthJA (the junction-to-ambient thermal resistance) since they incorporate the ambient temperature in their definition (see Eqs. (7) and (8)). Fig. 15 shows that the device resistances stay rather constant in the range of 100–1000 W/m2 K whereas more significant variations occur for larger heat transfer coefficients. This corresponds to the observed heat flow paths (see Fig. 14) which change with varying heat transfer coefficients. All defined thermal resistances saturate for very large heat transfer
Zth [K/W]
3
2
λTIM = 0.025 λTIM = 0.1 λTIM = 0.5 λTIM = 3 λTIM = 50 Ideal
1 0 1E-6
1E-4
1E-2 1E+0 Time [s]
1E+2
Fig. 13. Comparison of the thermal impedance derived from the ideal model and the extended model. If the module is attached in a very poor manner, the point of separation changes.
Fig. 14. Stationary temperature fields and heat flow paths for different qualities of the device attach (in terms of h, from top to bottom: 102, 103, 104, 105 and 1015 W/m2K).
8 7 Rth [K/W]
be surface heat sources so that a homogenous heat generation was assumed. This represents a simplification of the reality since the areas of highest current density and therefore highest thermal power generation are near the bond contacts. However, this was neglected for simulation purposes. All the other external boundaries were chosen to be adiabatic, i.e., no heat exchange was allowed. This is justified if no heat transport via radiation or convection is assumed. The only boundary which differed from this assumption was the one on the bottom of the MCPCB and the heat sink respectively at which room temperature was applied as fixed temperature boundary condition. The thermal conductivity of the TIM was then varied and by that, the points of separation were compared to the thermal impedance derived from the ideal model. The simulated curves in Fig. 13 show that the point of separation is shifted when the device attach to the heat sink is improved or deteriorated. This confirms the experimental findings of Section 3, and indicates that the point of separation has no real physical meaning but is a rather mathematical procedure for extracting RthJC of an LED module. Especially the derivation of one single junction-tocase thermal resistance independent of the boundary conditions (besides others, the quality of the device attach) seems doubtful, since the heat flow path inside of the device changes depending on the applied thermal boundary conditions at the case. In order to illustrate this dependency in more detail, a twodimensional simulation model as depicted in Fig. 14 was derived, and the evolving temperature fields inside the module were investigated. The quality of the device attach was modeled in terms of a heat outflow boundary condition at the bottom of the MCPCB using the heat transfer coefficient h (all other boundary were chosen according to the 3D models).
1009
6 5 4 3 2 1E+0
RMAMB RMA RMM
1E+2 h
RAAMB RAA RAM
1E+4
1E+6
[W/m2K]
Fig. 15. Defined thermal resistances vs. heat transfer coefficient. In certain regions, the defined junction-to-case thermal resistances change significantly depending on the module attach.
coefficients, i.e., when the module is attached ideally to the ambient so that the heat can propagate straightforward through the device and almost no heat spreading occurs. The special behavior of RMM and RAM, i.e., that the thermal resistance increases with increasing heat transfer coefficient, had also been shown by Schweitzer [5] in a similar form. Both resistances merge with RMA respectively RAA for large h since the average temperature at the case gets closer to the maximum temperature. Moreover, the thermal resistance of the environment becomes negligible when large heat transfer coefficients can be provided, which is shown by the merging of RMAMB and RMA respectively RAAMB and RAA.. The fact that all of the device resistances again decrease for very small heat transfer coefficients might be understood in terms of thermodynamics: It is always necessary to keep in mind that the thermal resistance of the complete system needs to be minimized at the end. Hence, e.g., in the interval [100;1000] the heat flows such that a lumped thermal resistance RMA or RAA is at a maximum whereas the overall system resistance is minimized. For heat transfer coefficients smaller than 100 W/m2 K, the heat needs to flow through a larger cross-sectional area at the bottom of the LED package in order to keep the system's thermal resistance as low as possible. In that instant, the defined lumped thermal resistances of the LED package might well be lower than for the case when heat transfer coefficients larger than 100 W/m2 K are present. At the end it should also be mentioned that these investigations need to be extended to include the electrical domain as well in order to give an even deeper insight into the accordance of experiment and simulation. It can well be understood that
1010
S. Müller et al. / Microelectronics Journal 44 (2013) 1005–1010
variously changing forward voltages of series connected LEDs inside a headlamp module cause a change in electrical and therefore thermal power consumption. This coupling can be taken into account by a combined electro-thermal simulation of high-power LED headlamp modules.
to a specific application, respectively heat sink solution. Ongoing work will focus on systematizing this approach and on investigating its feasibility for practical use.
References 5. Conclusions For both, experiments and simulations, the dual-interface method leads to varying results for RthJC depending on the thermal boundary conditions at the case. A more detailed analysis regarding the influence of the environment (summarized in the heat transfer coefficient h) on the thermal resistance of the module itself showed, that it is generally not possible to specify one single junction-to-case thermal resistance. Even though it is an improvement to the previous state-of-the-art to standardize the measurement and extraction procedure of RthJC (as it is stated in the new JESD51-14 standard), the intricate interplay of the module with the complete system assembly is not yet taken into account. A suggestion to improve this situation is to provide Rth vs. h graphs for a specific LED module so that the customer is able to estimate the thermal performance of the LED package with respect
[1] Electronic Industries Association, JEDEC 51 Standard: Methodology for the Thermal Measurement of Component Packages, 1995. [2] D. Schweitzer, Transient dual interface measurement of the Rth-JC of power packages, in: Proceedings of the Therminic Conference, Rome, 2008, pp. 14–19. [3] JEDEC Solid State Technology Association, Transient dual interface test method for the measurement of the thermal resistance junction to case of semiconductor devices with heat flow trough a single path, JEDEC Standard, 2010. [4] D. Schweitzer et al., How to evaluate transient dual interface measurements of the Rth-JC of power semiconductor packages, in: Proceedings of the 25th IEEE SEMI-THERM Symposium, 2009, pp. 172–179. [5] D. Schweitzer, The junction-to-case thermal resistance: a boundary condition dependent thermal metric, in: Proceedings of the 26th IEEE SEMI-THERM Symposium, 2010, pp. 151–156. [6] System Design and Analysis Inc., Spray cooling technologies market investigation, Indianapolis, 2000 〈http://www.navsea.navy.mil/nswc/crane/sd18/Public%20Documents/Reports/LibrarySprayCool.pdf〉. [7] C.J.M. Lasance, R.E. Simons, Advances in high-performance cooling for electronics, Electronics Cooling, November 2005. 〈http://www.electronics-cooling. com/2005/11/advances-in-high-performance-cooling-for-electronics/〉.
Contents lists available at ScienceDirect
Microelectronics Journal journal homepage: www.elsevier.com/locate/mejo
Evaluation of thermal transient characterization methodologies for high-power LED applications Stefan Müller a,n, Thomas Zahner a, Frank Singer a, Gabriele Schrag b, Gerhard Wachutka b a b
OSRAM Opto Semiconductors GmbH, Regensburg, Germany Institute for Physics of Electrotechnology, Munich University of Technology, Germany
art ic l e i nf o
a b s t r a c t
Article history: Received 30 November 2011 Received in revised form 8 July 2013 Accepted 21 August 2013 Available online 26 September 2013
In the past, thermal characterization methodologies for LED packages have mainly been derived from already existing solutions of the microelectronics industry. Within this paper, several issues regarding the determination of the junction-to-case thermal resistance RthJC for LED packages are addressed. The new JESD51-14 standard is taken into consideration and especially the so called “point of separation” of the underlying dual-interface method is investigated. Experiments and finite element simulations were carried out in order to investigate the environmental influences on this crucial point. The investigations reveal that the point of separation changes depending on the thermal boundary condition at the case of the LED module, viz the quality of the package attach. & 2013 Elsevier Ltd. All rights reserved.
Keywords: Junction-to-case thermal resistance LED package JESD51-14 standard Point of separation
ΔT f ¼ Z ef ð1ηÞ1 th P el P opt
1. Introduction
Z real th ¼
For LED packages, the measurement of the thermal impedance and resistance follows a widely accepted determination of the thermal transient which is specified in the JESD51 standard [1]. Furthermore, the transient dual-interface (TDI) method, proposed by Schweitzer [2], has currently been standardized [3]. This method is used in order to extract the junction-to-case thermal resistance RthJC of a single device from two thermal impedance measurements. In the first of these two measurements, a thermal interface material (TIM) is placed between the device and the heat sink, and in the second measurement this interface material is removed. The new JESD51-14 standard [3] suggests to use the point at which the thermal impedance graphs of both measuring setups separate as an approximation for RthJC.
Eq. (1) represents the original definition of the thermal impedance as specified by the JESD51 standard. Here, Zth is termed as effective thermal impedance because in reality, LED modules convert a significant amount of electrical power into optical power as well. Thus, the value of Zth has to be corrected accordingly, which can be expressed in terms of the LED efficiency η (see Eq. (2)). Hence, using Eq. (1) instead of Eq. (2) would result in a reduced amplitude of the thermal impedance and therefore in a reduced value for RthJC. Unfortunately, this is not standardized. Moreover, as it had already been mentioned by Schweitzer et al. [4], one significant problem with respect to thermal characterization is the notion of the thermal resistance itself. The thermal resistance is defined as the change in temperature between two equipotential surfaces of the same cross-section, divided by the amount of heat which flows through those surfaces per unit time. However, in reality the situation is quite different (see Fig. 1). Both junction and case level, show non-homogenous temperature profiles and first investigations taking this situation into account had already been performed by Schweitzer [5]. For this study, various thermal resistances were defined according to the temperature profiles introduced in Fig. 1.
2. Issues in determining RthJC for LED applications Since LEDs convert a significant amount of electrical power into optical power, ambiguities already arise when the thermal impedance is to be defined. f Z ef ¼ th
n
ΔT P el
Corresponding author. E-mail address: [email protected] (S. Müller).
0026-2692/$ - see front matter & 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.mejo.2013.08.014
ð1Þ
RMM RAA
T J;MAX T C;MAX P th
T J;AVG T C;AVG P th
ð2Þ
ð3Þ ð4Þ
1006
S. Müller et al. / Microelectronics Journal 44 (2013) 1005–1010
2.5
TJ,MAX
TJ,AVG
Temperature
Zth [K/W]
2.0 1.5 1.0 Hot Spot Average
0.5 TC,MAX
0.0 1E-6
TC,AVG TAMB
1E-2 Time [s]
1E+0
Fig. 2. Simulated thermal impedance graphs extracted at junction hot-spot (solid line) and averaged over all five LED chips (dashed line).
Lateral Dimension Junction
1E-4
Case
2.5 LED
2.0
Aluminum PCB Fig. 1. Illustration: Junction and case temperature profiles for a one-chip LED package. Characteristic temperatures are indicated by crosses and dashed lines. The inhomogeneity of the profiles is clearly observable.
Zth [K/W]
Ceramic
Simulation Experiment
1.5 1.0 0.5
T J;MAX T C;AVG RMA P th
ð5Þ
T J;AVG T C;MAX P th
ð6Þ
RAAMB
1E+0
Fig. 3. Comparison of thermal transient data obtained from experiment and simulation.
T J;MAX T AMB P th
ð7Þ
T J;AVG T AMB P th
ð8Þ
RMAMB
1E-4 1E-2 Time [s]
These thermal resistances are investigated in more detail in Section 4 in order to see if they depend on the thermal boundary conditions (die attach, package, cooling) and, thus, on the changing temperature profile inside the case. Moreover, this directly leads to another problem in RthJC determination. For multi-chip LED packages like headlamp modules, the LEDs are connected in series in order to improve the homogeneity of the light output. According to the JESD51 standard, the temperature is measured with respect to the change in forward voltage of the device. When the thermal performance of headlamp modules is determined in this way, it is important to be aware that an average temperature is measured and not the hotspot temperature which is, however, the crucial value for device failure. The notion of an average temperature originates from the fact that the forward voltages of the individual LEDs do not change simultaneously but are instead affected by the temperature profile along the series connection. The thermal impedance graphs of Fig. 2 had been derived from finite element analysis. The thermal response of the headlamp modules to an overall thermal power of 1 W was calculated and, both, the normalized hot-spot and the normalized averaged thermal transient are depicted in the graph. For stationary operation, the difference can be quite significant, e.g., for our simulation it was approximately 10%. Furthermore, when the same simulation model as shown in Fig. 2 is used and compared to results obtained from transient
[a.u.]
RAM
0.0 1E-6
1E+0
1E+3 1E+6 # Data Points
1E+9
Fig. 4. Illustration: Parasitic electrical transient (green line) superimposed on the actual thermal signal (red line) results in the real measurement signal (blue line) and artifacts in the Rth extraction. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
measurements, the thermal impedance graphs differ as shown below (Fig. 3). The discrepancy between experiment and simulation is either caused by wrong assumptions for the simulation model (e.g., a heat capacity which is assumed too low) or it occurs due to measurement parasitics as indicated in Fig. 4. Simulations revealed that the heat capacity of the pn-junction would have to be increased tenfold in order to compensate for the shift. Since this is far from physical reality, the mentioned discrepancy is expected to be caused by parasitic capacitances in the measurement setup (Fig. 4). Hence, the difference between
S. Müller et al. / Microelectronics Journal 44 (2013) 1005–1010
experiment and simulation rather represents an offset than a logarithmic time shift. In the meanwhile, this finding has also been incorporated in the new JESD51-14 standard via a square root extrapolation of the measurement data at small times.
1007
DUT
DUT
3. Experimental investigations In order to investigate the influence of the environment on the point of separation itself, a 5-chip headlamp module was characterized first by using the transient dual-interface methodology. Additionally, in a third measurement, the module was attached to the heat sink using highly thermal conductive glue (see Fig. 5). As it can be seen from Figs. 5 and 6, the point at which the graphs separate changes when the module attach is improved or deteriorated. This artifact was further investigated by building up a new heat sink setup in order to create a more ideal device cooling. This cooling solution was chosen due to its easy manufacturability and its large cooling capacity [6]. The goal of this kind of heat sink solution was to provide an optimized heat extraction from the device. For a perfect heat sink, no point of separation would be necessary to determine RthJC and the stationary value of the thermal impedance graph would correspond to the real junction-to-case thermal resistance of the device. However, this is not possible since the performance of current high-end cooling solutions is limited to heat transfer coefficients up to approximately 105 W/m2 K [7]. Therefore, the difference
Inlet
Outlet
Fig. 7. Schematic cross-section of the new jet impingement heat sink setup.
4
Zth [K/W]
3
Fig. 8. First prototype of the new heat sink setup. Shown are two 5-chip headlamp modules mounted right above the two jets (not visible).
w/o TIM w/ TIM Glued
3
2
0 1E-6
1E-4
1E-2 Time [s]
1E+0
1E+2
Zth [K/W]
1 2
w/o TIM w/ TIM Jet
1
Fig. 5. Thermal impedance graphs for a headlamp module attached to the heat sink without thermal grease (w/o TIM), with thermal grease (with TIM) and using highly thermal conductive glue.
0 1E-6 2.5
Zth [K/W]
1E-2 Time [s]
1E+0
Fig. 9. Thermal impedance graphs for a headlamp module attached to the heat sink without thermal grease (w/o TIM) and with thermal grease (with TIM). The dotted line corresponds to the thermal transient impedance derived by using the new jet impingement cooling setup.
2.3 2.1 1.9 w/o TIM w/ TIM Glued
1.7 1.5 2E-2
1E-4
2E-1 Time [s]
2E+0
Fig. 6. Magnified points of separation. Shift of the separation point is observable for the module which is glued to the heat sink.
between a point of separation derived from measurements using jet impingement cooling and the ideal RthJC from simulation was of further interest. The results obtained by using the new heat sink setup (Figs. 7 and 8) are shown below (Figs. 9 and 10). However, a perfect module attach can never be realized by an experimental setup. But it is possible to simulate this situation by e.g., Finite Element Analysis. By using this approach it is possible to analyze the correspondence of the point of separation and the real junction-to-case thermal resistance (Fig. 11).
1008
S. Müller et al. / Microelectronics Journal 44 (2013) 1005–1010
4. Results From Simulation
2.5
Zth [K/W]
2.3
w/o TIM w/ TIM Jet
2.1 1.9 1.7 1.5 2E-2
2E-1 Time [s]
2E+0
Fig. 10. Magnified point of separation. Shift in point of separation again present if stationary value of jet impingement cooling is not taken as RthJC.
In order to investigate and quantify the effects of the environment on the point of separation in more detail, simulation models of two geometries had been constructed. The ideal model was created in a way that it represented the real headlamp package as detailed as possible and an ambient temperature boundary condition was applied to the case of the package (Fig. 11, left image). In the second model (Fig. 11, right image) a copper plate and a heat sink were added in addition since this represents the real measurement setup as it is used at OSRAM OS. Fig. 12 shows a more detailed description of the model. For all simulation models, die attach (Solder), adhesive, thermal interface material (TIM) and the thermal grease layers were modeled as lumped thermal resistive layers with individual thermal conductivities in [W/m K]. LED chips were assumed to
Fig. 11. Simulation models created for spreading point analysis. On the left, the ideal model is shown, on the right, the ideal model is attached to a heat sink as it is the case for a real measurement setup.
5 Dies
Solder
Adhesive
MCPCB
Cu Plate
Thermal Grease
Ceramic submount
TIM
Heat Sink
Fig. 12. Domains of the extended simulation model incorporating the heat sink setup. Thin thermal resistive layers had been applied to Solder, Adhesive, TIM and Thermal Grease.
S. Müller et al. / Microelectronics Journal 44 (2013) 1005–1010
nðk∇T Þ ¼ hðT amb T Þ
ð9Þ
These investigations were extended by a larger batch of stationary simulations. Referring back to the thermal resistance definitions in chapter II, the following graph could be deduced. Here, RMM, RAA, RMA and RAM represent the thermal resistance of the device itself, i.e., in a way RthJC, whereas RMAMB and RAAMB rather correspond to RthJA (the junction-to-ambient thermal resistance) since they incorporate the ambient temperature in their definition (see Eqs. (7) and (8)). Fig. 15 shows that the device resistances stay rather constant in the range of 100–1000 W/m2 K whereas more significant variations occur for larger heat transfer coefficients. This corresponds to the observed heat flow paths (see Fig. 14) which change with varying heat transfer coefficients. All defined thermal resistances saturate for very large heat transfer
Zth [K/W]
3
2
λTIM = 0.025 λTIM = 0.1 λTIM = 0.5 λTIM = 3 λTIM = 50 Ideal
1 0 1E-6
1E-4
1E-2 1E+0 Time [s]
1E+2
Fig. 13. Comparison of the thermal impedance derived from the ideal model and the extended model. If the module is attached in a very poor manner, the point of separation changes.
Fig. 14. Stationary temperature fields and heat flow paths for different qualities of the device attach (in terms of h, from top to bottom: 102, 103, 104, 105 and 1015 W/m2K).
8 7 Rth [K/W]
be surface heat sources so that a homogenous heat generation was assumed. This represents a simplification of the reality since the areas of highest current density and therefore highest thermal power generation are near the bond contacts. However, this was neglected for simulation purposes. All the other external boundaries were chosen to be adiabatic, i.e., no heat exchange was allowed. This is justified if no heat transport via radiation or convection is assumed. The only boundary which differed from this assumption was the one on the bottom of the MCPCB and the heat sink respectively at which room temperature was applied as fixed temperature boundary condition. The thermal conductivity of the TIM was then varied and by that, the points of separation were compared to the thermal impedance derived from the ideal model. The simulated curves in Fig. 13 show that the point of separation is shifted when the device attach to the heat sink is improved or deteriorated. This confirms the experimental findings of Section 3, and indicates that the point of separation has no real physical meaning but is a rather mathematical procedure for extracting RthJC of an LED module. Especially the derivation of one single junction-tocase thermal resistance independent of the boundary conditions (besides others, the quality of the device attach) seems doubtful, since the heat flow path inside of the device changes depending on the applied thermal boundary conditions at the case. In order to illustrate this dependency in more detail, a twodimensional simulation model as depicted in Fig. 14 was derived, and the evolving temperature fields inside the module were investigated. The quality of the device attach was modeled in terms of a heat outflow boundary condition at the bottom of the MCPCB using the heat transfer coefficient h (all other boundary were chosen according to the 3D models).
1009
6 5 4 3 2 1E+0
RMAMB RMA RMM
1E+2 h
RAAMB RAA RAM
1E+4
1E+6
[W/m2K]
Fig. 15. Defined thermal resistances vs. heat transfer coefficient. In certain regions, the defined junction-to-case thermal resistances change significantly depending on the module attach.
coefficients, i.e., when the module is attached ideally to the ambient so that the heat can propagate straightforward through the device and almost no heat spreading occurs. The special behavior of RMM and RAM, i.e., that the thermal resistance increases with increasing heat transfer coefficient, had also been shown by Schweitzer [5] in a similar form. Both resistances merge with RMA respectively RAA for large h since the average temperature at the case gets closer to the maximum temperature. Moreover, the thermal resistance of the environment becomes negligible when large heat transfer coefficients can be provided, which is shown by the merging of RMAMB and RMA respectively RAAMB and RAA.. The fact that all of the device resistances again decrease for very small heat transfer coefficients might be understood in terms of thermodynamics: It is always necessary to keep in mind that the thermal resistance of the complete system needs to be minimized at the end. Hence, e.g., in the interval [100;1000] the heat flows such that a lumped thermal resistance RMA or RAA is at a maximum whereas the overall system resistance is minimized. For heat transfer coefficients smaller than 100 W/m2 K, the heat needs to flow through a larger cross-sectional area at the bottom of the LED package in order to keep the system's thermal resistance as low as possible. In that instant, the defined lumped thermal resistances of the LED package might well be lower than for the case when heat transfer coefficients larger than 100 W/m2 K are present. At the end it should also be mentioned that these investigations need to be extended to include the electrical domain as well in order to give an even deeper insight into the accordance of experiment and simulation. It can well be understood that
1010
S. Müller et al. / Microelectronics Journal 44 (2013) 1005–1010
variously changing forward voltages of series connected LEDs inside a headlamp module cause a change in electrical and therefore thermal power consumption. This coupling can be taken into account by a combined electro-thermal simulation of high-power LED headlamp modules.
to a specific application, respectively heat sink solution. Ongoing work will focus on systematizing this approach and on investigating its feasibility for practical use.
References 5. Conclusions For both, experiments and simulations, the dual-interface method leads to varying results for RthJC depending on the thermal boundary conditions at the case. A more detailed analysis regarding the influence of the environment (summarized in the heat transfer coefficient h) on the thermal resistance of the module itself showed, that it is generally not possible to specify one single junction-to-case thermal resistance. Even though it is an improvement to the previous state-of-the-art to standardize the measurement and extraction procedure of RthJC (as it is stated in the new JESD51-14 standard), the intricate interplay of the module with the complete system assembly is not yet taken into account. A suggestion to improve this situation is to provide Rth vs. h graphs for a specific LED module so that the customer is able to estimate the thermal performance of the LED package with respect
[1] Electronic Industries Association, JEDEC 51 Standard: Methodology for the Thermal Measurement of Component Packages, 1995. [2] D. Schweitzer, Transient dual interface measurement of the Rth-JC of power packages, in: Proceedings of the Therminic Conference, Rome, 2008, pp. 14–19. [3] JEDEC Solid State Technology Association, Transient dual interface test method for the measurement of the thermal resistance junction to case of semiconductor devices with heat flow trough a single path, JEDEC Standard, 2010. [4] D. Schweitzer et al., How to evaluate transient dual interface measurements of the Rth-JC of power semiconductor packages, in: Proceedings of the 25th IEEE SEMI-THERM Symposium, 2009, pp. 172–179. [5] D. Schweitzer, The junction-to-case thermal resistance: a boundary condition dependent thermal metric, in: Proceedings of the 26th IEEE SEMI-THERM Symposium, 2010, pp. 151–156. [6] System Design and Analysis Inc., Spray cooling technologies market investigation, Indianapolis, 2000 〈http://www.navsea.navy.mil/nswc/crane/sd18/Public%20Documents/Reports/LibrarySprayCool.pdf〉. [7] C.J.M. Lasance, R.E. Simons, Advances in high-performance cooling for electronics, Electronics Cooling, November 2005. 〈http://www.electronics-cooling. com/2005/11/advances-in-high-performance-cooling-for-electronics/〉.