IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 49, NO. 10, OCTOBER 2002 1715

Design and Fabrication of Highly EfficientGaN-Based Light-Emitting Diodes

Hyunsoo Kim, Seong-Ju Park, and Hyunsang Hwang, Member, IEEE

Abstract—A promising fabrication method and an innovativegeometrical design for highly efficient GaN-based light-emittingdiodes (LEDs) were investigated based on current spreadingphenomenon. Based on theoretical considerations, it was possibleto determine the critical transparent-electrode thickness, whichresulted in significant improvements in the electrical and opticalcharacteristics of LEDs. In addition, we were able to define con-ditions for an ideal geometrical design and the resulting productexhibited significant improvements in electrical and optical char-acteristics in spite of the fact that a transparent electrode, actingas a p-type current spreader, was not used. Considering the simplefabrication process and high device performance, the proposedfabrication methods, as well as the innovative geometrical design,have considerable promise for use in practical applications.

Index Terms—Current spreading, GaN, geometrical design,light-emitting diode, model.

I. INTRODUCTION

GaN-BASED light-emitting diodes (LEDs) have been thesubject of extensive investigations, and have been devel-

oped because of their potential applications in areas such as full-color displays, full-color indicators, and high-efficiency lamps[1]–[8]. However, the GaN-based LED has a critical weaknessin that its fabrication mainly involves the use of the lateral carrierinjection type due to the absence of an appropriate conductingsubstrate. In this case, a current crowding problem is often en-countered and impedes the development of the efficient GaN-based LEDs. Several groups have reported on studies relatingto this issue, both theoretically and experimentally [9]–[11].Based on the assumption that the transparent electrode repre-sents a perfect current spreader, Eliashevichet al. reported thatthe conductivity of an n-type GaN layer has a profound effecton uniform current spreading [9]. It has been shown that the uni-form current spreading is essentially attained above the criticaln-type carrier concentration. However, a thin metal film has amuch higher sheet resistance than its corresponding bulk mate-rial due to the reflection of conduction-electrons from defectsthat are trapped in the film during the deposition and from in-ternal surfaces [12]. Therefore, the resistivity of the transparentelectrode should not be ignored in the development of a moreaccurate model. Based on this consideration, more highly de-veloped theoretical model has recently been proposed and im-portant parameters such as the current density, the resistivities

Manuscript received February 1, 2002. This work was supported in part bythe Korea Energy Management Cooperation and the Brain Korea 21 Project.The review of this paper was arranged by Editor P. Bhattacharya.

The authors are with the Department of Materials Science and Engineering,Kwangju Institute of Science and Technology, Kwangju, 500-712, Korea(e-mail: hwanghs@kjist.ac.kr).

Publisher Item Identifier 10.1109/TED.2002.802625.

of the transparent electrode and n-type layers, and the effectivelength for the lateral current path were found to be importantfactors in uniform current spreading [10], [11].

From the standpoint of both uniform current spreading andhigh extraction efficiency of a generated light, the determina-tion of the proper thickness of the transparent electrode becomesvery important. However, no systematic study on the transparentelectrode has yet been reported because of a lack of under-standing of the relationships between the transparent electrodeand the n-type layer with respect to current spreading. In thisstudy, we report on a method for determining the critical trans-parent-electrode thickness for the realization of highly efficientLEDs.

Based on the effective length factor, which involves devicegeometry, significant improvements in LED characteristics havealso been demonstrated by local modification of the p-type pad(electrode) geometry [10], [11]. In addition, although it does notmainly concern the current spreading problem, the geometricaldesign such as the interconnected microdisk LED also showeda 60% increase in optical emission efficiency compared to theconventional broad-area LED [13]. These results indicate theimportance of geometrical design on device efficiency. In thisregard, in terms of the ideal geometrical design, which givesperfectly uniform current spreading conditions, we report onan attempt to realize highly efficient GaN-based LEDs in theabsence of a transparent electrode.

II. EXPERIMENT

Metalorganic chemical vapor deposition was used to grow a1.5- m-thick n-GaN : Si layer on a (0001) sapphire substrate.This was followed by the growth of 0.05-m-thick InGaN/GaNmultiple quantum well (MQW) layers with five periods, fol-lowed by the deposition of a 0.25-m-thick p-GaN : Mg layer.The procedure for the growth and epilayer structure of the MQWLED have been described elsewhere [14]. In terms of fabricationof the LED device, the p-type layer was selectively etched to ex-pose the n-type layer using an inductively-coupled plasma (ICP)etching system. A Ni/Au transparent layer was then depositedon the surface of the p-GaN layer. This was followed by the de-position of a Ni/Au (30 nm/100 nm) layer in order to achieve ap-ohmic pad. For an n-ohmic pad, a Ti/Al (30 nm/100 nm) layerwas deposited on the n-GaN, and the metal-deposited sampleswere then annealed at 450C for 30 s in a rapid thermal an-nealing system. All electrical and optical properties of the LEDwere evaluated via on-wafer probing of the devices. The cur-rent–voltage (– ) characteristics were measured using a pa-rameter analyzer (HP 4155A). The light output power of theLED was measured using a UV/VIS 818 photodiode.

0018-9383/02$17.00 © 2002 IEEE

1716 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 49, NO. 10, OCTOBER 2002

III. RESULTS AND DISCUSSIONS

A. Critical-Transparent Electrode Thickness

Fig. 1(a) shows a three dimensional schematic view of theGaN-based LED, which was designed using earlier device mod-eling results [10], [11]. In previous modeling studies, it wasshown that perfectly uniform current spreading is possible whenthe following equation is satisfied

(1)

where is the current density and ( ) the resistivity of thetransparent electrode (n-type layer). The geometrical parame-ters , , and are also defined in Fig. 1(b), which shows atop view of the fabricated LED for the device model. In order toderive (1), the voltage drop was expressed as , assuminga constant current density for the series mode of the device[9]–[11]. However, for developing a more accurate model, thecurrent should be constant, since the cross-sectional area

is significantly different for each component. Therefore, thebasic equation for the voltage drop can be expressed as

(2)

where is the resistance. Based on this expression, the re-sulting equation for a perfectly uniform current spreading canbe rewritten as

(3)

where and represent the thickness of the transparent elec-trode and the n-type layer, respectively, andand are as-sumed to be equal , for the sake of simplifica-tion. Since is defined as the sheet resistance , (3) can bepresented as

(4)

(5)

or

(6)

Equation (5) indicates that perfectly uniform current spreadingis possible when the sheet resistances of the transparentand n-type layer are identical. Regarding this condition, itis possible to determine the critical transparent-electrode thick-ness for the intentionally designed LED of Fig. 1.

Fig. 2 shows the sheet resistance variation of the transparentelectrode as a function of film thickness . The thick-ness of the transparent electrode (Ni/Au) was controlled by athickness monitor in the e-beam evaporator and corrected viathe use of X-ray reflectivity measurements. For measurementof the sheet resistance, a Ni/Au film with a 1 : 1 thickness ratiowas evaporated onto the nonactivated 1010 mm p-type GaN

Fig. 1. (a) Schematic view of the GaN-based LED. (b) Top view of thefabricated LED with anl=w ratio of 1.50, as measured by optical microscope.This geometry was designed for a device model with respect to currentspreading.

Fig. 2. Sheet resistance of the transparent electrode(� ) as a function offilm thickness(t ). The inset shows the geometry used for measuring the lightoutput power of the LED.

sample and then annealed at 450C for 30 s in an N ambient.The sheet resistance was then measured using a four-point probesystem. In Fig. 2, it can be seen that the sheet resistance of thetransparent electrode significantly increases as the thick-ness is reduced. This is due to the enhanced reflection ofconduction-electrons caused by a relative increase in disloca-tion, defect, and internal surface scattering with decreasing filmthickness , which decreases the mean-free-path of electronsand the conductivity of the metal film [12]. Based on this exper-imental behavior and (5), it is possible to determine the criticaltransparent-electrode thickness for LED samples withan arbitrary sheet resistance of the n-type layer . To ensure

KIM et al.: DESIGN AND FABRICATION OF LIGHT-EMITTING DIODES 1717

TABLE ISHEET RESISTANCE(� ) AND THE LIGHT TRANSMITTANCE (T ) OF THE TRANSPARENTELECTRODE AS AFUNCTION OF FILM THICKNESS(t ). THE

LIGHT TRANSMITTANCE (T ) OF THETRANSPARENTELECTRODEWAS MEASURED AT 470 NM WAVELENGTH USING A UV-VIS SPECTROMETER. LED1AND 2 REPRESENT THEFABRICATED LEDS WITH A � VALUE OF 90 AND 55= , RESPECTIVELY. THE SERIESRESISTANCE(R ) WAS EXTRACTED VIA

LINEAR FIT FROM THE MEASUREDI–V CURVE. THE LIGHT OUTPUT POWER OFALL LEDs WAS MEASURED AT A PHOTODIODEMEASUREMENTANGLE OF90 AS

SHOWN IN THE INSET OFFig. 2. IN ADDITION, FOR AN ACCURATE COMPARISON OF THEMEASUREDOUTPUT POWER FORALL LEDs,THE POSITIONS FOR THE

LED AND THE PHOTODIODE WERE FIXED. THE INTEGRATED LIGHT OUTPUT POWER (L ) WAS OBTAINED FROM THE INTEGRATION OF THELIGHT

OUTPUT POWER (L) MEASURED AT CURRENTSRANGING FROM 0 TO 100mA

the relation (5), various thickness schemes of the transparentelectrode were adopted for the fabrication of two LED samples,which have values of about 90 (corresponding to acarrier concentration of 1 10 cm ) and 55 510 cm as shown in Table I. For LED samples withvalues of 90 and 55 , it is predicted that the theoretical crit-ical thicknesses of the transparent electrode are near 40and 60 Å, respectively.

Considering the optical efficiency of LEDs, the light transmit-tance of the transparent electrode must also be takeninto consideration because it is strongly dependent on the thick-ness . Table I shows the light transmittance ofthe transparent electrode at the wavelength of 470 nm as a func-tion of electrode thickness . For measurement of the lighttransmittance, the samples, which were prepared for measure-ment of the sheet resistance, were used. It is clear that the lighttransmittance begins to significantly degrade abovea thickness of 60 Å.

In Table I, it can be seen that the series resistancesof the90 -LED1s are heavily dependent on the transparent-elec-trode thickness . This is an essential proof that the sheet re-sistance of the transparent electrode (i.e., the transparent-elec-trode thickness) is a very important factor in a device perfor-mance. The series resistance was calculated from the diodecurrent of a pn junction. When the series resistance contributesto device behavior, the diode equation can be written as [15]

(7)

where is the prefactor, the measured voltage from the–curve, the series resistance, andthe ideality factor. Equa-tion (7) can be rewritten as

(8)

Therefore, according to (8), the series resistance can be ex-tracted via a linear fit from the measured current–voltage (– )curve. For an accurate determination, a calculation of the seriesresistance was made based on the high current injection (1

mA) in the – curves, which is dominated by the series resis-tance [16].

For LED1 with a value of 90 , the series resis-tance of LEDs was minimized at a thickness of 60 Å, in-dicating that the experimentally determined critical thickness ofthe transparent electrode for 90 -LED is 60 Å as a result ofthe largely optimized current injection. In addition, the highestlight output power was observed at a thickness of 60 Å, whichis in good agreement with the electrical characteristics of LED.However, since the light output efficiency is also strongly de-pendent on the light transmittance , the highest lightoutput power at a thickness of 60 Å, can be attributed to thecombined effects of the most uniform current spreading andthe proper light transmittance. Therefore, it is more desirable todetermine the experimental critical transparent-electrode thick-ness based on electrical characteristics. According to above re-sults, it should be noted that the experimental critical thickness

for 90 -LED1 was shifted by 20 Å compared to thetheoretical prediction (40 Å). This deviation will be discussedlater.

For LED2 with a value of 55 , the series resistancewas gradually decreased with increasing transparent-elec-

trode thickness, indicating that the experimental critical thick-ness is larger than 60 Å. On the other hand, the inte-grated light output power was highest at a thicknessof 60 Å, and was significantly degraded at the 120-Å thickness.This relates to the significant deterioration of the light transmit-tance of the transparent electrode. According to the electricalcharacteristics, it is evident that the experimental critical thick-ness of the transparent electrode was shifted towardgreater thickness from the theoretical prediction (60 Å). How-ever, considering the relatively small difference in series resis-tance between 60 and 120 Å-thickness as well as signifi-cant discrepancies between electrical and optical performance,it is difficult to conclude that the experimental critical thickness

for 55 -LED is 120 Å.The systematic discrepancy between the experimental and the

theoretical predictions can be explained in several ways. First,for simplification, it was assumed that the device widths for ann-type layer and a transparent electrode are equalin (4). However, the actual values of and are 220 and

1718 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 49, NO. 10, OCTOBER 2002

260 m, respectively. Therefore, for an accurate determinationof the critical thickness , (5) must be modified so as

(9)

which leads to a correction in sheet resistance by a factor of15%. In addition, for the basic modeling in terms of the uniformcurrent spreading, the lateral voltage drop through the p-typelayer was neglected, on the assumption that the elec-trical resistivity of the p-type layer is much larger than thatof the transparent electrode [10], [11]. Although this as-sumption is valid in terms of the practical thickness of the trans-parent electrode, it is certain that this assumption might con-tribute to the shift in the experimental critical thicknesstoward thicker ranges. Furthermore, differences in measurementtechniques should be also noted. As described previously, thesheet resistance of the transparent electrode was measured usinga four-point probe system. However, the sheet resistance of a5 5 mm n-type GaN layer was measured using the In/Zncontact from a Hall measurement system. Therefore, an absolutecomparison of the measured sheet resistance is difficult due tothe existence of the In/Zn contact resistance for an n-type layeras well as the differences in the measurement systems. Basedon the previous considerations, it can be concluded that the de-termination of the critical transparent-electrode thickness basedon (5) must be corrected by about 46%, which was roughly esti-mated from the experimental data on the 90 -LED. There-fore, for the 55 -LED, it can be estimated that a more ac-curate critical thickness of the transparent electrodewill be near 100 Å. This estimation is plausible consideringthat the deterioration of the light transmittance by17.4% from 60 to 120 Å-thickness does not sufficiently ex-plain the degradation of the integrated light output power of55 -LED by 85% with an increase in the transparent-elec-trode thickness from 60 to 120 Å.

It is noteworthy that the electrical and optical characteristicsof the 55 -LEDs were much superior to those of the90 -LEDs irrespective of the determined critical trans-parent-electrode thickness. This result suggests, for the highestdevice efficiency, first, the resistance of the n-type layer shouldbe minimized in order to maximize the carrier injection intothe n-type layer. The critical transparent-electrode thicknessshould then be determined according to the proposed method,resulting in the maximization of the carrier injection into thep-type layer through the transparent electrode.

B. Innovative Geometrical Device Design

Equation (6) predicts that perfectly uniform current spreadingis possible when the geometrical factor of ratio approacheszero. In order to confirm this condition, LED samples with var-ious ratios were prepared as shown in the inset of Fig. 3(a).

Fig. 3(a) shows the– characteristics of LEDs with variousratios. For a reasonable comparison, we attempted to min-

imize the effect of the material-related factor by adopting themost efficient LED system, for both aspects of the electrical andoptical performance, which has an n-layer sheet resistance of55 and a transparent-electrode thickness of 60 Å. Fig. 3(a)

Fig. 3. (a) Current density–voltage (J–V ) and (b) the differential quantumefficiency-current density (�–J) characteristics of LEDs with variousl=wratios. The inset of (a) shows a top view of the fabricated LED with twoextremel=w ratios, as measured by optical microscope. The inset of (b) showsthe geometry used for measuring the light output power of the LED. It shouldbe noted that an absolute comparison of the differential quantum efficiency(�)for each measurement angle of 0 and 90is impossible due to the limitation ofthe measurement system(x 6= y). It is only possible to compare the relativedifferential quantum efficiency of LEDs with variousl=w ratios under thesame measurement angle.

clearly shows that the electrical characteristics are gradually im-proved with decreasing ratios. In order to investigate therelation between electrical and optical performance, we plottedthe differential quantum efficiency from the measured lightoutput power as shown in Fig. 3(b), calculated using the fol-lowing equation:

number of emitted photonsnumber of injected electrons junction area

(10)

where is the measured light output power (in Watts),theinjection current (A), the photon energy (eV) radiated fromthe LED, and the junction area (or p-mesa area). Fig. 3(b)clearly shows that the differential quantum efficiency isgreatly improved with decreasing ratios. Interestingly, wewere able to observe more prominent improvements of thedifferential quantum efficiency with decreasing ratioswhen measured at a photodiode measurement angleof 0 .This can be attributed to the relative change of the dominantlight-emitting area from through the transparent electrode

KIM et al.: DESIGN AND FABRICATION OF LIGHT-EMITTING DIODES 1719

Fig. 4. Photographs of the fabricated (a) conventional LED with al=w ratioof 1.00, (b) newly designed LED(new1), and (c) another newly designed LED(new2).

(vertical direction) to the sidewalls (lateral direction) withdecreasing ratios. Based on these experimental behaviors,the highest device efficiency is predicted when the ratioapproaches zero. This fact is very attractive because it providesthe possibility of constructing highly efficient LEDs withoutthe need for a transparent electrode.

Fig. 4 shows the conventional [Fig. 4(a)] and the novel LEDs[Fig. 4(b) and (c)], designed based on this consideration

. It should be noted that the LED (c) (which we will call“new2”) has larger emitting area than the LED (b) (which wewill call “ new1”) by a factor of 13%. Also, it is noteworthythatnew1has slightly larger p-pad area thannew2. For an exactcomparison of the device performance between the conventionaland the newly designed LEDs, the same overall dimensions of350 260 m were adopted for all LEDs and the same waferwas used for LED fabrication.

Fig. 5 shows the– characteristics of the conventional andthe newly designed LEDs with a value of 90 . For thenewly designed LEDs, a clearly improved– characteristicwas observed compared to the conventional LED, which resultsin 50.6 and 26.8% reduction in series resistance of thediodes fornew1andnew2, respectively. This result is very sur-prising, considering the fact that the junction area of the newlydesigned LED is much smaller than the conventional LED byabout 30%. Therefore, it can be concluded that the newly de-signed LEDs are very effective in enhancing the current injec-tion efficiency. As expected, the– characteristics show moreclear difference between the conventional and the newly de-signed LED as shown in the inset of Fig. 5. However, it shouldbe noted that the conventional LED shows abnormal– curvewith a high series resistance, which is attributed to the high sheetresistance of the n-type layer. Therefore, it is important to ad-dress the issue of whether the improvement of the electricalcharacteristics for the newly designed LEDs is sample-depen-dent or not.

Fig. 6 shows the– characteristics of the conventional andthe newly designed LED with a value of 55 . It isnoteworthy that the newly designed LEDs also show improved

Fig. 5. I–V characteristics of the conventional and the newly designed LEDswith a � value of 90= . For the fabrication of the conventional LED,60 Å-thickness of the transparent electrode was adopted. For an n- and p-typeelectrode pad, a Ti/Al (30 nm/100 nm) and a Ni/Au (30 nm/100 nm) werecommonly adopted for the fabrication of all LEDs. The inset shows the currentdensity–voltage(J–V ) characteristics of the conventional and the newlydesigned LEDs.

Fig. 6. I–V and the currentJ–V characteristics (inset) of the conventionaland the newly designed LEDs with a� value of 55= . For the fabricationof the conventional LED, 60-Å thickness of the transparent electrodes wasadopted.

electrical characteristics compared to the conventional LED,which results in a 56.2% and 43.7% reduction in series resis-tance fornew1andnew2, respectively. According to these re-sults, we conclude that the proposed geometrical design holdsconsiderable promise in terms of electrical characteristics andthat sample dependence is not a significant factor. In addition,it should be also noted that the electrical characteristics ofnew1are superior to those ofnew2, which is indicative of a sensitivepattern dependence.

Fig. 7 shows the power–current (– ) characteristics of theconventional and newly designed LEDs with a value of90 . These data clearly show that the newly designed LEDsexhibit improved light output power for the same injectedcurrent , indicating that the newly designed LEDs are veryeffective in extracting generated light. Compared to the–curves measured at 90, those of the newly designed LEDsmeasured at 0show much greater improvement in light outputpower . As discussed above, this is related to the variation of

1720 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 49, NO. 10, OCTOBER 2002

Fig. 7. Light outputL–I characteristics of the conventional and newlydesigned LEDs with a� value of 90= . As described in the inset ofFig. 3(b), it is not possible to compare the absolute light output power of LEDsbetween measurement angle of 90and 0 .

the dominant light-emitting area. That is, it was shown that thedominant light-emitting area changes from arising through thetransparent electrode (vertical direction) to the sidewalls (lateraldirection) with decreasing ratios. Therefore, based on thisconsideration, it is clear that while vertical light extraction ismost favorable for the conventional LED, lateral extraction isfavored for the newly designed LED. However, because of thehigh sheet resistance of the n-type layer, the optical efficiency ofthe conventional LED might be significantly degraded. There-fore, it is also important to consider the sample dependenceof optical efficiency in LEDs with the innovative geometricaldesign.

Fig. 8(a) shows the – characteristics of the conventionaland newly designed LEDs with a value of 55 . Com-pared to the data shown in Fig. 7 at a measurement angle of90 , the newly designed LED with a value of 55 didnot show a drastic improvement of output power comparedto the conventional LED and even the output power forwas slightly degraded. Considering this observed sample de-pendence, it would be expectable that, if a higher-quality LEDsample is prepared, compared to those investigated in this study,the conventional LED may show a superior light output powerat a measurement angle of 90. However, we also believe thatthe difference will not be large because our conventional LEDsamples possess sufficient electrical and optical performance,in that they showed ideal light extraction through all areas ofthe transparent electrode. Moreover, considering the prominentimprovement in light output power in the lateral direction, it islikely that the newly designed LEDs still have great potentialfor use in practical applications.

In order to investigate optical performance in detail, the–characteristics were plotted and are shown Fig. 8(b). In the ver-tical direction 90 , the light output power of thenewly designed LED is lower than that of the conventional LED.Considering the same current density , this is due to thereduced junction area for the newly designed LED. However,in spite of the reduced junction area, the newly designed LED(new2)showed a higher output power in the lateral direction,indicating an extremely improved device efficiency. Fig. 8(c)shows the differential quantum efficiency versus current den-

Fig. 8. (a) Light outputL–I , (b) the light output power–current density (L–J),and (c) the differential quantum efficiency-current density (��J) characteristicsof the conventional and newly designed LEDs with a� value of 55= .

sity characteristics for all LEDs. According to this plot,it is clear that the newly designed LEDs dramatically maximizethe optical efficiency of the devices.

It is also noteworthy that, although the newly designed LEDs(new1andnew2) involve the common concept of perfect currentinjection, a clearly different behavior was observed in opticalas well as in electrical characteristics. That is, while thenew1showed electrical characteristics superior tonew2, new2showedsuperior optical characteristics compared tonew1. This signifi-cant pattern influence on device performance suggests that, forgeometrical LED design as a practical application, more consid-eration should be shown for design and additional testing mustbe taken into consideration.

KIM et al.: DESIGN AND FABRICATION OF LIGHT-EMITTING DIODES 1721

Fig. 9. I–V characteristics for both conventional Nichia-type (right side) andthe advanced type-LEDs (left side). (b) The light output power–current(L–I)characteristics for both LED types. The overall dimension for both LEDs is400� 400�m .

Finally, we attempted to employ the innovative geometricaldesign for a commercial Nichia type LED. For this work,conventional Nichia type and the advanced Nichia type LEDwere fabricated on the same wafer with a value of 55as shown in the inset of Fig. 9(a). It is clear that the advancedNichia type leads to improved electrical characteristics, re-sulting in a reduction in series resistance by 43.2%. In addition,improved optical characteristics were also observed for theadvanced type as shown in Fig. 9(b). These results indicate thatthe proposed innovative design has great potential for use inthe practical applications.

In order to explain the improved device performance of thenewly designed LEDs, we mainly focused on maximized cur-rent injection from the viewpoint of current spreading. How-ever, it is also noteworthy that the enhanced recombination andextraction efficiency might constitute an improved device per-formance as the result of the microsizing effect [13]. This expla-nation is plausible considering that the conventional line widthof the active area discussed in this study is below 45m.

IV. CONCLUSION

Based on the current spreading phenomenon in the GaN-based LEDs, promising fabrication method and design ruleswere investigated. It was possible to maximize the device

performance from the determination of the critical trans-parent-electrode thickness during the fabrication process. Forthis work, first, the resistance of the n-type layer should beminimized in order to maximize the carrier injection into then-type layer. The critical transparent-electrode thickness shouldthen be determined according to the relation of ,resulting in the maximization of the carrier injection into thep-type layer through the transparent electrode. This resultwill be very useful for the fabrication of highly efficientdevices in the conventional LED process. In addition, we alsoinvestigated the geometrical design rule for the highly efficientLED in terms of a perfect current spreading. Based on thisconsideration, it was even possible to realize the ideal LEDgeometry without the need for a transparent electrode, whichresulted in improvement in LED performances, compared witha conventional LED. Finally, it is concluded that, although thedetermination of the critical transparent-electrode thickness hasa great influence on the device performance, the design of idealgeometry is a better method to fabricate the highly efficientLED considering the extremely improved device performanceas well as the simple fabrication process.

ACKNOWLEDGMENT

The authors would like to thank N.-M. Park and J.-S. Jang,Kwangju Institute of Science and Technology, Kwangju, Korea,for many useful discussions.

REFERENCES

[1] S. Nakamura, T. Mukai, and M. Senoh, “Candela-class high-brightnessInGaN/AlGaN double-heterostructure blue-light-emitting diodes,”Appl. Phys. Lett., vol. 64, pp. 1687–1689, 1994.

[2] R. P. Vaudo, I. D. Goepfert, T. D. Moustakas, D. M. Beyea, T. J. Frey, andK. Meehan, “Characteristics of light-emitting diodes based on GaN p-njunctions grown by plasma-assisted molecular beam epitaxy,”J. Appl.Phys., vol. 79, pp. 2779–2783, 1996.

[3] S. Nakamura and G. Fasol,The Blue Laser Diode. Berlin, Germany:Springer, 1997.

[4] P. Perlin, M. Osinski, P. G. Eliseev, V. A. Smagley, J. Mu, M. Banas, andP. Satori, “Low-temperature study of current and electroluminescence inInGaN/AlGaN/GaN double-heterostructure blue light-emitting diodes,”Appl. Phys. Lett., vol. 69, pp. 1680–1682, 1996.

[5] I. Mártil, E. Redondo, and A. Ojeda, “Influence of defects on the elec-trical and optical characteristics of blue light-emitting diodes based onIII–V nitrides,” J. Appl. Phys., vol. 81, pp. 2442–2444, 1997.

[6] G. Y. Zhao, G. Yu, T. Egawa, J. Watanabe, T. Jimbo, and M. Umeno, “En-ergy-gap narrowing in a current injected InGaN/AlGaN surface light-emitting diode,”Appl. Phys. Lett., vol. 71, pp. 2424–2426, 1997.

[7] Y. Arakawa, T. Someya, and K. Tachibana, “Progress in GaN-basednanostructures for blue light-emitting quantum dot lasers and verticalcavity surface emitting lasers,”IEICE Trans. Electron., vol. E83-C, pp.564–572, 2000.

[8] H. X. Jiang, S. X. Jin, J. Li, J. Shakya, and J. Y. Lin, “III-nitrides bluemicrodisplays,”Appl. Phys. Lett., vol. 78, pp. 1303–1305, 2001.

[9] I. Eliashevich, Y. Li, A. Osinsky, C. A. Tran, M. G. Brown, and R. F.Karlicek, Jr., “InGaN blue light-emitting diodes with optimized n-GaNlayer,” Proc. SPIE, vol. 3621, pp. 28–36, 1999.

[10] H. Kim, J.-M. Lee, C. Huh, S.-W. Kim, D.-J. Kim, S.-J. Park, and H.Hwang, “Modeling of a GaN-based light-emitting diode for uniform cur-rent spreading,”Appl. Phys. Lett., vol. 77, pp. 1903–1904, 2000.

[11] H. Kim, S.-J. Park, and H. Hwang, “Effects of current spreading on theperformance of GaN-based light-emitting diodes,”IEEE Trans. ElectronDevices, vol. 48, pp. 1065–1069, 2001.

1722 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 49, NO. 10, OCTOBER 2002

[12] J. E. Siewenie and L. He, “Characterization of thin metal films pro-cessed at different temperatures,”J. Vac. Sci. Technol. A, vol. 17, pp.1799–1804, 1999.

[13] S. X. Jin, J. Li, J. Y. Lin, and H. X. Jiang, “InGaN/GaN quantum wellinterconnected microdisk light-emitting diodes,”Appl. Phys. Lett., vol.77, pp. 3236–3238, 2000.

[14] D.-J. Kim, Y. T. Moon, K. M. Song, I.-H. Lee, and S.-J. Park, “Effectof growth pressure on indium incorporation during the growth of InGaNby MOCVD,” J. Electron. Mater., vol. 30, pp. 99–102, 2001.

[15] D. K. Schroder,Semiconductor Material and Device Characteriza-tion. New York: Wiley, 1990.

[16] P. G. Eliseev, P. Perlin, J. Furioli, P. Sartori, J. Mu, and M. Osinski, “Tun-neling current and electroluminescence in InGaN:Zn,Si, AlGaN/GaNblue light-emitting diodes,”J. Electron. Mater., vol. 26, pp. 311–319,1997.

Hyunsoo Kim was born in Ulsan, Korea, on October 6, 1974. He received theB.S. degree in metallurgical engineering from Pusan National University, Pusan,Korea, in 1997, and the M.S. degree in materials science from Kwangju Instituteof Science and Technology, Kwangju, Korea, in 1999, where he is currentlypursuing the Ph.D. degree in materials science and engineering.

His current research interests are device design, modeling, and reliability ofGaN-based light-emitting diodes.

Seong-Ju Parkwas born in Korea in 1952. He received the B.S. degree in chem-istry from Seoul National University, Seoul, Korea, in 1976, and the M.S. andPh.D. degrees in physical chemistry from Seoul National University and Cor-nell University, Ithaca, NY, in 1979 and 1985, respectively.

From 1985 to 1987, he worked as a Postdoctor with the IBM T. J. Watson Re-search Center, Yorktown Heights, NY. From 1987 to 1995, he was with the Elec-tronics and Telecommunications Research Institute, Daejon, Korea, as a Prin-cipal Researcher. In 1995, he joined the Faculty at Kwangju Institute of Scienceand Technology, Kwangju, Korea, as a Professor in the Department of MaterialsScience and Engineering. Presently, he is a Professor with the department andthe Director of the Center for Photonic Materials Research. He has engaged inresearches on growth and characterization of semiconductor epitaxial structures,characterization of nanoelectronic and photonic materials, atomic and electronicstructures of semiconductor surfaces, and plasma etching and reaction mecha-nisms.

Hyunsang Hwang(M’93) was born in Korea in 1966. He received the B.S. de-gree in metallurgical engineering from Seoul National University, Seoul, Korea,in 1988, and the Ph.D. degree in materials science from the University of Texas,Austin, in 1992.

From 1992 to 1997, he was with the LG Semiconductor Corporation, Korea,as a Principal Researcher. In 1997, he joined the Faculty at Kwangju Institute ofScience and Technology, Kwangju, Korea, as a Professor in the Department ofMaterials Science and Engineering. Presently, he is a Professor with the depart-ment. His research interests are process and device design of deep submicronMOSFET, MOSFET device reliability, ultrathin dielectric, and optoelectronicdevices.