Sei sulla pagina 1di 10

IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 24, NO.

8, AUGUST 2009 1967

A General Photo-Electro-Thermal Theory for Light


Emitting Diode (LED) Systems
S. Y. (Ron) Hui, Fellow, IEEE, and Y. X. Qin

Abstract—The photometric, electrical, and thermal features of applications such as automobile headlights and compact lamps,
LED systems are highly dependent on one another. By considering the ambient temperature could be very high and the size of the
all these factors together, it is possible to optimize the design of heatsink is limited. This serious thermal problem has been ad-
LED systems. This paper presents a general theory that links the
photometric, electrical, and thermal behaviors of an LED system dressed in [10] and [11]. The drop in luminous efficacy due
together. The theory shows that the thermal design is an indispens- to thermal problem would be serious, resulting in reduction of
able part of the electrical circuit design and will strongly influence luminous output [12].
the peak luminous output of LED systems. It can be used to explain Photometric parameters such as luminous flux and luminous
why the optimal operating power, at which maximum luminous flux efficacy, electrical parameters such as electric power, current,
is generated, may not occur at the rated power of the LEDs. This
theory can be used to determine the optimal operating point for an and voltage of an LED, and thermal parameters such as junc-
LED system so that the maximum luminous flux can be achieved tion and heatsink temperature and thermal resistance are closely
for a given thermal design. The general theory has been verified linked together. In [7] and [8], the relationship between the lumi-
favorably by experiments using high-brightness LEDs. nous output (photometric variables) and thermal behavior has
Index Terms—Electronic control, lighting-emitting diodes, ther- been reported. Reference [13] highlights the highly nonlinear
mal design for LEDs. thermal behavior of the junction-to-case thermal resistance of
LED with electric power consumption of LED. The junction-
I. INTRODUCTION to-case thermal resistance is affected by many factors such as
IGHT emitting diodes (LEDs) have emerged as promis- the mounting and cooling methods [14], [15], the size of the
L ing lighting devices for the future. However, LEDs are
still primarily restricted to decorative, display, signage, and sig-
heatsink and even the orientation of the heatsink [13]. Thus,
analysis on the junction thermal resistance [13], [16], [17] and
naling applications so far and have not reached the stage of thermal management [18], [19] have been major LED research
massively entering the general and public illumination markets. topics. To deal with various factors that affect the luminous out-
In photometry, one important factor commonly used for com- put, control methods have been proposed to control the luminous
paring different lighting devices is the luminous efficacy (lumen output of LED systems [20], [21]. An LED device model has
per Watt) [1]. One major hindrance to the widespread of LED been proposed to model the thermal junction resistance and the
applications in general and public illumination is that the lumi- light output [22]. But this model is for the LED device and not
nous flux of LEDs will decrease with the junction temperature for the LED system including the thermal design of the heatsink
of the LEDs [2]–[4]. This phenomenon leads to observations and the electric power control.
by some researchers and users that the maximum luminous out- In this paper, a general theory that links the photometric,
put of LEDs in some designs occurs at an operating power less electrical, and thermal aspects of an LED system is presented.
than the rated power of the LEDs [5]. It is rightly pointed out This theory is based on a simple thermal model of the LED
that the quantum efficiency and junction thermal resistance of and heatsink and can be used to predict the optimal operating
LED are the two limiting factors in LED technology [6]. The point (i.e., maximizing the luminous output) and provide design
luminous efficacy of various LEDs typically decreases by ap- parameters for optimal thermal design. Tests have been carried
proximately 0.2–1% per degree Celsius rise in temperature [5]. out to verify the general theory. Examination of the theory also
Due to the aging effect, the actual degradation of luminous ef- provides clear explanation on why the optimal operating points
ficacy could be higher than this quoted figures. Recent research of some LED systems occur in an operating power less than the
reports [2]–[4], [7], [8] have highlighted the relationship of ef- rated power of the LED. Practical results obtained in the experi-
ficacy degradation and the junction temperature of the LEDs. ments also highlight the major limitations of the existing LEDs.
Accelerated age tests carried out in [9] show that the light output Both the theory and practical results provide useful insights for
can drop by further 45%. For aged LEDs, the efficacy degra- LED system designers and allow users to determine the advan-
dation rate could be up to 1% per degree Celcius. In some tages and disadvantages of using LED in different applications.

Manuscript received August 3, 2008; revised November 28, 2008 and


February 10, 2009. Current version published August 12, 2009. This paper
II. GENERAL PHOTO-ELECTRO-THERMAL THEORY
was supported by the Strategic Research Grant of CityU SRG Project 7001993.
Recommended for publication by Associate Editor M. Alonso.
A. General Analysis
The authors are with the Department of Electronic Engineering, Centre for Let φv be the total luminous flux of an LED system consisting
Power Electronics, City University of Hong Kong, Kowloon, Hong Kong, China.
Color versions of one or more of the figures in this paper are available online of N LED devices.
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TPEL.2009.2018100 φv = N × E × P d (1)

0885-8993/$26.00 © 2009 IEEE


1968 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 24, NO. 8, AUGUST 2009

Fig. 2. Typical relationship of heatsink temperature and power dissipation.


Fig. 1. Typical relationship of efficacy of LEDs versus junction temperature.

where E is the luminous efficacy (lumen/Watt) and Pd is the


real power of one LED (W)
The emission intensity (I) of LEDs decreases with increas-
ing temperature. Near room temperature, the emission intensity
follows an exponential decay function [25].
−(T − 25 ◦ C)
I = I|25 ◦ C exp (2a)
T1
where T1 is the characteristic temperature.
In photometric terms, this relationship is reflected in the
temperature-dependent luminous efficacy (E) of LEDs as shown
in Fig. 1 (as in LED manufacturer data sheets). Being semi-
conductor devices, LEDs should not be operated at a junction
temperature above 125 ◦ C. The exponential curve of (2a) within
this practical range of operating temperature is fairly linear and
will be approximated as

E = Eo 1 + ke (Tj − To ) for Tj ≥ To E≥0 and


(2b)
where Eo is the rated efficacy at the rated temperature To (typ-
ically 25 ◦ C in some LED data sheets) and ke is the relative Fig. 3. (a) Simplified dynamic thermal equivalent circuit of N LEDs mounted
rate of reduction of efficacy with increasing temperature. For on the same heatsink. (b) Simplified steady-state thermal equivalent circuit with
example, if E reduces by 40% over a temperature increase of N LEDs mounted on the same heatsink.
100 ◦ C, then ke = −0.004.
In general, the LED power can be defined as Pd = Vd × Id , sponding thermal capacitances are needed if dynamic response
where Id is the diode current and Vd is the diode voltage. But is studied.
only part of the power will be dissipated as heat. Thus, the heat Under steady-state conditions, the thermal model can be
generated in one LED is defined as simplified into a steady-state model as shown in Fig. 3(b).
In practice, heatsink compound or equivalence may be used
Pheat = kh Pd = kh Vd Id (3) between the LEDs and the heatsink to ensure good ther-
mal contact. The thermal resistance of such thermal conduc-
where kh is a constant less than 1, and it represents the portion
tor/electric insulator is relatively small when compared with
of the LED power that turns into heat. For example, if 85%
Rjc of LEDs and is neglected in Fig. 3(b) and the following
of the LED power is dissipated as heat, then kh = 0.85. The
equations.
measurement of kh can be found in [23].
Based on the model in Fig. 3(b), the steady-state heatsink
Now consider a typical relationship of the heatsink temper-
temperature can be expressed as
ature and the heat generated in the LED system as shown in
Fig. 2. A simplified dynamic thermal equivalent circuit of the Ths = Ta + Rhs (N Pheat ) = Ta + Rhs (N kh Pd ) (4)
LED system is shown in Fig. 3(a), assuming that 1) the N LEDs
where Ta = ambient temperature.
are placed on the same heatsink with a thermal resistance of Rhs ;
From Figs. 3(b) and (4), the junction temperature of each
2) the LED has its junction to case thermal resistance Rjc ; and
LED is therefore
3) a thermal conductor with electrical isolation (such as heatsink
compound) is used to isolate the LEDs from the heatsink and it
Tj = Ths + Rjc (Pheat ) = Ths + Rjc (kh Pd ) (5a)
has a thermal resistance of Rins . A distributed thermal model is
used for the heatsink due to its relatively large size. The corre- Tj = Ta + (Rjc + N Rhs ) kh Pd . (5b)
HUI AND QIN: GENERAL PHOTO-ELECTRO-THERMAL THEORY FOR LIGHT EMITTING DIODE (LED) SYSTEMS 1969

thermal measurement reported in [23], it has been shown that


Now, Tj obtained in (5) can be used in (2b): kh will reduce slightly for a few percents under dimming condi-
E = Eo [1 + ke (Tj − To )] tions. From LED manufacturer data sheets [5], the degradation
of the efficacy with junction temperature is usually assumed to
E = Eo {1 + ke [Ta + (Rjc + N Rhs ) kh Pd − To ]} be linear and thus ke is assumed to be constant. This first ap-
E = Eo [1 + ke (Ta − To ) + ke kh (Rjc + N Rhs ) Pd ]. (6) proximation is acceptable for ke and kh and will be relaxed to
accommodate the changing nature of Rjc in the analysis later.
So, the total luminous flux φv is Based on this assumption, (8) can be simplified as
φv = N EPd dφv
= NE o {[1+ke (Ta −To )] +2 [ke kh (Rjc + N Rhs )] Pd } .
φv = N {Eo [1+ke (Ta − To ) +ke kh (Rjc + N Rhs ) Pd ]} Pd dPd
  (9)
φv = N Eo [1+ke (Ta − To )] Pd +ke kh (Rjc + N Rhs ) Pd2 . Therefore, the maximum-φv point can be obtained by putting
dφv /dPd = 0 and
(7)
[1 + ke (Ta − To )]
Several important observations can be made from (7). Pd∗ = − (10)
2ke kh (Rjc + N Rhs )
1) Equation (7) relates the luminous flux (φv ) to the electrical
power of the LED (Pd ) and the thermal resistance of the where Pd∗ is the LED power at which maximum φv occurs.
heatsink (Rhs ) and the LED junction (Rjc ) together. It is (Note that ke is a negative value.)
an equation that integrates the photometric, electrical, and From (3), the corresponding LED current at which maximum
thermal aspects of the LED system together. φv occurs can be obtained as
2) For a given heatsink (that may be restricted in size by [1 + ke (Ta − To )]
a specific application), the operating point Pd∗ at which Id∗ = − . (11)
2ke kh (Rjc + N Rhs ) Vd
maximum φv occurs can be determined. On the other
hand, it can be used for thermal design to optimize the Several significant observations can be made from (10) and
size of the heatsink (Rhs ) for a given LED system. (11).
3) Because ke is negative and less than 1, (7) is in the form 1) Equations (10) and (11) relate the operating points Pd∗ and
of φv = α1 Pd − α2 Pd2 where a1 and a2 are two positive Id∗ , respectively, to the thermal design of the LED system
coefficients. As Pd is increased from zero, φv increases (i.e., thermal resistance of the heatsink Rhs and Rjc ).
almost linearly because the second term is negligible when 2) The maximum luminous flux will occur approximately at
Pd is small. As Pd increases, the second negative term, a lamp power Pd∗ specified in (10). This Pd∗ will shift to
which is proportional to the square of Pd , will reduce a lower value if (Rjc + N Rhs ) is increased. This leads to
φv significantly. After reaching the maximum point, the the possibility that the Pd∗ may occur at a power level that
φv will drop faster as Pd and Rjc increase [due to the is less than the rated power Pd(rated) of the LED.
increasing significance of the negative terms in (7b)]. This 3) Based on the above comment, one should expect that
means that the parabola of φv is not symmetrical. Since the Pd∗ could be shifted to higher power level if a larger
the luminous flux function is a parabola and therefore has heatsink with lower Rhs is used.
a maximum value, this maximum point can be obtained 4) For many applications such as vehicle headlamps and
from dφv /dPd = 0. compact LED lamps (for replacement of incandescent
By differentiating (7) with respect to Pd , lamps), the size of the heatsink is highly restricted and
  the ambient temperature is high. In these cases, there is

 [1+ke (Ta −To )] +2ke kh (Rjc +N Rhs ) Pd  a high possibility that Pd∗ will occur at a power level less

   


 + (Ta −To ) Pd +kh (Rjc +N Rhs ) Pd2   than the rated power unless proper design is adopted.
 
dφv
= NE o dke  
2 dkh .
dPd  × + ke (Rjc +N Rhs ) Pd 

 dPd dPd 
 C. Effects of Junction-to-Case Thermal Resistance Rjc of LED
 
 + k k P 2
dRjc




 In practice, Rjc of the LED increases with lamp power [13].
e h d
dPd Therefore, a vigorous equation can be obtained from (8) as
(8)

It should be noted that the first two terms on the right-hand- dφv
side of (8) do not have derivatives, while the remaining three = N Eo [1 + ke (Ta − To )] + 2ke kh (Rjc + N Rhs ) Pd
dPd
terms do. While ke and kh do not vary significantly with Pd and

can thus reasonably be assumed constant, it must be noted that
dR jc
Rjc will indeed increase significantly and nonlinearly with the + ke kh Pd2 . (12)
dPd
LED power [13].
The function of Rjc is highly complex and dependent on sev-
B. Simplified Equations
eral factors such as thermal resistance of the heatsink, ambient
As a first approximation, it is assumed that ke , kh , and Rjc temperature, the LED size and mounting structure, and even the
are constant for the time being. From the simple open-system orientation of the heatsink [13]. Equation (7) in fact provides
1970 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 24, NO. 8, AUGUST 2009

heatsink with a thermal resistance of 6.3 ◦ C/W. The luminous


flux and efficacy of the LEDs are measured in an integrating
sphere. Parameter ke and E0 are measured from the same type
of LED samples that have been used for at least one month.
Parameter kh is shown to be 0.85 at dimmed situation and 0.9
at the rated power of LED [23]. Because of its relatively small
variation, it is kept constant at 0.85, although a more accurate
result can be obtained if kh is expressed as a function of Pd . The
parameters required for (7) are ke = −0.005, kh = 0.85, Ta =
28 ◦ C, T0 = 25 ◦ C, E0 = 41 Lumen/Watt, N = 8, Rhs =
6.3 ◦ C /W Rjco = 10 ◦ C/W and kjc = 0.1 ◦ C/W2 .
If the Rjc is assumed to be constant (i.e., Rjc = Rjco ), the
luminous flux equation follows the form of (7):
φv = 323.08 × Pd − 84.2 × Pd2 . (15)
Fig. 4. Assumed linear function of junction-to-case thermal resistance Rjc .
If Rjc is assumed to be linearly temperature-dependent (13),
the luminous flux equation follows the form of (14):
the physical meaning of the effects of the temperature-dependent φv = 323.08 × Pd − 84.2 × Pd2 − 1.39Pd3 . (16)
Rjc . Since Rjc increases with lamp power Pd , the two negative
The measured total luminous flux for eight LEDs is used for
terms (with ke , which is negative) in (7) will accelerate the re-
comparison with calculated values. The measured and calcu-
duction of the luminous flux as Pd increases. This effect should
lated total luminous flux values are plotted, not against the total
be noticeable when Pd exceeds the Pd∗ , resulting in a slightly
power sum of eight LEDs but against one LED power. (Note:
asymmetric parabolic luminous flux function.
the eight LEDs are identical and are connected electrically in
The general theory can accommodate nonlinear junction-to-
series). Using the power of one LED in the x-axis allows one to
case thermal resistance Rjc in principle. Since Rjc is a com-
check easily if the Pd∗ operating point is at the rated LED power
plex and nonlinear function of the lamp’s heat dissipation Pheat
or not. The measured results and calculated results from (15)
(which is equal to kh Pd ) and the thermal design of the mount-
and (16) are plotted in Fig. 5(a).
ing structure, the theoretical prediction is based on a simplified
Several points should be noted:
linear function as follows:
1) The asymmetric theoretical curves of φv are in good agree-
Rjc = Rjco (1 + kjc Pd ) (13) ment with the measured curve. This confirms the validity
of the general theory.
where Rjco is the rated junction-to-case thermal resistance at 2) The maximum lumen/Watt point occurs at about Pd =
25 ◦ C and kjc is a positive coefficient. A typical linear approx- 1.9 W, which is less than the rated power of 3 W. This
imation of Rjc is shown in Fig. 4. Putting (13) into (7), the lu- result shows that the general theory can predict accurately
minous flux equation that incorporates a temperature-dependent the Pd∗ operating point, which may not be the rated power.
Rjc becomes: Equation (10) indicates that a large N Rhs term will shift
Pd∗ to the low power level of the LED.
φv = N Eo {[1 + ke (Ta −To )] Pd + [ke kh (Rjco + N Rhs )] Pd2
3) The two negative terms in this example can also be seen in
+ [ke kh kjc Rjco ] Pd3 }. (14) (16). The asymmetry after the peak luminous output point
is more noticeable in the theoretical curve obtained from
III. EXPERIMENTAL VERIFICATION (16) than from (15). Comparison of (15) and (16) shows
that the effect of the variation of Rjc , which is reflected in
Three types of LEDs are used to verify the theory. They are
the extra third term in (16) is the reason for the noticeable
Luxeon K2-3W cool white LEDs and 5 W cool white LEDs
asymmetry of φv .
[5], and CREE XLamp XR-E LEDs. They are mounted on
4) In summary, the simplified model (7), which is the basis
several heatsinks with thermal resistance of 6.3 ◦ C/W, 4.5 ◦ C/W,
for (15), has the form of φv = α1 Pd − α2 Pd2 , while the
and 2.2 ◦ C/W so that experiments can be performed to evaluate
more vigorous model (14), which is the basis for (16), has
their luminous output under different lamp power operations.
the form of φv = α1 Pd − α2 Pd2 − α3 Pd2 . Therefore, the
Light measurements are conducted in an integrating sphere at
model in (14) is more accurate than the model in (7) par-
room temperature. Measurements are taken 20 minutes after the
ticularly when Pd has exceeded Pd∗ . However, since both
begining of each test to ensure that steady-state conditions have
simplified and vigorous models are accurate enough for
been reached.
the power less than Pd∗ , which is also the recommended
useable power range of LEDs, both equations can be
A. Tests on Luexon K2 Cool-White 3W LEDs
used in the design optimization procedure as explained in
1) On a Heatsink With Thermal Resistance of 6.3 ◦ C/W: Section IV.
Eight Luxeon K2 Cool-white 3W LEDs are mounted on a Based on (6), the efficacy function can also be obtained.
HUI AND QIN: GENERAL PHOTO-ELECTRO-THERMAL THEORY FOR LIGHT EMITTING DIODE (LED) SYSTEMS 1971

Fig. 5. (a) Calculated and measured total luminous flux versus lamp power
Fig. 6. (a) Calculated and measured total luminous flux versus lamp power
for eight Luxeon 3W LEDs mounted on a heatsink with thermal resistance
for eight Luxeon 3W LEDs mounted on a heatsink with thermal resistance
of 6.3 ◦ C/W. (b) Calculated and measured total luminous efficacy versus lamp
of 4.5 ◦ C/W. (b) Calculated and measured total luminous efficacy versus lamp
power for eight Luxeon 3W LEDs mounted on a heatsink with thermal resistance
power for eight Luxeon 3W LEDs mounted on a heatsink with thermal resistance
of 6.3 ◦ C/W.
of 4.5 ◦ C/W.

Assuming Rjc is constant (i.e., Rjc = Rjco ),


Fig. 6(a). The corresponding measured and calculated efficacies
E = 40.39 − 10.52Pd (17) are in good harmony as shown in Fig. 6(b). The Pd∗ is 2.4 W at
an efficacy of 21 lumens/Watt in this case. The use of a larger
Assuming Rjc is linearly temperature dependent (13), heatsink with a smaller thermal resistance means that the NRhs
E = 40.39 − 10.52Pd − 0.17Pd2 . (18) term in the denominator of (10) is smaller than that in the pre-
vious case (with Rhs = 6.3 ◦ C/W). Therefore, Pd∗ has increased
The measured efficacy values and the calculated values from from 1.9 to 2.4 W as expected from (10) and the efficacy from
(17) and (18) are displayed in Fig. 5(b). It is noted that the 21 to 23 lumens/Watt.
calculated values are generally consistent with measurements, 3) On a Heatsink With Thermal Resistance of 2.2 ◦ C/W:
except at very low power where the light output is low and the Another eight 3-W LEDs are mounted on an even larger heatsink
relative measurement error is large. The results obtained from with thermal resistance of 2.2 ◦ C/W for evaluation. The mea-
(18) are more accurate than those from (17) when Pd is large. sured and calculated luminous output as function of LED power
It is important to note that Pd∗ = 1.9 W at which the efficacy is Pd are shown in Fig. 7(a) and the corresponding results of the
only 20 lumens/Watt. If these LEDs are operated at rated power efficacy are included in Fig. 7(b).
(3 W), the efficacy will even drop to 8 lumens/Watt, which is The theoretical Pd∗ is now about 3.5 W, which is higher than
worse than that of incandescent lamps of 10–15 lumens/Watt. the rated power of 3 W, and the efficacy is 20 lumens/Watt. This
2) On a Heatsink With Thermal Resistance of 4.5 ◦ C/W: again confirms the prediction by the theory (10) that Pd∗ will
Eight 3-W LEDs are mounted on a larger heatsink with thermal shift to the higher power level with a decreasing term of N Rhs
resistance of 4.5 ◦ C/W. The measured and calculated total lumi- (i.e., a larger heatsink with a lower Rhs and/or a reduction of N ).
nous outputs as functions of single-LED power Pd are shown in Therefore, the theory can be used to design the optimal heatsink
1972 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 24, NO. 8, AUGUST 2009

Fig. 7. (a) Calculated and measured total luminous flux versus lamp power
for eight Luxeon 3W LEDs mounted on heatsink with thermal resistance of Fig. 8. (a) Calculated and measured total luminous flux versus lamp power for
2.2 ◦ C/W. (b) Calculated and measured total luminous efficacy versus lamp two Luxeon 5 W LEDs mounted on heatsink with thermal resistance of 10 ◦ C/W.
power for eight Luxeon 3W LEDs mounted on a heatsink with thermal resistance (b) Calculated and measured total luminous efficacy versus lamp power for two
of 2.2 ◦ C/W. Luxeon 5 W LEDs mounted on a heatsink with thermal resistance of 10 ◦ C/W.

(20), respectively, and they are plotted with practical measure-


for a particular operating power within the rated power range.
ments in Fig. 8(a) and (b), respectively. Despite only two 5 W
On the other hand, it can also be used to predict the optimal
LEDs are used, the theoretical predictions based on the averaged
operating power for a given heatsink. The practical optimal
values are in general agreement with the measurements.
lamp power should not exceed the rated power.
φv = 75.2Pd − 7.57Pd2 − 0.296Pd3 (19)
B. Tests on Luexon K2 Cool-White 5 W LEDs
E = 37.6 − 3.78Pd − 0.149Pd2 . (20)
In order to ensure that the theory can be applied to LEDs of
different power, 5 W LEDs are used for evaluation. They are
mounted on two heatsinks with thermal resistance of 6.8 and 2) On a Heatsink with Thermal Resistance of 6.8 ◦ C/W: The
10 ◦ C/W, respectively. previous experiments are repeated by mounting the two 5 W
1) On a Heatsink With Thermal Resistance of 10 ◦ C/W: Two LEDs on a larger heatsink with a thermal resistance of 6.8 ◦ C/W.
5 W LEDs are mounted on a heatsink with thermal resistance Fig. 9(a) and (b) shows the comparisons of the measured and
of 10 ◦ C/W. For the theoretical calculation, the parameters used calculated luminous flux and efficacy, respectively. In general,
in (10) are ke = −0.0036, kh = 0.85, Ta = 28 ◦ C, T0 = 25 ◦ C, calculated and measured results are in good agreement. Compar-
E0 = 38 Lumens/W, N = 2, Rhs = 10 ◦ C/W, Rjc = 13 ◦ C/W, isons of the peak luminous flux in Figs. 8(a) and 9(a) confirms
and kjc = 0.13 ◦ C/W2 . Fitting these parameters into (7) and as- once again that using a larger heatsink (with lower thermal re-
suming that Rjc will rise linearly with temperature, the luminous sistance) can shift the optimal operating point to the high-lamp
flux equation and the efficacy equation are expressed as (19) and power level. For the heatsink with Rhs = 10 ◦ C/W, Pd∗ occurs
HUI AND QIN: GENERAL PHOTO-ELECTRO-THERMAL THEORY FOR LIGHT EMITTING DIODE (LED) SYSTEMS 1973

Fig. 10. (a) Calculated and measured total luminous flux versus lamp power
for two CREE 3W LEDs mounted on heatsink with thermal resistance of
Fig. 9. (a) Calculated and measured total luminous flux versus lamp power for 6.3 ◦ C/W. (b) Calculated and measured total luminous efficacy versus lamp
two Luxeon 5 W LEDs mounted on heatsink with thermal resistance of 6.8 ◦ C/W. power for two CREE 3W LEDs mounted on a heatsink with thermal resistance
(b) Calculated and measured total luminous efficacy versus lamp power for two of 6.3 ◦ C/W.
Luxeon 5 W LEDs mounted on a heatsink with thermal resistance of 6.8 ◦ C/W.

corresponding results of the efficacy are included in Fig. 11(b)


at about 3.8 W. For the heatsink with Rhs = 6.8 ◦ C/W, Pd∗ has based on the following parameters:
shifted to about 6 W. ke = −0.0045, kh = 0.83, Ta = 28 ◦ C, T0 = 25 ◦ C, E0 =
78.5 Lumen/Watt, N = 8, Rhs = 4.5 ◦ C /W Rjco = 8 ◦ C/W.
C. Tests on CREE X lamp XR-E LEDs Pd∗ is now about 3 W at which the efficacy is 42 lumens/W.

1) On a Heatsink With Thermal Resistance of 6.3 ◦ C/W:


Eight cool white CREE X lamp XR-E LEDs are mounted on a IV. OPTIMAL DESIGN PROCEDURE OF LED SYSTEMS
heatsink with a thermal resistance of 6.3 ◦ C/W. The parameters In this section, a new design guideline [24] is introduced. The
required for (7) are results [Figs. 5(a), 6(a), and 7(a)] of the tests on the Luexon
ke = −0.0045, kh = 0.83, Ta = 28 ◦ C, T0 = 25 ◦ C, E0 = 3 W high-brightness LEDs mounted on different heatsinks and
78.5 Lumen/Watt, N = 8, Rhs = 6.3 ◦ C /W Rjco = 8 ◦ C//W. the theoretical curve with Rhs = 0 ◦ C/W are plotted against the
Fig. 10(a) and (b) show the measured and calculated lumi- power of one LED in Fig. 12. Typical results of Fig. 12 can
nous flux and the efficacy plots, respectively. They are in good be redrawn as those in Fig. 13 for explanation. As the cooling
agreements confirming that the general theory applies well to effect is increased (e.g., by increasing the size of the heatsink
CREE LEDs. The peak flux occurs at 2.4 W at an efficacy of and effectively reducing Rhs ), Pd∗ will shift to the high-power
39 lumens/Watt. level (e.g., shifting from A to D progressively). Three important
2) On a Heatsink With Thermal Resistance of 4.5 ◦ C/W: An- points are highlighted here.
other eight cool white CREE X lamp XR-E LEDs are mounted 1) The maximum φv is the point of inflexion of the luminous
on an even larger heatsink with thermal resistance of 4.5 ◦ C/W flux function (7) or (14). As Pd increases from zero, the
for evaluation. The measured and calculated luminous output as positive slope of the curve (i.e., dφv /dPd ) is gradually
a function of LED power Pd are shown in Fig. 11(a), and the decreasing to zero when the peak of the curve is reached. A
1974 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 24, NO. 8, AUGUST 2009

Based on the above observations, the following design rules


[24] can be used to maximize the luminous output from a LED
system:
Rule 1: The function of the luminous flux φv versus LED
power Pd is a parabolic curve with a maximum point. The
operating point Pd should preferably be chosen at or below
the maximum point Pd∗ if the LED system is designed to op-
erate at a static-operating point. This means that for a given
luminous flux output, the operating point with a lower-LED
power should be chosen. Within this recommended power
range, either (7) or (14) can provide sufficiently accurate
predictions.
Rule 2: If the thermal design is restricted by limited space
for the heatsink so that the Pd∗ occurs at a power less than or
equal to the rated power Pd ( r a t e d ) , then the LED system should
be operated at Pd∗ for each LED device. [For example, points A
and B are optimal operating points for the respective curves as
their Pd∗ values do not exceed Pd(rated) .]
Rule 3: If the thermal design is flexible, then the LED system
should be designed in such a way that 1) the theoretical maxi-
mum φv point (or Pd∗ ) occurs at a power higher than Pd(rated) of
the LED and 2) the intersection point of the theoretical φv –Pd
curve and the rated power line should have a value of about 80%–
96% of the theoretical maximum φv value. The rated power
should be chosen as the operating power for each LED.
∆φv |pu = 0.04 ∼ 0.20 of the maximum φv point in the
curve.
This 4%–20% range for ∆φv from the maximum φv point
offers a good compromise of the light output and the size and
thus cost of the heatsink.
If forced cooling is applied, the φv − Pd curve will change
Fig. 11. (a) Calculated and measured total luminous flux versus lamp power dynamically. This can be visualized as having a dynamically
for two CREE 3W LEDs mounted on heatsink with thermal resistance of changing thermal resistance Rhs . The optimal operating point
4.5 ◦ C/W. (b) Calculated and measured total luminous efficacy versus lamp should follow the three rules explained previously. It should be
power for two CREE 3W LEDs mounted on a heatsink with thermal resistance
of 4.5 ◦ C/W. kept along the operating lines as highlighted in the red bold solid
lines in Fig. 15 in order to maximize the luminous flux output.
The general theory presented here points to useful information
for the LED device and system structures that may improve the
luminous performance. Equations (10) and (7) or (14) suggest
large positive slope means that a relatively small increase that luminous output will increase if Rjc and Rhs can be reduced.
of Pd can result in a relatively large increase of φv . So the These implications mean that
initial linear portion of the curve results in good efficacy. 1) On the device level, using multichip LED structures with
As Pd is moved to the region at and around Pd∗ , the slope more relatively low-power LED chips, and therefore larger
is zero or relatively small. Therefore, a relatively large contact area for heat transfer (with an overall lower Rjc )
increase in Pd will give a relatively small increase in φv . may be better than using single-chip high-power LED
2) The LED power Pd must not exceed the rated LED power structures of the same power rating. The trend for further
Pd(rated) . Otherwise, the lifetime of the LED will be short- increasing the device power in a single-chip LED device
ened. Therefore, the intersection points of these curves may not be appropriate, unless the heat dissipation prob-
with the rated power limit should indicate how the light lem can be reduced or solved.
output can be maximized. The intersection points of these 2) On the system level, using a higher number of LED devices
curves and the rated power line are denoted as “a”, “b”, of lower power may be better than using less LED devices
“c,” and “d” as shown in Fig. 14. of higher power for the same system power.
3) The values of φv at “c” and “d” are higher than that at “b.” These structural issues and experimental confirmation will
But the curve with peak φv at D requires a much larger be presented in the near future. With continuous improvements
heatsink than that with peak φv at C. The difference of φv in LED technology [26], it is envisaged that LED will replace
at “c” and “d” may not be significant enough to justify an mercury-based linear and compact fluorescent lamps in some
increase in cost and size of the heatsink. general illumination applications.
HUI AND QIN: GENERAL PHOTO-ELECTRO-THERMAL THEORY FOR LIGHT EMITTING DIODE (LED) SYSTEMS 1975

Fig. 12. Luminous flux curves with different thermal resistance R h s .

Fig. 13. Effects of cooling effects on the luminous flux curves. Fig. 15. Optimal operating point highlighted in bold solid lines.

V. CONCLUSION
A general theory that links the luminous, electrical, and ther-
mal aspects of an LED system has been developed. This theory
can predict the optimal operating point at which maximum lu-
minous flux can be generated for a given thermal design. It can
also be used to set criteria for the optimal thermal design for
the appropriate heatsink for a given application. Based on this
theory, one can explain why the maximum luminous flux may
not occur at the rated power of the LED system. It is found that
the maximum luminous flux will occur at an LED power level
that is dependent on the thermal resistance of the heatsink (or
more generally, the cooling effect of the thermal design). For a
small heatsink with high thermal resistance, the maximum flux
tends to occur at a lower-power level. This theory also provides
insights into the effects of various parameters such as the vari-
ations of the temperature-dependent junction-to-case thermal
resistance, the heat dissipation factor, and the efficacy degrada-
Fig. 14. Proposed light maximization design method. tion rate of the LEDs. It is found that the temperature-dependent
1976 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 24, NO. 8, AUGUST 2009

junction–to-case thermal resistance will cause the asymmetry of [19] Q. Cheng, “Thermal management of high-power white LED package,” in
the luminous flux function and affect adversely the degradation Proc. 8th Int. Conf. Electron. Packag. Technol. (ICEPT2007), Shanghai,
China, Aug., pp. 1–5.
of efficacy. The general theory has been verified by experimental [20] B. Ackermann, V. Schulz, C. Martiny, A. Hilgers, and X. Zhu, “Control
results. In spite of using a simple thermal model in this theory, of LEDs,” in Proc. IEEE IAS 2006, Tampa, FL, Oct., pp. 2608–2615.
good agreements between theoretical predictions and practical [21] M. Nishikawa, Y. Ishizuka, H. Matsuo, and K. Shigematsu, “An LED
drive circuit with constant-output-current control and constant-luminance
measurements have been achieved. control,” in Proc. Telecommun. Energy Conf. (INTELEC 2006), Province,
RI, Sep., pp. 1–6.
ACKNOWLEDGMENT [22] P. Baureis, “Compact modeling of electrical, thermal and optical LED
behavior,” in Proc. 35th Eur. Solid-State Device Res. Conf., Grenoble,
The authors would like to thank the Centre for Power Elec- France, Sep. 2005, pp. 145–148.
tronics of the City University of Hong Kong for its support in this [23] Y. X. Qin, D. Y. Lin, and S. Y. R. Hui, “A Simple method for comparative
project. They also thank Mr. W. H. Mak for his help in setting study on the thermal performance of light emitting diodes (LED) and
fluorescent lamps,” IEEE APEC 2009 and IEEE Trans. Power Electron.,
up the testing facilities and Mr. Deyan Lin for the constructive pp. 1811–1818, Jul. 2009.
discussions. [24] S. Y. R. Hui, “Methods for optimal operation of light-emitting diodes,”
U.S. Patent 12/370,101, Feb. 2009.
REFERENCES [25] E. F. Schubert, Light-Emitting Diodes, 2nd ed. Cambridge, U.K.: Cam-
bridge Univ. Press, 2006, ch. 5.
[1] R. Simpson, Lighting Control: Technology and Applications. Boston, [26] J. Xu, M. F. Schubert, A. N. Noemaun, D. Zhu, and J. K. Kim, “Reduction
MA: Focal Press, 2003. in efficiency droop, forward voltage, ideality factor, and wavelength shift
[2] J. M. Zhou and W. Yan, “Experimental investigation on the performance in polarization-matched GalnN/GalnN multi-quantum-well light-emitting
characteristics of white LEDs used in illumination application,” in Proc. diodes,” Amer. Phys. Lett., vol. 94, pp. 01113-1–01113-3, 2009.
PESC2007, Orlando, FL, Jun., pp. 1436–1440.
[3] J. H. Cheng, C. K. Liu, Y. L. Chao, and R. M. Tain, “Cooling performance
of silicon-based thermoelectric device on high power LED,” in Proc. 24th
Int. Conf. Thermoelectrics, Clemson, SC, Jun. 2005, pp. 53–56.
[4] T. Zahner, “Thermal management and thermal resistance of high power S. Y. (Ron) Hui (F’03) was born in Hong Kong, in
LEDs,” in Proc. 13th Int. Workshop Thermal Investigation ICs Syst. 1961. He received the B.Sc degree (with honors) from
(THERMINIC 2007, Budapest, Hungary, Sep., pp. 195–195. the University of Birmingham, Birmingham, U.K., in
[5] Luxeon Power Light Source, Data Sheet DS51. Available: http:// 1984, and the D.I.C. and Ph.D degrees from the Im-
www.philipslumileds.com/pdfs/DS51.pdf perial College of Science and Technology, University
[6] S. Buso, G. Spiazzi, M. Meneghini, and G. Meneghesso, “Performance of London, London, U.K., in 1987.
degradation of high-brightness light emitting diodes under DC and pulsed From 1987 to 1990, he was a Lecturer in
bias,” IEEE Trans. Device Mater. Rel., vol. 8, no. 2, pp. 312–322, Jun. power electronics at the University of Nottingham,
2008. Nottingham, U.K. In 1990, he went to Australia and
[7] J. Garcia, D. G. Lamar, M. A. Costa, J. M. Alonso, and M. R. Secades, took up a lectureship at the University of Technology,
“An estimator of luminous flux for enhanced control of high brightness Sydney, where he became a Senior Lecturer in 1991.
LEDs,” in Proc. IEEE PESC 2008, Rhodes, Greece, Jun., pp. 1852–1856. He joined the University of Sydney in 1993 and was promoted to Reader of
[8] C. Biber, “LED light emission as a function of thermal conditions,” in electrical engineering in 1996. He is currently a Chair Professor of electronic
Proc. IEEE Semicond. Thermal Meas. Manag. Symp., San Jose, CA, Mar. engineering at the City University of Hong Kong (CityU), Kowloon, China.
2008, pp. 180–184. From 1999 to 2004, he was an Associate Dean of the Faculty of Science and
[9] L. Trevisanello, M. Meneghini, G. Mura, M. Vanzi, M. Pavesi, Engineering at CityU. He is the author or coauthor of more than 200 technical
G. Meneghesso, and E. Zanoni, “Accelerated life test of high brightness papers, including more than 120 refereed journal publications. He also holds of
light emitting diodes,” IEEE Trans. Device Mater. Rel., vol. 8, no. 2, more than 40 patents.
pp. 304–311, Jun. 2008. Dr. Hui received the Teaching Excellence Award in 1999 and the Grand
[10] J. Bielecki, A. S. Jwania, E. Khatib, and T. Poorman, “Thermal consider- Applied Research Excellence Award in 2001 from the City University of Hong
ations for LED components in an automotive lamp,” in Proc. Twenty Kong, and the Best Paper Award from the IEEE IAS Committee on Production
Third Annu. IEEE Semicond. Thermal Meas. Manag. Symp. (SEMI- and Applications of Light in 2002. He is a Fellow of the IEE and has been an
THERM2007), San Jose, CA, Mar., pp. 37–43. Associate Editor of the IEEE TRANSACTIONS ON POWER ELECTRONICS since
[11] J. F. Van, D. Michele, and M. Colgan, “White LED sources for vehicle 1997. Since 2007, he has been an Associate Editor of the IEEE TRANSACTIONS
forward lighting,” in Proc. SPIE 2002, vol. 4776, pp. 195–205. ON INDUSTRIAL ELECTRONICS. Since October 2002, he has been an At-Large
[12] J. Petroski, “Thermal challengs facing new generation LEDs for lighting member of the IEEE PELS AdCom. He has been appointed as an IEEE Distin-
applications,” Proc. SPIE, vol. 4776, pp. 215–222, 2003. guished Lecturer by the IEEE PELS for 2004–2007.
[13] J. Lalith, Y. M. Gu, and N. Nadarajah, “Characterization of thermal resis-
tance coefficient of high-power LEDs,” in Proc. 6th Int. Conf. Solid State
Lighting, San Diego, CA, Aug. 2006, pp. 63370–63377.
[14] L. Yuan, S. Liu, M. X. Chen, and X. B. Luo, “Thermal analysis of high
power LED array packaging with microchannel cooler,” in Proc. 7th Int.
Conf. Electron. Packag. Technol. (ICEPT2006), Shanghai, China, Aug.,
pp. 1–5.
[15] J. Petroski, “Spacing of high-brightness LEDs on metal substrate PCB’s
for proper thermal performance,” in Proc. Ninth Intersoc. Conf. Thermal Y. X. Qin (S’08) was born in China, in 1985. He re-
Thermomech. Phenom. Electron. Syst. (ITHERM 2004), Las Vegas, NV, ceived the B.S. degree in electrical engineering from
Jun., pp. 507–514. Shandong University, Jinan, China, in 2007, and he is
[16] Z. L. Ma, X. R. Zheng, W. J. Liu, X. W. Lin, and W. L. Deng, “Fast currently working toward the Master of Philosophy
thermal resistance measurement of high brightness LED,” in Proc. 6th degree at the Department of Electronic Engineer-
Int. Conf. Electron. Packag. Technol. (ICEPT2005), Shenzhen, China, ing, City University of Hong Kong, Kowloon, Hong
Aug., pp. 614–616. Kong, China.
[17] B. Siegal, “Practical considerations in high power LED junction temper- His current research interests include the design
ature measurements,” in Proc. 31st Int. Conf. Electron. Manuf. Technol. and development of switching-mode power supplies,
(IEMT 2006), Kuala Lumpur, Malaysia, Nov., pp. 62–66. LED driving circuits, and electronic ballasts.
[18] M. Arik, C. Becker, S. Weaver, and J. Petroski, “Thermal management of Mr. Qin has served as a Reviewer for the IEEE
LEDs: Package to system,” Proc. SPIE, vol. 5187, pp. 64–75, 2004. TRANSACTIONS ON INDUSTRY ELECTRONICS and several conferences.

Potrebbero piacerti anche