In recent years, high-power, high-brightness Light Emitting Diodes (LEDs) have penetrated into an ever-increasing number of lighting applications. For such devices, maintaining a low die temperature is becoming a huge challenge because of the escalating power density (e.g., 200-300 W/cm2 for the latest generation). Active cooling solutions are rarely considered as a viable solution due to cost, noise and reliability concerns. Furthermore, heat removal through radiation, which is effective in traditional light sources, has a small to negligible effect in LEDs due to their relatively low die temperature.
The die temperature of any LED devices, often addressed as the “junction” temperature, has a profound effect on the optical performance and the reliability. For instance, higher die temperature is typically associated with decreased luminous efficacy (measured in lumen per electrical wattage) and color points/color temperature shifts. The deterioration in luminous efficacy is in large part due to a drop in internal quantum efficiency. Extended exposure to elevated temperature could also cause the degradation of the ohmic contacts  or the phosphor materials [1-2]. The degradation usually leads to a gradual decay of luminous efficacy and sometimes catastrophic failures .
The temperature measurement of LED die often involves the measurement of certain electrical or optical parameters of the device under test; the temperature can then be derived by knowing its relation with the measured parameters. The forward-voltage based method is by far the most popular method . Infrared imaging can generate temperature maps of the die surface and is therefore advantageous when temperature gradients are important. Methods based on peak wavelength shift , liquid crystal coating  or spectral analysis  are also reported in the literature.
Despite its popularity, single-valued junction temperature causes ambiguity when temperature gradient becomes significant. It is commonly believed that junction temperature approximates the mean die temperature, although this is yet to be proven. The validity of associating junction temperature with device reliability is also being questioned . In high-power LED devices, overheating related failures are often localized phenomena that may be best correlated with the temperature at the very same location.
As in IC packages, thermal resistances are widely used to characterize LED packages. ΘJB and ΘJC are far more popular than ΘJA. This is because of the presence of heat slugs in nearly all high-power LED packages. When using current JESD51 thermal standards  for LEDs, ambiguity often arises in terms of environmental conditions and/or the reference temperatures. There is an urgent need to standardize thermal resistance measurement and reporting for LEDs. The standards should benefit both the LED vendors and the end users by reducing inconsistency in thermal data.
LED Die Temperature Measurement
The forward voltage of a LED die can be related to its temperature via a coefficient called K-factor. Within even the same manufacturing batch, the variation in K-factor for LED packages is often such that calibration for all individual packages is required.
Measuring the forward voltage response to a power step function, the transient method can generate the so-called structure functions. These structure functions, when combined with the actual device structure, can provide physical insights into the heat transfer path .
IR thermography is another method to measure the LED die temperature. The main drawback is that only the surface temperature can be accurately measured. In naked LED dice, die surface temperature approximates the junction temperature well due to small temperature gradient within the die thickness. For phosphor encapsulated dice, however, the measured temperature is usually higher than the junction temperature due to the additional thermal dissipation in the phosphor as a result of optical losses.
In IR thermography, accurate emissivity is the key to achieve high measurement precision. One complication is that the LED die material could be transparent to the wavelength of the IR camera. Most relevant materials are opaque to long wavelengths (8-10 �m) cameras. Medium wavelength cameras (2.5-5 �m) offer better spatial resolution but are usually more difficult to apply since many ceramic and semiconductor materials are transparent or partially reflective in this wavelength range. The minimal spatial resolution that can be achieved with long wavelength cameras is around 20 �m.
Spectral-based methods measure the peak wavelength shift of the LEDs and converts it to temperature shift by the known linear relation between the two. It is often difficult to implement in practice, however, because of the small magnitude of the effect (0.05 – 0.10 nm/K for AlInGaP devices and 0.04 � 0.05 nm/K for InGaN devices). With the relatively broad LED emission peaks, it is very difficult to measure the wavelength shift accurately enough to derive an accurate temperature estimate.
Figure 1. Ratio of peak die temperature and mean die temperature, as a function
of current, for a high-power LED. All temperatures measured by an IR camera.
The True Meaning of “Junction Temperature”
In high-power LEDs, the temperature gradient is caused by the current crowding near the ohmic contacts, the poor lateral heat spreading within the thin-film structured die and by the limited heat carrying capability of the interconnect. Using an IR camera, we measured the die surface temperature of a high-power, single die LED package, which is heat sunk at the case. Figure 1 plots the ratio of the peak die temperature and the mean die temperature as a function of current. The curve indicates that the higher the current, the higher the temperature gradient at the die. For comparison, the same set of packages is measured using the forward voltage based method. Figure 2 illustrates the ΘJC comparison between the two methods for four sample devices.
Figure 2. ΘJC comparison between IR method and forward voltage method. For IR, “junction”
temperature is taken both as mean die temperature and peak die temperature.
For the IR method, the “junction” temperature is taken as both the mean die temperature as well as the peak die temperature. Contrary to common belief, in this particular case, the junction temperature is closer to the peak die temperature than to the mean die temperature. In a broader study that involves multiple LED package types, the junction temperature is shown to approximate the mean die temperature with reasonable accuracy. Pending further study, the inconsistency in correlation is believed to be related to the different current density distributions.
Component-Level Thermal Characterization for LED Packages
The thermal resistance-based metrics, defined in JESD51 for IC packages, can be largely extended to cover LEDs with some revisions. For example, the standard thermal test boards defined in JESD 51 no longer apply to LED packages due to their non-conforming foot-prints. A universal, LED package -friendly test board needs to be defined and standardized. In addition, the measurement of the board temperature and the case temperature must be re-interpreted in the context of LEDs. Finally, for non -surface mountable LED packages, the attachment procedure, which often involves the use of thermal interface materials and mounting screws, needs to be standardized in order to minimize the unnecessary variation in temperature measurement.
Multiple-die containing LED packages bring a new issue to thermal resistance-based metrics: No thermal resistance can be defined for a system that contains multiple heat sources. Fortunately, in most such packages, all individual dice are the identical type, they are connected in series and dissipate an equal amount of heat. Furthermore, all dice are attached to a common carrier or a heat slug. To stick to the classical thermal resistance definition, a common practice is to measure the pseudo junction temperature by measuring voltage drop across all dice. The pseudo junction essentially lumps all individual dice into one and discards any temperature gradient among dice.
For a densely populated LED array, due to strong thermal cross talk, the pseudo junction temperature could deviate significantly from the “true” junction temperature, measured at the die having the highest mean temperature. There are two remedies to overcome this: 1. Measure junction temperature when only one die is active (individual addressability is required) and use the superposition method to calculate the peak junction , or 2. Use a non-voltage based technique such as the IR method.
Given the ever-increasing energy “loss” in the form of light, how and whether to correct the total power when calculating thermal resistance is another unique issue in LED thermal characterization. In the most rigorous definition, heat equals total electrical power in minus the optical power out. This may sound simple but could be more difficult to implement practically because: 1. The additional optical characterization increases the complexity of the measurement system, and 2. Optical correction as a function of current and case temperature needs to be provided, together with all thermal resistance values, so that the junction temperature can be correctly “recovered” by the end user.
Realizing the complexity involved, the authors favor the use of total electrical power with no optical correction. No ambiguity would arise from such definition should the end user use the same power term. However, because the optical efficiency is strongly junction temperature dependent, the inclusion of the optical power causes the thermal resistance to “inherit” this dependency as well. In other words, part of the change in thermal resistance at different junction temperature is not real and does not reflect any change in heat transfer behavior of the system. For LEDs that operate at multiple current levels/heat sink temperatures, thermal resistance should be measured at all possible working conditions.
System-Level Thermal Characterization for LED Packages
The thermal resistances reported by LED vendors can be used in system-level analysis only to a certain extent. This is because of the difference between the environmental conditions at which LEDs are characterized and the real-life condition under which they operate. Besides that, in many applications more than one LED package or multi-die packages are applied. A Boundary-Condition Independent (BCI) Compact Thermal Model (CTM) is an ideal way to exchange thermal information between the LED vendors and the end users . In the case of high-power, single die LED packages, the CTM reduces to a single thermal resistance when thermal behavior is dominated by a designed -in, low resistant thermal path. In multi-die packages, the situation is more complicated since the CTM will not only have to describe the behavior of the individual dice, but also the interactions between them causing the single-dice BCI approach to fail .
High-power LEDs are some of the most heat intensive applications of semiconductor devices. As such the temperature gradient at their dice can be significant causing the classical junction temperature definition to fail to capture the true thermal behavior. IR thermography can be complementary to the popular electrical method when temperature gradient is too important to ignore.
The rapid technical advancement in the field calls for suitable thermal standards. IC-devices oriented JESD51 standards can be modified to serve single-die containing LED packages. For multi-die devices with certain regularity in power pattern and packaging structure, it is also possible to extend the classical thermal resistance definition as long as the thermal cross-talk can be captured. When calculating thermal resistances, the correction of total energy by subtracting optical power makes the most physical sense but is deemed impractical. The inclusion of optical power in the total energy may lead to stronger dependency of thermal resistance to the junction temperature. This dependency needs to be reflected by performing thermal resistance measurement at multiple current/heat sink conditions.
- Meneghesso, G., Levada, S., Pierobon, R., Rampazzo, F., Zanoni, E., Cavallini, A., Castaldini, A., Scamarcio, G., Du, S., and Eliashevich, I., “Degradation Mechanism of GaN-based LEDs after Accelerated DC Current Aging,” IEDM, 2002, pp. 103-106.
- Narendran, N., Gu, Y., Freyssinier, J., Yu, H. and Deng, L., “Solid-state Lighting: Failure Analysis of White LEDs,” Journal of Crystal Growth, Vol. 268, 2004, pp 449-456.
- Barton, D., Osinski, M., Perlin, P., Helms, C., and Berg, N., “Life Test and Failure Mechanisms of GaN/AlGaN/InGaN Light Emitting Diodes,” Reliability Physics Symposium, 35th. Annual Proceedings, 1997.
- Poppe, A., Gabor, F., Horvath, G., “Electrical, Thermal and Optical Characterization of Power LED Assemblies,” Proceedings of 12th International Workshop on Thermal Investigations of ICs, THERMINIC 2006, p. 197-202.
- Shatalov, M., Chitnis, A., Yadav, P., Hasan, Md. F., Khan, Adivarahan, V., Maruska, H. P., Sun, W., and Khan, M., “Thermal Analysis of Flip-Chip Packaged 280nm Nitride-Based Deep Ultraviolet Light-Emitting Diodes,” Applied Physics Letters, Vol. 86, 201109, 2005.
- Park, J., and Lee, C., “A New Thermal Measurement Technique Using Nematic Liquid Crystals with IR Laser Illumination for Visible Light Emitting Devices,” Proc. Electronic Components and Technology Conference, Vol. 2, 2005, pp. 1607-1610.
- Gu, Y., and Narendran, N., “A Non-Contact Method for Determining Junction Temperature of Phosphor-Converted White LEDs”, Third International Conference on Solid State Lighting, Proceeding of SPIE 5187, 2004, pp. 107-114.
- Lasance, C., “Thermally Driven Reliability Issues in Microelectronic Systems: Status-Quo and Challenges,” Microelectronics Reliability, Vol. 43, 2003, pp. 1969-1974.
- JEDEC Standard JESD51, “Methodology for the Thermal Measurement of Component Packages (Single Semiconductor Device),” available for download at www.jedec.com.
- Kim, L., Shin, M., “Implementation of Side Effects in Thermal Characterization of RGB Full-Color LEDs,” IEEE Electronics Device Letters, Vol. 28, No. 7, 2007, pp. 578-580.
- Lasance, C., “Thermal Ten Years of Boundary Condition Independent Compact Thermal Modeling of Electronic Parts: A Review,” Heat Transfer Engineering, Vol.29, No.2, 2008, pp.149-169.
- Treurniet, T., Lammens, V., “Thermal Management in Color Variable Multi-Chip LED Modules”, Proceedings of the XXII-nd SEMI-THERM Symposium, San Jose, Calif., March 12-16, 2006, pp. 186-190.