solid-state lighting: failure analysis of white leds

ARTICLE IN PRESS

Journal of Crystal Growth 268 (2004) 449–456

*Corresp

518-687-71

E-mail

0022-0248/

doi:10.101

Solid-state lighting: failure analysis of white LEDs

N. Narendran*, Y. Gu, J.P. Freyssinier, H. Yu, L. Deng

Lighting Research Center, Rensselaer Polytechnic Institute, 21 Union Street, Troy, NY 12180, USA

Abstract

Long life, on the order of 50,000–100,000 h, is one of the key features of light-emitting diodes (LEDs) that has

attracted the lighting community to this technology. White LEDs have yet to demonstrate this capability. The goal of

the study described in this manuscript was to understand what affects the long-term performance of white LEDs.

Different types of LEDs have different degradation mechanisms. As a starting point, this study considered a commonly

available commercial package, the 5 mm epoxy-encapsulated phosphor-converted (YAG:Ce) white LED. Based on past

studies, it was hypothesized that junction heat and the amount of short-wavelength emission would influence the

degradation rate of 5 mm type white LEDs, mainly due to yellowing of the epoxy encapsulant. Two groups of white

LEDs were life-tested. The LEDs in one group had similar junction temperatures but different amplitudes for the short-

wavelength radiation, and the LEDs in the second group had similar amplitudes for the short-wavelength radiation but

different junction temperatures. Experimental results showed that the degradation rate depends on both the junction

temperature and the amplitude of short-wavelength radiation. However, the temperature effect was much greater than

the short-wavelength amplitude effect. Furthermore, the phosphor medium surrounding the die behaves like a

lambertian scatterer. As a result, some portion of the light circulates between the phosphor layer and the reflector cup,

potentially increasing the epoxy-yellowing issue. To validate this theory, a second experiment was conducted with LEDs

that had the phosphor layer both close to the die and further away. The results showed that the LEDs with the

phosphor layer away from the die degraded at a slower rate.

r 2004 Elsevier B.V. All rights reserved.

PACS: 85.60.Jb; 85.60.�q; 85.60.Bt

Keywords: A1. Degradation; A1. Reliability; B1. Nitrides; B2. Semiconducting materials; B3. Light emitting diodes; B3. White LEDs

1. Introduction

Since the first demonstration of the high-bright-ness blue light-emitting diode (LED) and subse-

onding author. Tel.: +1-518-687-7176; fax: +1-

20.

address: [email protected] (N. Narendran).

$ - see front matter r 2004 Elsevier B.V. All rights reserve

6/j.jcrysgro.2004.04.071

quently the white LED, the interest in using solid-state light sources for general illumination hasbeen rapidly growing [1,2]. Presently, manymanufacturers around the world are producingwhite LEDs. One of the most common methodsfor producing white light with LEDs is to use acerium-doped yttrium aluminum garnet (YAG:Ce)phosphor with a gallium nitride (GaN)-based blue

d.

ARTICLE IN PRESS

N. Narendran et al. / Journal of Crystal Growth 268 (2004) 449–456450

LED. Typically, the phosphor is embedded insidean epoxy resin that surrounds the LED die. Someportion of the short-wavelength radiation emittedby the GaN LED is down-converted by thephosphor, and the combined light is perceived aswhite by the human visual system.

Long life, on the order of 50,000–100,000 h, isone of the key features of LEDs that has attractedthe lighting community to this technology. WhiteLEDs have yet to demonstrate this capability.Early devices that took the form of indicator-stylepackages, 5 mm and surface mount devices(SMD), degraded very rapidly and reached a50% light output level within 10,000 h, even atdrive currents recommended by their manufac-turers [3,4]. However, some of the recent whiteLED packages that were designed specificallyfor illumination applications exhibit slower degra-dation rates than the indicator devices [4–8]. Thegoal of the study presented here is to understandwhat affects the long-term performance of whiteLEDs.

LEDs rarely fail catastrophically; instead, theirlight output degrades slowly over time. The causeof slow light output degradation was investigatedin this study. Different types of LEDs havedifferent degradation mechanisms, and thus dif-ferent light output degradation rates. As a startingpoint, this study considered the 5 mm epoxy-encapsulated phosphor-converted (YAG:Ce)white LED. Although many studies have investi-gated the degradation of InGaN-type LEDs, veryfew studies have analyzed the degradation ofpackaged white LEDs [9–12]. According to thefew that have studied the packaged devices (blueand white LEDs), the primary reason cited for therapid degradation of light output is the yellowingof the epoxy encapsulant [3,13–15]. Excessive heatat the p–n junction is pointed to as the main causeof epoxy yellowing [13,14]. Ambient temperatureand the ohmic heating at the bandgap add up tocreate the heat at the LED junction.

Polymeric materials and epoxies also yellowover time due to photodegradation. This topic hasbeen studied extensively by many polymer che-mists. Although ultraviolet (UV) radiation withwavelengths less than 300 nm has the strongesteffect on the yellowing of polymer materials in

photodegradation, it has been shown that evenvisible light in the range of 400–500 nm inducesphotodegradation [16]. The amount of photode-gradation depends on the amount of radiation andtime of exposure. Therefore, even visible light ofsufficient quantity can cause polymer and epoxymaterials to degrade [17].

Another observation made from past studies isthat the 5 mm type phosphor-converted whiteLED degrades faster than the similar type of blueLED [4]. If heat and the amount of short-wavelength radiation were the only reasons forthe yellowing of the epoxy, then the blue LEDshould degrade faster than the white LED becausethe total amount of short-wavelength radiationwould be much higher for the blue LED comparedwith the white LED at the same drive current. Themain difference between a 5 mm type blue and asimilar white LED is the phosphor that is mixed inwith the epoxy. Therefore, the presence ofphosphor could be a reason for the excessdegradation. An obvious question then is: Is thephosphor degrading? If the YAG:Ce phosphorwere degrading, then the energy emitted by thephosphor should decrease with respect to theexciting short-wavelength emission. This in turnwould result in a chromaticity shift toward blue asa function of time. However, literature shows thatfor 5 mm white LEDs, the chromaticity of thewhite light as a function of time shifts towardyellow rather than blue [3]. Furthermore, YAG:Cephosphors are considered very stable and do notdegrade easily [7]. Therefore, phosphor degrada-tion is an unlikely scenario. A possible explanationfor the additional degradation of white LEDs isthat the phosphor particles mixed into the epoxyemit light isotropically and also scatter theunconverted short-wavelength light, as illustratedin Fig. 1. Therefore, at any given time only afraction of the light will travel outward from thephosphor layer; the balance is redirected towardthe phosphor layer, the reflector cup, or the die.Since the radiant energy travels through the epoxyregion of the white LED more often than in theblue LED, the epoxy would yellow more.

Based on past studies, the authors of thismanuscript hypothesized that the light outputdegradation of white LEDs caused by the

ARTICLE IN PRESS

N. Narendran et al. / Journal of Crystal Growth 268 (2004) 449–456 451

yellowing of the epoxy depends on three factors:the heat at the p–n junction, the amount of short-wavelength radiation, and the location of thephosphor layer. By placing the phosphor layer farfrom the die, light is more likely to escape thedevice rather than being trapped where it willcause excessive yellowing of the epoxy.

Fig. 1. Schematic of a 5 mm LED.

Fig. 2. (a) Photo of the life-test setup

2. Experiment

Two laboratory experiments were conducted toverify the above-mentioned hypotheses. The firstexperiment investigated the effects of junctiontemperature and short-wavelength emission onthe degradation rate of white LEDs, and thesecond experiment studied the effect of placing thephosphor layer away from the die and the epoxysurrounding it.

2.1. Experiment 1

The goal of the first experiment was toinvestigate the relative effects of junction tempera-ture and short-wavelength emission on the degra-dation rate of white LEDs. Two groups of whiteLEDs were life-tested. The LEDs in one group hadsimilar junction temperatures but produced differ-ent amounts of short-wavelength radiation, andthe LEDs in the second group produced similaramounts of short-wavelength radiation but differ-ent junction temperatures.

The LEDs were tested in specially designed life-test chambers that had two functions: one, to keepthe ambient temperature constant, and two, to actas light-integrating boxes to enable light outputmeasurements [5]. Fig. 2 illustrates the experimental

and (b) schematic of the setup.

ARTICLE IN PRESS


setup. Several life-test chambers were built andplaced inside a temperature-controlled room.

The chambers measured 23 cm on each side. Theinterior of the chamber was painted matte white. A2.5 cm diameter round aperture was cut into thefront panel of each chamber and was fitted with atransparent acrylic window to enable spectralmeasurements. These apertures were covered withblack fabric during the life-test to keep externallight from entering the box. Each chambercontained one LED array mounted at the centerof the inside top surface. A photodiode thatcontinuously measured the light output wasattached to the center of the left panel, whichhad a small white baffle that blocked the directlight so that only the reflected light reached thephotodiode. A resistance temperature detector(RTD) placed on top of the baffle, which was alsolocated at the midpoint in the vertical direction,measured the ambient temperature within thechamber and controlled the heater through thetemperature controller (T-CTL). This ensured thatthe temperature inside the box remained within71�C. The heater was attached to a raisedaluminum plate with a matte-white cover thatsat on the bottom of the chamber. To estimatethe LED junction temperature, a thin wirethermocouple (TC, J-type) was soldered to thecathode pin of one LED in the array. Junc-tion temperature of the LED was estimated fromthe cathode pin temperature, the power dissipatedat the p–n junction, and the thermal resistance

Table 1

Life-test conditions for each LED array

Group # Array # Ambient temperature

(�C)

Drive

(mA)

Group 1 1 35 59

2 45 43

3 55 35

4 65 28

Group 2 5 35 31

6 40 36

7 45 36

2 45 43

8 65 39

9 70 41

coefficient of the white LEDs [18]. For eachchamber, an external LED driver controlledthe current flow through the LEDs. A halogenlight source placed inside the top left of eachchamber was used to calibrate the chamber atregular intervals throughout the life-test to ensurethat the light output degradation was only fromthe LEDs and not due to degrading paint insidethe box. Outside the chambers, two sensorsmeasured the ambient temperature of the labora-tory. A multimeter measured the input current andvoltage of each LED. A computer-controlled dataacquisition system gathered six channels of in-formation (light output, board temperature, vol-tage, current, power, and power factor) every hourfor each box.

Fifty-four white LEDs (Nichia NLPB-500) wereselected from the same batch, acquired in April2003. Six LEDs with the same peak wavelength forthe short-wavelength emission (around 462 nm)were connected in series to form an array.Altogether, nine LED arrays were constructedand each had its own life-test chamber. Theoperating conditions of the nine arrays are shownin Table 1.

As mentioned earlier, the arrays were dividedinto two groups. Group 1 (Arrays 1–4) had similarjunction temperatures (approximately 95�C) butproduced different short-wavelength amplitudes(ranging from 89% to 124% of the measuredemission amplitude at the standard condition,20 mA, 25�C ambient temperature). Group 2

current Relative amplitude of

short-wavelength emission

Junction

temperature (�C)

1.24 95.9

1.14 96.0

0.99 94.8

0.89 94.7

1.09 68.7

1.14 78.6

1.15 84.5

1.14 96.0

1.15 106.5

1.13 114.5

ARTICLE IN PRESS

0.000000

0.000200

0.000400

0.000600

0.000800

0.001000

0.80 0.90 1.00 1.10 1.20 1.30 1.40

Dec

ay c

onst

ant

Relative amplitude ofshort-wavelength radiation

Fig. 4. Decay constant a as a function of short-wavelength

amplitude for the LED arrays in Group 1.


(Arrays 2 and 5–9) produced similar short-wavelength amplitudes (approximately 114% ofthe emission amplitude at the standard condition),but had different junction temperatures (rangingfrom 69�C to 115�C). The operating conditions ofthe different arrays are summarized in Table 1.These conditions were achieved by adjusting theLED drive current and the ambient temperatureinside the chambers.

Usually, LEDs undergo rapid changes during aninitial period of several hundred hours, and thensettle into a more systematic degradation pattern.Therefore, the LED arrays were initially seasonedfor 400 h at their respective operating conditionsprior to collecting data. The junction temperaturesof each array remained fairly constant over timewith small fluctuations (no more than 2�C),mainly caused by the fluctuations of the roomtemperature and the ambient temperature in thethermal chambers.

2.1.1. Results of Experiment 1

Fig. 3 shows the relative light output as afunction of time for the LED arrays in Group 1,which had similar junction temperatures butdifferent short-wavelength amplitudes. For eacharray, the relative light output over time wasnormalized to its initial value. The horizontal axisis on a log scale. The light output decrease overtime is exponential in nature and therefore, thelight output, L; can be expressed as:

L ¼ e�at; ð1Þ

where a is the light output degradation rate or thedecay constant, and t is the operation timemeasured in hours.

0.400

0.600

0.800

1.000

10 100 1000 10000

Time (hrs)

Rel

ativ

e L

ight

Out

put

A = 1.24

A = 1.14

A = 0.99

A = 0.89

Fig. 3. Light output variation as a function time for the LED

arrays in Group 1.

Exponential curve fits yielded different decayconstants for each LED array. Fig. 4 illustrates thevariation of the decay constant as a function ofshort-wavelength amplitude of the LED emission.A higher decay constant means a faster degrada-tion rate. As the short-wavelength amplitudeincreased, the decay constant also showed anincreasing trend (Fig. 4). Therefore, as hypothe-sized, the light output degradation of 5 mm typewhite LEDs exhibits an increasing trend withincreased short-wavelength radiation.

Fig. 5 shows the relative light output as afunction of time for the LED arrays of Group 2,which had similar short-wavelength amplitudesbut different junction temperatures. Here, too, thelight output decrease over time is exponential innature. As before, exponential curve fits yieldedthe decay constants for each LED array. Fig. 6illustrates the variation of the decay constant as afunction of junction temperature. With increasingjunction temperature, the decay constant increasedexponentially (Fig. 6). As hypothesized, the lightoutput degradation rate of 5 mm type white LEDsincreased with increased junction temperatures.

2.2. Experiment 2

The goal of the second experiment was toinvestigate the impact of placing the phosphoraway from the die and the epoxy. Three LEDarrays, each containing 16 5 mm LEDs, werehoused in separate aluminum cylinders, each witha 5 cm diameter, as shown in Fig. 7. Two of thearrays were assembled using blue LEDs, and onewas assembled with white LEDs. All of the LEDswere purchased in June 2003 from a single

ARTICLE IN PRESS

0.400

0.600

0.800

1.000

10 100 1000Time (hrs)

Rel

ativ

eL

ight

Out

put

10000

Tj = 69˚C

Tj = 79˚C

Tj = 85˚C

Tj = 96˚C

Tj = 107˚C

Tj = 115˚C

Fig. 5. Light output variation as a function time for the LED arrays in Group 2.

0.00000

0.00020

0.00040

0.00060

0.00080

0.00100

60 70 80 90 100 110 120

Junction Temperature (deg C)

Dec

ay c

onst

ant

Fig. 6. Decay constant a as a function of junction temperature

for the LED arrays in Group 2.


manufacturer (white: NSPW510BS; blue:NSPB510S). The inside surfaces of the cylinderswere painted white. Two cylinders, one with a blueLED array and another with a white LED array,were fitted with a 5 cm round transparent glasswindow. The third cylinder with a blue array wasfitted with a similar glass window coated with aYAG:Ce phosphor on one side. This glass windowwas installed with the phosphor side facing theLED array. The light source assemblies weremounted vertically facing downward. The LEDarrays were housed inside three chambers, similarto the ones used earlier, and were life-tested.

The experiment was conducted twice, once witha set of LED arrays operating at 60 mA and thenwith a second set of LED arrays operating at40 mA. All LEDs were seasoned for a few hundredhours at their respective operating currents priorto collecting data. Light output data were recordedtwice each day. Data for the 60 mA drive currentwere recorded for approximately 650 h. Then theexperiment was repeated with the second set of

LED arrays operating at a 40 mA constantcurrent.

2.2.1. Results of Experiment 2

Fig. 8 illustrates the relative light output of theLED arrays as a function of time. As hypothe-sized, the white LED array had the highestdepreciation rate while the blue and blue-plus-phosphor arrays had lower depreciation rates. Asit can be seen in the plot, placing the phosphorlayer close to the die causes the LEDs to degradefaster.

Fig. 8 shows that the blue-plus-phosphor LEDsdegraded at a rate slightly higher than the blueLEDs at 60 mA. This may be due to the fact thatthe phosphor layer was not hermetically sealedand moisture could have degraded the phosphor.At 40 mA, both degraded at the same rate for awhile, and then the blue-plus-phosphor LEDsstarted to slow down with respect to the blue.When the pin temperatures of the two LED arrayswere examined, they showed a few degrees changein temperature during this period. Therefore, thedeviation is possibly due to the fact that the lightoutput change as a function of temperature isdifferent for the two LED arrays.

3. Discussion

Although junction heat and amplitude of short-wavelength radiation influenced the yellowing ofthe epoxy, and hence, the light output degradationrate of these types of white LEDs, the junction

ARTICLE IN PRESS

Fig. 7. Photo of an LED array enclosed in an aluminum cylinder, and the schematics of the three LED arrays used in Experiment 2.

0.5

0.6

0.7

0.8

0.9

1.0

0 200 400 600 800 1000 1200Time (hours)

Rel

ativ

elig

htou

tput

Blue @ 40 mA Blue+ph @ 40 mA White @ 40 mA

Blue @ 60 mA Blue+ph @ 60 mA White @ 60 mA

Fig. 8. Lumen depreciation data for the three LED arrays at 40

and 60 mA.


temperature had a much greater effect than theshort-wavelength amplitude. One aspect thatneeds further investigation is the interactionbetween junction temperature and the amount ofshort-wavelength radiation on the degradation ofthese types of white LEDs. The authors of thismanuscript are presently studying this issue, andthe results will be published at a later time.

Results obtained thus far show that the lumenmaintenance of white LEDs can be improved byusing epoxy materials that have lower photode-gradation characteristics by extracting the heatmore efficiently from the die to keep it cooler, andby not placing the phosphor layer very close to thedie. Some white LEDs in the marketplace haveexploited the first two facts and have showndramatic improvements in lumen maintenance[5,6]. It should be noted that there are othercomponents that can also cause light outputreduction, including degradation of the dieattach!e epoxy and discoloration of the metalreflector due to thermal effects. Once the photo-degradation phenomenon is addressed, the degra-dation of the semiconducting element will become

the primary cause for long-term lumen deprecia-tion. By that time, the useful life of white LEDswill have significantly improved, and the whiteLED will be closer to delivering its long-lifepromise.

Acknowledgements

The authors would like to thank the USDepartment of Energy and the University ofCalifornia at Santa Barbara for their support inthe form of a grant (DOE Grant DE-FC26-01NT41203). Dr. Madis Raukas of OSRAMSYLVANIA is thanked for supplying the glassplates with YAG:Ce phosphor coatings. RichardPysar of the LRC is thanked for designing andimplementing the LED life-test apparatus. JenniferTaylor of the LRC is thanked for helping us toprepare this manuscript.

References

[1] S. Nakamura, T. Mukai, M. Senoh, Appl. Phys. Lett. 64

(1994) 1687.

[2] S. Nakamura, G. Fasol, The Blue Laser Diode, GaN

Based Light Emitters and Lasers, Springer, Berlin, 1997.

[3] N. Narendran, J. Bullough, N. Maliyagoda, A. Bierman,

J. Illum, Eng. Soc. 30 (1) (2001) 57.

[4] N. Narendran, L. Deng, Proceedings of the IESNA

Annual Conference, 2002, p. 157.

[5] N. Narendran, L. Deng, R.M. Pysar, Y. Gu, H. Yu, Proc.

SPIE 5187 (2004) 267.

[6] D.A. Steigerwald, J.C. Bhat, D. Collins, R.M. Fletcher,

M.O. Holcomb, M.J. Ludowise, P.S. Martin, S. Rudaz,

IEEE J. Sel. Top. Quantum Electron. 8 (2) (2002) 310.

[7] R. Mueller-Mach, G.O. Mueller, M.R. Krames, T.

Trottier, IEEE J. Sel. Top. Quantum Electron. 8 (2)

(2002) 339.

ARTICLE IN PRESS


[8] M.R. Krames, et al., Phys. Stat. Sol. (a) 192 (2) (2002) 237.

[9] T. Egawa, H. Ishikawa, T. Jimbo, M. Umeno, Appl. Phys.

Lett. 69 (6) (1996) 830.

[10] L. Sugiura, J. Appl. Phys. 81 (4) (1997) 1633.

[11] L. Sugiura, J. Appl. Phys. 70 (10) (1997) 1317.

[12] M. Fukuda, Reliability and Degradation of Semiconduc-

tor Lasers and LEDs, Artech House, Boston, 1991.

[13] D.L. Barton, M. Osinski, P. Perlin, C.J. Helms, N.H. Berg,

Proc. SPIE 3279 (1998) 17.

[14] D. Barton, M. Osinski, IEEE LEOS 2 (1998) 440.

[15] F.M. Steranka, et al., Phys. Stat. Sol. (a) 194 (2) (2002)

380.

[16] J. Akhavan, E. Koh, S. Waring, E. Kronfli, Polymer 42

(2001) 7711.

[17] A. Torikai, H. Hasegawa, Polym. Degradation Stability 63

(1999) 441.

[18] Y. Gu, N. Narendran, Proc. SPIE 5187 (2004) 107.

solid-state lighting: failure analysis of white leds

Documents