Estimating the junction temperature and its dynamic behavior in dependence of various operating conditions is an important issue, since these properties influence the optical characteristics as well as the aging processes of a light-emitting diode (LED). Particularly for high-power LEDs and pulsed operation, the dynamic behavior and the resulting thermal cycles are of interest. The forward voltage method relies on the existence of a time-independent unique triple of forward-voltage, forward-current, and junction temperature. These three figures should as well uniquely define the optical output power and spectrum, as well as the loss power of the LED, which is responsible for an increase of the junction temperature. From transient FEM-simulations one may expect an increase of the temperature of the active semiconductor layer of some 1/10 K within the first 10 μs. Most of the well-established techniques for junction temperature measurement via forward voltage method evaluate the measurement data several dozens of microseconds after switching on or switching off and estimate the junction temperature by extrapolation towards the time of switching. In contrast, the authors developed a measurement procedure with the focus on the first microseconds after switching. Besides a fast data acquisition system, a precise control of the switching process is required, i.e. a precisely defined current pulse amplitude with fast rise-time and negligible transient by-effects. We start with a short description of the measurement setup and the newly developed control algorithm for the generation of short current pulses. The thermal characterization of the LED chip during the measurement procedures is accomplished by an IR thermography system and transient finite element simulations. The same experimental setup is used to investigate the optical properties of the LED in an Ulbricht-sphere. Our experiments are performed on InGaN LED chips mounted on an Al based insulated metal substrate (IMS), giving a comprehensive picture of the transient behavior of the forward voltage of this type of high power LED.

A precise estimation of the junction temperature is an important issue for all types of semiconductor devices, in general. Particulary, the current trend towards “high-power” components implicates an increasing power conversion (transfer) within a decreasing active volume. This fact requires an appropriate thermal management of the whole system on the one hand, and reliable methods for the determination of the junction temperature on the other hand. Hence, theoretical considerations combined with appropriate tools and measurement data become more and more important for thermal analysis and thermal characterization.1,2 Furthermore, the increasing power density of today’s devices, generating heat in an ever smaller volume, leads to short thermal time constants representing the temperature dynamics within the semiconductor’s junction.3,4

For light-emitting diodes (LEDs), the heat dissipation5 and the resulting temperature distribution within the chip are important, not only in terms of drop out due to failures, but also in terms the temperature dependent optical properties (like peak emission wavelength, spectral width, etc.), the decrease of efficiency,6 and the aging behavior.7,8

If color conversion elements (phosphors) are used to generate white light, it has to be considered that the color conversion process is temperature dependent and generates significant additional losses. Besides a reversible dependency of the phosphors’ characteristics, an increasing temperature activates irreversible degradation effects.9,10

In order to evaluate the thermal properties of the LED under different environmental conditions, 3-dimensional thermal simulations provide an important tool from the design phase to the practical application. Theoretically, the geometric structure of the chip and the thermal properties of all the implemented materials have to be known with sufficient accuracy for a reliable calculation of the junction temperature and the temperature distribution within the chip. But for most of the practical applications the required data (e.g. the thermal conductivity of some materials) is not fully available. In this context, it has to be considered that the properties of a thin layer with interfaces to the neighboring layers may strongly differ from the bulk material’s properties. However, when the junction temperature and the temperatures at the surfaces of the chip are known, a (reverse) thermal simulation can be applied to estimate the missing geometric or material parameters and calculate the corresponding temperature profile.

Determining the functional relation between the junction temperature and the measurable electrical quantities (i.e. electrical voltage and current) at the terminals of the LED is an essential task in thermal management. Whereas a couple of established junction temperature measurement techniques already exist with a dynamic of several dozens of microseconds, a reliable measurement in the range of microseconds and the interpretation of the electrical, thermal and optical effects is still a challenge.

The temperature in the active region of the LED, the junction temperature, is responsible for the optical and electrical properties of LEDs. This junction temperature, however, is not directly accessible for temperature measurement. Therefore, the forward voltage method11,12 is chosen as an indirect measurement principle to determine the behavior of the junction temperature. The three figures forward voltage, forward current, and junction temperature should completely determine the LED’s operation, the emitted light power and spectrum, the efficiency, and the losses. One important argument for this method is the fact that the forward voltage in a semiconductor’s junction changes instantaneously with the corresponding temperature as described by the Shockley equation for ideal p-n junctions. Even in cases where internal resistances of the chip and phonon assisted recombination/generation processes are present, the exponential relation between forward current and forward voltage as given by the Shockley equation can be used as good approximation. Hence, the relation between forward current IF, forward voltage VF, and (absolute) junction temperature TJ can be described as

IF=I0(exp(eVFnkTJ)1),
(1)

with the parameters (dark) saturation current I0, which is responsible for the main temperature dependence of LEDs, the ideality factor n, and the physical constants of electron charge e and Boltzmann constant k. While Shockley defined the ideality-factor n between 1 and 2, according to the different carrier recombination mechanisms in a single pn-junction, much higher ideality-factors have been reported in literature for multiple hetero-junction systems or multiple-quantum-well (MQW) systems.13 

The calibration of the relation between forward voltage VF, forward current IF, and junction temperature ϑJ is done by pulse measurements. The main objective is to apply a rectangular constant current pulse for a duration of typically 50 μs with a repetition time of 1 s at a controlled constant sample temperature, which can be measured at the LED-submount. The heat generated by this pulse is considerably small, so that there is no significant increase in temperature in regions next to the active layer. Consequently, even when applying the current pulse, the temperature of the whole LED chip, including the junction, can reliably be assumed as constantly equal to the defined temperature. First, an accurate estimation of the junction temperature requires a high-performance control of the forward current such that the time dependent IF(t) approximates an ideal current step with a constant amplitude. Second, the data acquisition of the forward voltage VF(t) has to be fast and precise.

Most of the commercially available measurement systems evaluate the measurement signal some multiples of 10 μs after the edge of the current pulse, until all transient effects have vanished, to get reliable values. The signals immediately after the pulse’s edge are extrapolated ex-post in order to estimate the junction temperature at the pulse start-time. Furthermore, an additional calibration measurement step in a climatic chamber is required to relate the measured forward voltage to a junction temperature value.

For all of the following investigations, a Cree® EZ900TM LED chip is used as experimental object. The LED chip with an active InGaN MQW-layer-structure metallically bonded to a silicon substrate has a square shape area of 880 x 880 μm, a thickness of 170 μm, and is mounted on an Al-based insulated metal substrate (IMS). Detailed information about the LED chip can be found in the corresponding data sheet.14 

Fig. 1 shows the simplified structural layout of the LED chip bonded to the IMS submount, which represents the basis for thermal simulations.

FIG. 1.

Schematic cross-section of the LED chip mounted on IMS.

FIG. 1.

Schematic cross-section of the LED chip mounted on IMS.

Close modal

1. Basic principle of the measurement

Since the measurement principle is based on a high-performance current control and a fast data acquisition, a real-time PC measurement system is used to reliably set output data and read input data at a samplerate of 1 MHz. In order to realize the required sampling rates, the National Instrument® I/O-cards are connected via PXI (PCI eXtensions for Instrumentation) bus to a PC running a real-time operating system.

The whole experimental setup is sketched in Fig. 2.

FIG. 2.

Schematic view of the measuring setup.

FIG. 2.

Schematic view of the measuring setup.

Close modal

In order to provide a well-defined thermal environment for the measurement, the LED-IMS-sample is mounted in a Cu-sample holder. Beside a precise and reliable fixation of the sample, the main task for the sample holder device is to provide a constant temperature of the IMS submount of the LED chip to achieve repeatable boundary conditions for the measurement. The IMS temperature is measured by a small Pt100 element, a Peltier element is used to control this temperature to an accuracy of about 0.01 K. The control algorithm is realized in the PC-DAQ-system. This high accuracy in temperature control allows a precise estimation of the emissivity during the calibration of the IR thermography system. Further, a water cooler keeps the back side of the Peltier element on a nearly constant level of temperature in order to ease the environmental condition for the temperature control. The copper blocks on the hot side and the cool side of the Peltier device ensure a homogeneous temperature distribution.

For the measurement of the temperature on the front surface of the LED, an IR thermovision camera AGEMA THV® 900 LW is used. The IR spectral range of 8 - 10 μm ensures that the LED chip appears as plain IR-emitting surface to the thermography system, only the emissivity of this surface has to be determined to interpret the IR-signal as temperature. The LED chip can be regarded as gray-body. For smaller IR-wavelengths, the LED chips top-layers would be semi-transparent, so that the measured IR radiation profile could not be assigned to a defined surface, neither the surface of the top-layer nor the surface of the active LED-region (junction). The issue of IR-transparency becomes strongly apparent when the LED chip is covered by a glob top of significant thickness.

2. Temperature measurement

In summary, the following temperatures are recorded during the measurement via Pt100 elements and miniaturized K-type thermocouples:

  • ϑCool - temperature of the IMS substrate to which the LED chip is soldered, sensed by a Pt100 sensor. The IMS acts as heat sink for the LED chip. In our measurement setup, the Peltier element is used to control this temperature.

  • ϑPeltier - temperature of the Cu block between Peltier and IMS. The IMS is attached to the Cu block by thermal grease. For most practical applications this temperature would correspond to the cooling element or heat sink.

  • ϑAmb - ambient temperature.

  • ϑAir - air temperature in front of the LED (5 mm in front of the LED surface on the top edge of the chip), used to quantify the convective cooling of the LED for thermal simulations.

As described in the previous section, the IR thermography-system detects the temperature ϑIR on the top surface of the LED, which is only some μm above the junction. For a correct interpretation of the received radiation as a temperature, the emission coefficient on the LED’s surface has to be determined via calibration measurement steps. The Pt100-element is used to control the surface temperature ϑCool of the IMS, i.e. the LED-submount, to several constant values between 25°C and 70°C. From the recorded thermograms the emissivity of the LED surface is calculated, when the LED is inactive.

As the emission-corrected ϑIR is an adequate representation of the junction temperature, it is used to check that there is no cumulative heating of the LED as a consequence of several current pulses (50 μs pulse width) applied in sequence (1 s period).

3. Control of the forward current pulse

The objective of the pulse control algorithm is to realize a constant current pulse with a precisely defined target amplitude IF* and pulse width to the LED chip within the MHz sampling times. As one main parameter, the junction capacitance CLED has to be considered. The basic principle of the current pulse control process is shown in Fig. 3.

FIG. 3.

Schematic diagram of the pulse control principle: the amplifier output voltage VG is adjusted to realize a constant current pulse IF(a) Adjust the target value of the current pulse amplitude, (b) optimize the rising edge of the current pulse.

FIG. 3.

Schematic diagram of the pulse control principle: the amplifier output voltage VG is adjusted to realize a constant current pulse IF(a) Adjust the target value of the current pulse amplitude, (b) optimize the rising edge of the current pulse.

Close modal

Assuming that the measurement current pulse edge should be applied at time t = 0, the amplifier output voltage VG is set to a value just below the conducting state voltage of the LED at t = −10 μs. The junction capacitance is charged to the maximum possible extent, the charging current approaches to zero after CLED has been fully charged, i.e. VF=VG and IF=0 at t = 0. From the measured voltage step response the relevant time constant τLED=RSCLED can be estimated to help with the control of the following forward-current pulse.

Starting with the pre-charged junction capacity, the pulse generation algorithm works iteratively. In a first step, a rectangular voltage pulse is applied at t = 0, so that the LED current increases exponentially according to the time constant τLED. The amplitude of this voltage pulse is increased with every iteration step, until the value of the LED current at t3τLED corresponds to the target value IF*.

In a second step, the amplifier output voltage for t<3τLED is iteratively increased such that the LED current approaches an ideal rectangular step function with the fastest possible rise-time without introducing significant oscillations.

In the third step the voltages for times t>3τLED are iteratively adjusted to generate an almost perfectly constant current level. Due to the diode characteristic any deviation of the forward current from the chosen value IF* would generate a variation of the measured forward voltage, which must not be interpreted as temperature effect. This optimized waveform is stored in the system for different operating conditions, and applied for the subsequent experiments.

For the evaluation of the optical properties of the LED, an Ulbricht integrating sphere is used. The complete sample-setup with the water-cooler and the Peltier-element is placed in the Ulbricht-sphere. Depending on the pulse width tPulse (ranging from 3 μs to 50 μs) and the temperature on the LED’s submount ϑCool (ranging from 15°C to 75°C), the emitted light power and the corresponding spectral density are determined. The amplitude of the current pulse is 350 mA for all the data shown in the following figures.

From the results of the power spectral density measurements shown in Fig. 4 it can be concluded that the center frequency shifts with the cooler temperature, but does not depend on the applied current pulse width, since all the curves with different tPulse and the same ϑCool are almost exactly superimposed. This is also reflected in Fig. 5, which depicts the center wavelength λCenter of the power spectral density depending on tPulse and ϑCool.

FIG. 4.

Power spectral density measurement via Ulbricht integrating sphere. Results for different pulse widths and cooler temperatures.

FIG. 4.

Power spectral density measurement via Ulbricht integrating sphere. Results for different pulse widths and cooler temperatures.

Close modal
FIG. 5.

Center frequency of power spectral density in dependence of cooler temperature and pulse width.

FIG. 5.

Center frequency of power spectral density in dependence of cooler temperature and pulse width.

Close modal

In principle, the emitted optical power POptic as well as the electrical input power PElec decrease with increasing cooler temperature ϑCool. Due to temperature dependent electron phonon interactions (generally leading to a decreasing band-gap energy with increasing temperature and thus to an increasing center wavelength λCenter) and temperature dependent non-radiative losses, the emitted optical power decreases with temperature. Besides Shockley-Read-Hall (SRH) recombination,15 also other mechanisms should be considered, such as temperature-assisted electron leakage16 with possible trap- and field-assisted effects,17 as well as exciton delocalization18 that is mainly observed at low temperatures. The electrical input power decreases with increasing temperature, because of the negative temperature coefficient of the LED’s forward voltage at constant forward current. Furthermore, the ratio between the emitted light power and the electrical input power POptic/PElec gives the efficiency η of the LED. As shown in Fig. 6, the efficiency decreases approximately linear with the temperature level, and slightly with decreasing pulse width. One part of this apparent decrease with decreasing pulse width is due to the 1 MHz pulse control-rate, which reduces the effective electrical pulse width for short pulses. Another part could be explained by the time-constants of the radiative recombination, which effects a delay of the optical output with respect to the electrical power input.19 The optical pulse appears shorter than the electrical pulse by some nanoseconds, accordingly the apparent efficiency is reduced.

FIG. 6.

LED efficiency in dependence of cooler temperature and pulse width.

FIG. 6.

LED efficiency in dependence of cooler temperature and pulse width.

Close modal

However, besides the already known temperature dependent optical effects, the LED’s efficiency also depends on the pulse width related to dynamic effects, particularly for pulse durations below 10 μs. A contemporary investigation of simultaneously measured optical and electrical responses of quantum-well LEDs is presented in Ref. 20.

In principle, the electrical properties of the measuring setup can be described with the equivalent circuit model depicted in Fig. 7.

FIG. 7.

Equivalent circuit diagram of the measurement setup.

FIG. 7.

Equivalent circuit diagram of the measurement setup.

Close modal

The LED chip itself can be modeled as an ideal diode Dideal described by the Shockley equation 1, an intrinsic path resistance RLED, and a junction capacitance CLED. The latter has to be considered in the current control algorithm, particularly when a current pulse with steep edges should be applied. Further, the measurement results give reason to include a parasitic inductance LM related to the wiring of measurement setup. Strictly, CLED contains the capacitance of the LED and the LED submount.

This equivalent circuit model is used to estimate the output voltage of the power amplifier, which is necessary to gain a steep edged current pulse, reaching the desired pulse-amplitude within a few sampling points, i.e. after a few microseconds, and avoiding oscillations. Furthermore, the total forward current can be split up into the part passing the semiconductors junction and the part for recharging the junction capacity.

The LED’s path resistance can be identified from the deviation of the measured IF(VF) characteristic of the LED from the ideal Shockley relation. For the identification of the junction capacitance, a voltage step below the LEDs threshold voltage is applied so that the resulting current charges the capacitor only, without acting as light producing forward current.

Finally, the identified values for the parameters of the equivalent circuit are:

RS1.02ΩRLED501mΩCLED10nFLM1.5mH

As already mentioned, the parameters of the equivalent circuit model result from fitting the measurement data on the LED module to the behavior of an ideal LED. Thus, they need not necessarily reflect physical properties of the semiconductor.

Based on the measurement setup shown in Fig. 2 and the equivalent circuit model according to Fig. 7 the resulting electrical measurement data and temperatures for a typical pulse-experiment are presented in the this section.

As an example, the measured voltage and current values for 100 μs pulses with a repetition rate of 1 s at ϑCool= 50.7 °C are given in Fig. 8. From the forward current curve IF(t), it can be concluded that the target value is reached in the first μs with a reasonable accuracy of about 1 mA. The corresponding forward voltage VF(t) decreases about 20 mV within the pulse width of 100 μs.

FIG. 8.

Forward current and forward voltage of a typical pulse with 350 mA amplitude at ϑCool= 50.7 °C.

FIG. 8.

Forward current and forward voltage of a typical pulse with 350 mA amplitude at ϑCool= 50.7 °C.

Close modal

If dynamic electrical effects are disregarded, the voltage decrease can be fully assigned to an increase in the junction temperature. Under this assumption, the time course of the junction temperature is shown in Fig. 9.

FIG. 9.

Calculated junction temperature for a typical pulse (350 mA current pulse amplitude).

FIG. 9.

Calculated junction temperature for a typical pulse (350 mA current pulse amplitude).

Close modal

The corresponding temperature values for the pulse measurements are:

ϑAmb26.8 °CϑAir27.1 °CϑCool50.7 °CϑIR51.2 °C-51.5 °C

From these temperature measurements it can be proved that the self-heating during the pulsed operation of the LED amounts to less than 1 K on average, and the temperature variation is about 0.3 K.

A typical thermogram for a corresponding DC-measurement at IF = 350 mA and ϑCool = 25 °C in thermal equilibrium is depicted in Fig.10, showing a LED temperature of ϑIR 50 °C.

FIG. 10.

Typical emission-corrected thermogram of the LED for DC operation with IF = 350 mA at ϑCool = 25 °C. The xy-coordinates are shown as pixel values with ≈ 150 μm/pixel.

FIG. 10.

Typical emission-corrected thermogram of the LED for DC operation with IF = 350 mA at ϑCool = 25 °C. The xy-coordinates are shown as pixel values with ≈ 150 μm/pixel.

Close modal

Due to its generic approach of model definition and its efficient calculation algorithms, the finite element package GetDP (distributed under GNU General Public Licence) is used for transient heat-transfer FEM simulations.21 According to the internal geometric structure of the LED given in Fig. 1, a transient thermal 3D-model was created to show the time-dependency of the temperature distribution within the complete LED-sample.

The FEM model is based on the following assumptions:

  • The temperature on the bottom of the Al substrate is constant 25 °C.

  • On the top surface of the LED natural convection (h = 20 W/m2K) is assumed with respect to the ambient temperature ϑAir = 25 °C.

  • Heat generation is uniformly distributed across the active layer according to the losses of the LED. The losses PLoss=PElec(1η) are calculated based on the electrical input power PElec and the temperature-dependent efficiency η(T) taken from the LED’s data-sheet. The electrical input power is determined by the constant forward current pulse of amplitude IF = 350 mA and the temperature-dependent forward voltage VF(T) taken from the pulse measurements.

The material parameters (thermal conductivity, mass density, specific heat) for the different layers are chosen as typical values for the corresponding materials. Certain variations in the parametrization have influence on the absolute temperature values, but the qualitative temperature characteristics remains the same.

Based on the structural FEM model from Fig. 1, the temperatures’ rise along the central z-axis is given in Fig. 11.

FIG. 11.

FEM results of the time-dependent temperatures along the vertical center line through LED chip and submount (350 mA current pulse amplitude).

FIG. 11.

FEM results of the time-dependent temperatures along the vertical center line through LED chip and submount (350 mA current pulse amplitude).

Close modal

From the upper plot it becomes clear that even after 50 ms no thermal equilibrium has been reached. The lower plot shows in detail the fast temperature rise in the active LED layer and the LED chip, while the bottom of the LED chip, the adhesive layer, remains at the initial temperature level within the first 30 μs. Obviously the complete loss energy can be stored in the thermal capacity of the LED chip during 30 μs at 350 mA.

Summing up, if a current pulse is applied to the LED, the temperature in the active layers (junction temperature) is expected to increase by some tenths of K within dozens of μs. In this time-frame the bottom surface of the LED chip, which is soldered or glued to the LED submount heat-sink, does not carry any heat flux and does not show any temperature increase. It becomes clear that the accuracy of the junction temperature measurement depends on the settling time between two current pulses, which is necessary to dissipate the heat stored in the LED chip by conduction, and to reach the initial thermal equilibrium again.

In this paper an innovative principle for the measurement of the LED junction temperature via the forward voltage method is presented. In contrast to the existing techniques, the investigations explicitly concentrate on the dynamic behavior within the first tens of microseconds after the rising edge of a constant current pulse.

The basic idea is to precisely control a constant forward current pulse within microseconds. The length of the pulse defines the loss energy dissipated in the LED chip, and consequently the quality of the assumptions about the LED junction temperature and the accompanying temperature effect on the LED efficiency.

The temperature of the LED can be precisely controlled only before applying the pulse. FEM simulations allow to estimate the effect of the pulse on the LED temperature, and consequently the quality of the assumption of a constant and defined LED junction temperature.

For this purpose, a special measurement setup has been designed. The primary measurement goal is to strictly synchronously determine forward current and forward voltage for a defined temperature. On the one hand, a precise temperature control of the LED chip together with an accurate temperature measurement is provided. On the other hand, a control algorithm for the forward current is introduced, which considers parasitic reactances of the measurement setup as well as the internal electrical properties of a real LED.

The complete sample-setup with the LED-module, the electrical connectors, the sensors and the temperature control-system, has been designed as a compact unit. Consequently the complete setup can easily be integrated into different experimental systems, e.g. for IR-thermography or for optical measurement in an Ulbricht integrating sphere.

An IR thermography system is used to evaluate the quality of the assumption of an isothermal experiment by observing the temperature of the top surface of the LED chip, which is very close to the junction temperature.

Transient thermal FEM simulations provide the time-dependent temperature distributions inside the LED module. Both techniques confirm a not-significant self-heating of the LED chip within the first microseconds. Nevertheless, as presented in Ref. 22, there can be a considerable temperature change observed inside the active layer within the first 20 ns (in the charge carrier plasma) and the first 2 μs (in the crystal lattice) respectively.

In addition, optical measurements within an Ulbricht integrating sphere show that the spectral properties of the emitted light are independent of the used pulse widths.

For the assignment of the measured forward voltage to a corresponding value of the junction temperature, it has to be assumed that even in the transient case the forward voltage depends only on the forward current and the junction temperature as given by the Shockley equation, for example. Up to now the authors have not been able to prove this assumption in full extent. Transient effects of electrical nature may influence the dynamic characteristics of the LED’s forward voltage, which must not be interpreted as stationary relation between voltage, current, and temperature.

As a consequence, care has to be taken in the application of the results of forward voltage measurements to control the junction temperature and the optical output behavior for pulsed LED operation at frequencies above some 10 kHz or pulselengths below 100 μs.

Future work will be dedicated to this issue.

This research work was funded by the Klima- und Energiefonds (KLIEN) and the Österreichische Forschungs-förderungsgesellschaft mbH. (FFG), project no. 843877 “Intelligent Color Conversion” (ICOL), program “Energy Mission Austria” (e!MISSION.at).

1.
H. K.
Lee
and
J. S.
Yu
,
Jpn. J. Appl. Phys.
49
(
4S
),
04DG11
(
2010
).
2.
D. H.
Lee
,
H. K.
Lee
,
J. S.
Yu
,
S. J.
Bae
,
J. H.
Choi
,
D. H.
Kim
,
I. C.
Ju
,
K. M.
Song
,
J. M.
Kim
, and
C. S.
Shin
,
Semicond. Sci. Technol.
26
(
5
),
1
(
2011
).
3.
M.
Weilguni
,
J.
Nicolics
,
R.
Medek
,
M.
Franz
,
G.
Langer
, and
F.
Lutschounig
,
Conf. Proc. of the 33rd Int. Spring Seminar on Electronics Technology
,
238
(
2010
).
4.
W.
Lai
,
X.
Liu
,
W.
Chen
,
X.
Lei
, and
X.
Cao
,
IEEE Trans. Electron Devices
62
(
11
),
3715
(
2015
).
5.
D. S.
Kim
and
B.
Han
,
J. Appl. Phys.
119
,
125104
(
2016
).
6.
J.
Hu
,
L.
Yang
, and
M. W.
Shin
,
J. Phys. D: Appl. Phys.
41
,
035107
(
2008
).
7.
M.
Meneghini
,
A.
Tazzoli
,
G.
Mura
,
G.
Meneghesso
, and
E.
Zanoni
,
IEEE Trans. Electron Devices
57
(
1
),
110
(
2010
).
8.
S.-C.
Yang
,
P.
Lin
,
C.-P.
Wang
,
S. B.
Huang
,
C.-L.
Chen
,
P.-F.
Chiang
,
A.-T.
Lee
, and
M.-T.
Chu
,
Microelectron. Reliab.
50
(
7
),
959
(
2010
).
9.
P.
Fulmek
,
C.
Sommer
,
P.
Hartmann
,
P.
Pachler
,
H.
Hoschopf
,
G.
Langer
,
J.
Nicolics
, and
F. P.
Wenzl
,
Adv. Optical Mater.
2013
(
1
),
753
(
2013
).
10.
M.
Meneghini
,
M. D.
Lago
,
N.
Trivellin
,
G.
Meneghesso
, and
E.
Zanoni
,
IEEE Trans. Device Mater. Rel.
13
(
1
),
316
(
2013
).
11.
Y.
Xi
and
E. F.
Schubert
,
Appl. Phys. Lett.
85
,
2163
(
2004
).
12.
A.
Keppens
,
W. R.
Ryckaert
,
G.
Deconinck
, and
P.
Hanselaer
,
J. Appl. Phys.
104
,
093104
(
2008
).
13.
J. M.
Shah
,
Y.-L.
Li
,
T.
Gessmann
, and
E.
Schubert
,
J. Appl. Phys.
94
(
4
),
2627
(
2003
).
14.
Cree-Inc., Data sheet: CPR3DX, Rev. B (2015 (accessed September, 2016)), http://www.cree.com.
15.
D. S.
Meyaard
,
Q.
Shan
,
Q.
Dai
,
J.
Cho
,
E. F.
Schubert
,
M.-H.
Kim
, and
C.
Sone
,
Appl. Phys. Lett.
99
,
041112
(
2011
).
16.
K.-T.
Lam
and
S.-J.
Chang
,
J. Display Technol.
10
(
12
),
1078
(
2014
).
17.
C. D.
Santi
,
M.
Meneghini
,
M. L.
Grassa
,
B.
Galler
,
R.
Zeisel
,
M.
Goano
,
S.
Dominici
,
M.
Mandurrino
,
F.
Bertazzi
,
D.
Robidas
,
G.
Meneghesso
, and
E.
Zanoni
,
Appl. Phys. Lett.
119
(
9
),
094501
(
2016
).
18.
S. F.
Chichibu
,
M.
Sugiyama
,
T.
Onuma
,
T.
Kitamura
,
H.
Nakanishi
,
T.
Kuroda
,
A.
Tackeuchi
,
T.
Sota
,
Y.
Ishida
, and
H.
Okumura
,
Appl. Phys. Lett.
79
(
26
),
4319
(
2001
).
19.
R.
Windisch
,
A.
Knobloch
,
M.
Kuijk
,
C.
Rooman
,
B.
Dutta
,
P.
Kiesel
,
G.
Borghs
,
G. H.
Döhler
, and
P.
Heremans
,
IEEE J. Quantum Electron.
36
(
12
),
1445
(
2000
).
20.
A.
David
,
C. A.
Hurni
,
N. G.
Young
, and
M. D.
Craven
,
Appl. Phys. Lett.
109
,
033504
(
2016
).
21.
P.
Dular
and
C.
Geuzaine
, “
GetDP reference manual: the documentation for GetDP, a general environment for the treatment of discrete problems
,” http://getdp.info.
22.
W. G.
Scheibenzuber
and
U. T.
Schwarz
,
Appl. Phys. Lett.
98
,
181110
(
2011
).