This action might not be possible to undo. Are you sure you want to continue?
IEEE TRANSACTIONS O N ELECTRON DEVICES,
8 , AUGUST 1979
with convex mirror windows were fabricated. lp~ak The emission wavelength was 1.17 pm andthe full widthat half maximum was about 1000 A. The I-L characteristics exhibited fairly good linearity. ACKNOWLEDGMENT
Y. Suematsu, Prof. 13. The authors wish to thank Prof. Fukuyo, and Prof. S. Furukawa, as well as Associate I? .of. N. Oura, Tokyo Institute of Technology, and T. Yamamloto, KDD Research Laboratory, helpful for discussion of :his work.
[ l ] H. Osanai, T. Shioda, T. Moriyama, S. Araki, M. Horiguchi, T. Izawa, and H. Takata,“Effect of dopants on transmission loss oflow-OH-contentopticalfibers,” Electron. Lett., vol. 12, pp.
549-550, Oct. 1976. T. P. Pearsall, B. I. Miller, R. J. Capik, and K. J. Bachmann, “Efficient lattice-matched double-heterostructure LED’s at 1.1 Appl. Phys. Lett., vol. 28, pp. pm fromGaxInl-xAsyP1-y,” 499-501, May 1976. A. G.Dentai, T. P. Lee, C. A. Burrus, andE. Buehler, “Smallarea, high-radiance C.W. InGaAsP L.E.D.s emitting at 1.2 to 1.3 pm,” Electron. Lett., vol. 13, pp. 484-485, Aug. 1977. K. Oe, S. Ando, and K. Sugiyama,“Surface emitting LED’s for the 1.2-1.3 pm wavelength with GaInAsP/InP double heterostructures,” Japan. J. Appl. Phys., vol. 16, pp. 1693-1.694, Sept. 1917. H. Nagai and Y. Noguchi, “InP/Ga,Inl-,As P1-,, doubleheterostructure for 1.5 pm wavelength,” Appl. h y s . Lett., vol. 32, pp. 234-236, Feb. 1978. I. Umebu, 0. Hasegawa, and K. Akita, “InGaAsP/InP D.H. L.E.D.s Electron. Lett., vol. 14,pp. forfiber-opticalcommunications,” 499-500, Aug. 1978. K. Wakao, K. Moriki, T. Kambayashi, and K. Iga, “GaInAsP/InP DH laser grown by newly designed vertical LPE furnace,” Japan. J. Appl. Phys., V O ~ .16,pp. 2073-2074,Nov. 1977.
Comparison of Surface- and Edge-Emitting LED‘s for Use in Fiber-Optical Communications
Abstract-The performance of state-of-the-art double-heterojunclion
(DH)surface and edge emitters are compared with respect to their use
in high-data-rate fiber-opticalcommunication systems. Thick-window (20-25-pm) surface emitters with 2-2.5-pm thick active layers and emitting up to 15-mW optical power at 300 mA have been fabricated. For edge emitters, we use very-high-radiance-type devices with 5(313-8 thick active layers. For these two types of LED’s we examine differences in structure and light coupling efficiency to fibers of various numerical apertures (NA). For typically good devices wecompare the diodes’ output power capabilities, the powers coupled into step- :md graded-index fibers of various NA, and their respective frequency response. For the same drive current level, we find that edge emitters couple more power than surface emitters into fiberswith NA 50.3. The edge emitters also have a 5 times larger bandwidths. We estimate that an edge emitter can couple 5-6 times more power into low numerical aperture (NA 50.2) fibers than a surface emitter of the same bandwidth. We conclude that edge emitters are preferred to surface emitters for optical data rates above 20 Mbits/s. Manuscript received January 2, 1979;revised January 30, 1979. ::his work was supported in part by the Department of the Air Force, Rmme Air Development Center, Hanscomb Air Force Base, MA, and in pas’ by RCA Laboratories, Princeton, NJ. The authors are with RCA Laboratories, Princeton, NJ 08540.
I. INTRODUCTION IGH-RADIANCE light-emitting diodes (LED’s) have been the subject of intensive research and development for fiber-optical communications, due to their linearity, small temperature sensitivity, and inherently small sensitivity to gradual degradation (compared to injection lasers). As a result of these efforts, two basic types of diodes have emerged: surface-emitting LED’s [ l ] and edge-emitting LED’s [ 2 ] . We presentin this paper a comparison betweenstate-of-the-art surface and edge emittersfabricatedinourlaboratory.The edge emitters are of the “very-high-radiance” type on which we reported previously . Thesurface emitters are “thickwindow”typediodes whose structure and performance are presented below. The two types of diodes are first compared from a general point of view; differencesin structure, emission angular distribution, and the consequent differences coupling in efficiency to optical fibers. Fortypically gooddiodes we compare the optical powers emitted into air, the powers coupled into optical fibers of various numerical apertures (NA), as well
0018-9383/79/0800~1230$00.75 0 1979 IEEE
ETTENBERG: AND BOTEZ
COMPARISON OF SURFACE- AND EDGE-EMITTING LED’S
p+ p p
0 0.2 0.4 X
DOPING (cm-3, Ge Ge Ge
0 . 3 &
- AIXGaI.XAs- AlxGal-xAs- A\xGal+As-
- GaAs -
THICK- WINDOW S U R F A CE M I T T E R E 22oc
C P O O p m J
3 5 0 p W COUPLED INTO 0.14 NA FIBER
Fig. 1. “Thick-window” DH surface emitter:structure; A1 content (x) variation; and doping concentrations.
as their respective frequency response. We find that a surface emitter can couple almost as much power as an edge emitter into low NA fibers, but at a high price in bandwidth. A discussion of the frequency response for the two types of diodes is also presented. 11. THICK-WINDOW (NGa)As SURFACE EMITTERS The structure of our surface-emittingLED is shown in Fig. 1. Thefour-layerdoubleheterojunction (DH) was grown by liquid-phase epitaxy (LPE) between 900 and 840°C in a previously described LPE multibinboat . Formostsurface emittersthe n-(A1Ga)As confining layer the of DH structure is also thewindowfortheradiationgenerated in the active layer. We use the name “thick-window” for our diodes, since the n-AlGaAs layer was grown much thicker than for a regular DH laser structure (i.e.,20-25pm versus 2-3pm). The relatively thick window layer is grown by cooling between 900 and 855°C. As a result of A1 depletion from the solution, aswell as thedifference in segregationcoefficientsbetween Al and As  a graded Al,Gal-,As layerresults with an Al concentration x varying be.tween 0.34 and 0.22 over -22 pm (see Fig. 1). The thick-window layer provides a sturdy diode base, which in turn considerably eases diode handling and potentially gives a more reliable device. Furthermore, the thickness of the window layer does not seem to affect the diode external efficiency since the light emitted in the active layer (x = 0.05-0.07) is only negligibly absorbed inn-Al,Ga,-,As material with x > 0.2 and n < 1 X 10” ~ m - ~ . et al. have Abe reported  on a similar thick-window structure, which they fabricate by using a modified LPE boat . Theactivelayerthicknessinourstructure is in the 2.02.5-pmrange,which hasbeenshownexperimentally  as well as predicted theoretically , [ 9 ] to be optimal for maximum diode external quantum efficiency. The doping concentrationintheactivelayer was chosen to be 5 X 10’cm-3 (Ge) in order to improve the device frequency response without affecting external its quantum efficiency. doping The concentrations of the other layers are a compromise between the requirements of good current confinement [lo] and low device series resistance. The difference in Al content between the active layer and the passive layers was kept at relatively small values (Ax 0.17)sinceforourrelativelythickactive layers (i.e., 2 2 pm) the leakage currents are negligible when normallyoperatingcurrentdensitiesareconsidered . In turn, the relatively low AI content of the passive p layers insures a low device series resistance [ 111. By using a thermally
/ 8800 1
8600 8500 8700 ~ 1
8300 8400 8200 1 ,
, 8000 1
WAVELENGTH, X (&
(b) Fig. 2. (a) Output optical power (into air) and voltage drop as a funcLED tion of dc drive current for the best“thick-window”surface (Fig. 1). At the 12-mW output level 350 p W were coupled into flatended 0.14 NA step-index fiber of 90-Mm-diameter core. (b) Typical spectrum for “thick-window” surface LED’s (Fig. 1).
deposited mask an SiOz and evaporated metallic trilayer (Ti/Pt/Au), 50-pm-diameter dot contacts were placed on the p side of the wafer. The wells were etched the into n-type GaAs substrate with the mixture 95HzO2 (30 percent): 5 N H 4 0 H (29 percent N , , which at H) stops the GaAs-nAlo.3Gao.,As interface. The performance of the best “thick-window” surface emitters thusfabricated is showninFig. 2(a) and (b). The output optical power was measured by placing a l-cm diameter calibrated Si photodiode in close proximity to the etched-wellrim of the LED. The output powers at 100 and 300 mA were as high as 5.8 and 15 mW, respectively.Above200mA,the power versus current curve deviates from a straight line due to Joule heating. In the linear output regime, an external quantum efficiency as high as 4.5 percent is obtained. These results are comparable to the bestresults reported in the literature for surface emitters with an active layer dopant concentration inthemiddle1017cm-3range , , -.Atthe 12-mW output power level, 350 pW were coupled into 0.14 NA flat-ended fiber of 9O-pm core diameter (i.e., - 15.4-dB coupling without loss), the use of index-matching fluids. Plotted in the same figure is the diode I-V curve, which shows that the device electrical characteristics were quite good (i.e., 1.7-V potential drop at 100 mA and =,2-Sl series resistance). Fig. 2(b) showsatypicalspectrumofoursurfaceemitters. The spectral halfwidth is approximately 350 A. Many diodes
EMITTER EDGE EMITTER SURFACE
IEEE TRANSACTIONS ON
ELECTRON DEVICES, VOL. ED-26, NO. 8 , AUGUST 1979
BEAM ANGLE FROM NORMAL
EMITTED LIGHT ANGULAR DISTRIBUTIONS
Fig. 3. Surface- andedge-emitting LED's. Top: Schematicreprescntames tions. Bottom: Isolated active layerswith light-generating v o l ~ marked by dashed lines, and the diodes' respective emission angular distributions.
showed interference peaks similar to the ones previously reported by King et al. and Shimano et al. 1151, Tllese peaks can be represented by the relation
45O 0 4' 5 9' 0 B E A MA N G L EF R O MN O R M A L
(b) Fig. 4. Typical far-field intensity patterns in the junction plane (611) and in a plane perpendicular to the junction (@I) for: (a) very-highradiance edge emitters [ 31 ; (b) surface emitters.
where Ah is the spacing between the interference peaks, h is the vacuum wavelength, ne&) is an effective index of ref raction (which takes into accountdispersion effects), and L is the difference in optical path length between the interfering beams. For Ah EE 34 A, h = 8380 A, and n,ff 2 3.9 we obtain = 53pmfrom (1). This corresponds t o interferencebetween transmitted beams and twice-reflected the (at air-window interface and metal-cap interface) beams, in agreement v i t h the findings of Shimano et al. [ 151 . However, our result:; as well as the ones reported by King et al. , disprove those authors' assertion thatinterference peaks cannotoccur for activelayerthicknesses larger thanthe electro1uminesce:lce wavelength (0.6 pm).Interferencepatterns shouldoccur as long as the grown layers interfaces are parallel to one another.
surface emitters is more difficult and of much lower yield than that of edge emitters. The active layers for both typesof diodes are shown isolated at the bottom of Fig. 3. The respective light generating volumes for both diodes are indicatedwith dashed lines. In a surface emitterthe generatedlightspends little timein the active region and thus is not seriously subjected to reabsorption. By contrast, in the edge emitter light must travel most of thediodelength beforebeing emitted into the air. As a consequence, edge emitters have lower external quantum efficiencies than surface emitters. However, the effect of absorption in the active layer can be drastically reduced in the edge emitter by making symmetric DH structures of very thin active layers (300-500 A) .Thusthe randomly excited 111. SURFACE VERSUS EDGE EMITTER: fundamental transverse modes have relatively little overlap GENERAL CONSIDERATIONS of the field intensitywiththe absorbing active region (as Fig. 3 displays the geometry and emission distribution for shown in Fig. 3); consequently, the diode external quantum the two types of LEDs. For (A1Ga)As DH structure surface efficiency is increased. The effect can be modeled with emitters, a well must be etched through the GaAs substrate to the help of the radiation confinement factor ro [I61 applied prevent heavy absorption of the emitted radiation. The labto a wavelength-averaged backgroundabsorption loss coeffirication also requires precise alignment of the dot contact>u,.ith cient [ 171 . the etched well. After etching, the relatively fragile diode base The most striking difference between the two diodesappears is usually strengthened by electroplating a thick Au layer li:.2] when comparing the angular distributions of theiremission. onthediode pside. By contrast,the edge-emitting LEI) is For asurface LED, the emitted light has atwo-dimensional fabricated by standard oxide-stripe laser technology without Lambertian distribution with a 120" full width at half power post-contact etching and careful contact alignment. The icm- (FWHP) beamwidth (Fig. 3 and 4(b)). The corresponding tact stripes for the edge emitters are 50-70pm wide, to maich radiation pattern is thus virtuallyspherical (Fig. 3). For the typicalfiber diameters (50-90pm).The diodes are usudly edge emitter, the presence of a transverse waveguide has the 100-150 pm long. The front emitting facet has an antirefk ct- effect of funneling most of the emitted radiation in relatively ing (AR) coating. Similar coatings and/or index-matching narrow angles in the plane perpendicular to the junction . epoxies are used insurface emitters for more efficient light As shown in Fig. 4(a) in the plane parallel to the junction the extraction . We generally findthatthefabrication of far-field intensity profile is still Lambertian with 120" beam-
BOTEZ AND ETTENBERG: COMPARISON OF SURFACE- AND EDG.E-EMITTING LED’S
SURFACE OR “BURRUS” LED
,EDGE EMITTING L E D ‘Ot 8-
0.3 0.4 0.5 FIBER N U M E R I C A L APERTURE
Fig. 5. Relative coupling loss to opticalfibersof various numerical aperturesfor surface emitters, edge emitters,and lasers. Thedata surface and edge emitters, this work (Fig. 9); pointsare: 0 and 0,surface emitters, from [ 1 , [ 61 , [ 121. 1
Fig. 6. Output optical power versus dc drive current for typically good surface and edge emitters. The surface emitters have 50-pm-diameter dot contacts and the edge emitters have 65-pm wide stripe contacts of 100-pm length.
J= 2 3kA/cm2 .
CURRENT ( m A 1
width (FWHP), while in the plane perpendicular to the junction far-field patterns beamwidths with of -30’ can be fiber coupling efficiencies than surface emitters it is also true that surface emitters have higher external quantum efficiencies. obtained . The two-dimensional radiationpatternofthe Thus it is more relevant to compare the absolute coupled edge emitter looks then as an oblate spheroid (Fig. 3). It must optical powers for typically good surface and edge emitters. bestressed that as opposed to large-optical-cavity-like edge emitters [I 71 ,  , for DH edge emitters the light escaping Iv. SURFACE VERSUS EDGE EMITTERS: the active layer gives only a negligible contribution to the deTYPICALLY GOOD DEVICES vice angular distribution . Possible explanations are light A. Power-Current Characteristics absorption in the “cap” layer and substrate and/or spontaneousWithin the same batch and/or from wafer to wafer the fabriemission-rate enhancement for the guided modes . performance. We The difference in radiationpatternsbetween surface and cateddiodes displayacertainvariationin edge emitters strongly affects their relative coupling to optical find that the scatter in device external efficiency is especially fibers of various NA. Fig. 5 shows the estimated coupling effi- pronounced for surface emitters. We believe that the differciencies for surface emitters, edge emitters, and injection lasers ence in yield between surface and edge emitters is due on one as a function of the NA of flat-ended optical fiber. Coupling hand to intrinsically simpler processing to access the generated efficiency calculations were done for the case when the source light in the edge emitter, and on the other hand we find that size is smaller than the fiber corearea. For surface emitters dot contacts are less adherent than stripe contacts. Lee and variawith Lambertian angular distributions the coupling efficiency Dentai also have reported surfaceLEDefficiency tions (up to 50 percent) for devices from the same batch. The is simply given by  ,  . For edge emitters and lasers one has to numerically compute the couplingintegral choice of a typically good surface emitter was based on results  (i.e., the percentage oftheemittedpowercontained from a large number of diodes obtainedfrom nine wafers. withinthe fiber acceptancecone angle). Thewidths of the Specifically, the chosen diodehas -70 percent of the best shaded regions shown in Rg. 5 represent estimated variations diode efficiency (Fig. 2(a)), which, as mentioned previously, in coupling efficiency introduced by several factors: the differ- compares very well with the best results reported in literature ence between fiber core and source diameters when the latter  ,  , [I21 - for middle lo” cm-3 active region is the larger of the two  ; deviations from a Lambertian doping levels. We would like to point out that, although lower angular distribution for the emission of surface emitters; and doping levels (e.g., 10l6 cm-3 and n-type) would provide variations in beamwidth for edge emitters andlasers. somewhat higher diode efficiencies , such devices would For low NA fibers (i.e., NA < 0.2) the’ couplingefficiency not be relevant for our comparison since their relatively low for edge emitters can be 5 to 7 dBbetterthanfor surface (electrical) bandwidths (i.e., < 6 MHz) prohibit their use in emitters (Fig. 5 ) . We also plot experimental data (circles for fiber-optical communications applications  . surface emitters and squares for edge emitters) obtained by us For edge emitters a typically good device was also chosen. The edge-emitting LEDs had undoped active layers -500 A (solid points) by or other workers [ l ] ,  ,  (open points). As can be seen, the experiments agree reasonably well thick, spectral halfwidths between 200-250 A , , and the output power capawiththe calculated curves. While edge emitters have higher 100-pm diode lengths.Fig.6shows
IEEE TFLiNSACTIONS ON ELECTRON DEVICES, VOL. ED-26, NO. 8 , AUGUST 1 9 7 9
EDGE EMITTER WITH LENS LENSED FIBER
CLEAVED END NO LENS
Fig. 7 . Near-field photographs (top) and intensity profiles (botto n) for the surface and edge emitter chosen for comparison (Fig. 6 )
bilities of the two devices. At 150-mA bias the surface LED emits 2.6 times more power than the edge-emitting LED. That occurs,however, at approximately three times more current density for the surface emitter. No index matching epoxjes or conformal coatings were used. The near-field photographs and intensity profiles for the two devices are shown in Fig. 7. The surface emitter had a 50-pm contact diameter while the edge emitter had a 65-,um wide contact stripe. In both cases, the emission intensity is quite uniform over most of the emitting region. For coupling to optical fibers both devices were operated at a dc drive current of 150 mA.
Fig. 8. Light detected at the end of a 1-km 0.28 NA graded-index fiber (left scale) as a function of the drive current applied to the diodes of Fig. 6 . Solid and dashed lines correspond to flatended fiber (bottom right inset) and lensedend fiber (top left inset). The right scale shows the coupled power computed by using the fiber loss coefficient (5 dB/km).
SURFACE EMITTER (50prn 0IA.I EDGE E M I T T E R (65pm S T R I P E ) G R A D E D INDEX F I B E R
B. Coupling to Optical Fibers o f VariousNA Both graded- and step-index fibers were used in the coupling experiments. Mode-stripped fibers were placed with the: help of amicropositioner inside theetched well of thesu~face emitters and in close proximity to the air-n-AlGaAs intedace. For edge emitters the fibers were positioned close to the diode cleaved facet. The micropositioner was then used to maximize the light coupling. No matching lenses were employed. With one exception (i.e., a lensed 0.28 NA graded-index fiber) all coupling experiments were made with flat-ended (cleared) fibers. Fig. 8 shows the results of lightcoupling to a1-km-iong graded-index fiber of 60-bm core diameter and 0.28 NA. The optical power detected at the end of the 1-km fiber is plo :ted versus drive current for surface and edge emitters in two cases: flat-ended fiber and lensed-end fiber. The use of a long flber ensures that the measured power is only that which propag,ites in the fiber core. Then, by using the fiber loss paramete- (5 dB/km in this case) we extrapolate to find the core-coupled powers (right scale in Fig. 8). Over the whole range of drive currents (0-300 mA) the edge emitter is found to couple =:20 percent more power than the surface emitter. Lensing of the fiber receiving end brings about a IO-percent improvemen: in coupling efficiency for both LED types, This relatively snlall difference between the lensed and the flat-ended fibers is most probably due to the fact that the fiber core size is comparable to the size of the sources. Also, due to the radial variatiort of the refractive index, lensing of agraded-indexfiber is less effective in enhancing the coupling efficiency than in the case - of a step-index into . fiber
N U M E R I C AA P E R T U R E L
Fig. 9. Coupled optical powers versus fiber numerical aperture for the LED's of Fig. 6, atthe 150-mA drive current level. Thephotoresponse curves correspond to powers measured as the photodetector was moved at various distances away from the sources. The equivalent NA was calculated by using (2). The data for 0.28 NA gradedindex fiber are obtained from Fig. 8.
The difference in emittedpower angular distribution between surface and edge emitters is evidenced in Fig. 9 by results of coupling experiments to fibers of various NA, and by photoresponse measurements of 1-cm-diameter a detector at various distances from the two sources. For the latter experiment we calculated the solid cone angle subtendedbythe detector at a given distance from the source, and translated it an equivalent NA. The relation used is
BOTEZ AND ETTENBERG:
COMPARISON O F SURFACE- AND EDGE-EMITTING LED'S
0 = 8, = sin-' (NA)
where 8 is the half-angle of the cone covered by the detector area and 8, is the fiber critical angle, which from a coupling point of viewis thefiberacceptance half-angle. For a stepindex fiber NA = where n n,2 are the refractive l indices of the fiber core and cladding, respectively. The photoresponse curves (solid for the edge emitter and dashed for the surface emitter) cross around 0.25 NA, which correspondsto a 14" acceptance half-angle. For NA < 0.25the edge emitter providesmorepower thanthe surface emitter into the angle covered by the measurement. The opposite is truefor NA > 0.25. These results are supportedby light coupling experiments for the two types of diodes to several different NA fibers (round data points for the surface emitter and square data points for the edge emitter). The data points shown in Fig. 9 were obtained for flat-ended step- and gradedindex fibers. Specifically, we used 50-pm core diameter stepindex fibers of NA values 0.22, 0.28, 0.3, and 0.4; a 90-pm core diameter 0.14 NA step-index fiber; and graded-index fibers of 0.14 and 0.28 NA with 90- and 6Oym core diameters, respectively. The fibers were relatively short (3-4 ft) and mode stripped so as to eliminate cladding modes from the measurement. As can be seen from Fig. 9 the surface and edge emitters coupled the same amount of power for NA = 0.3. Just as in the case of the photoresponse curves, for NA values lower than the cross point, the edge emitter couples increasingly better than the surface emitter. Thus for 0.14 NA fibers, the edge LED couples ~ 4 percent more power than the sur0 face LED. Similar ratios are obtained when coupling into 0.14 NA graded-index fibers. The couplingefficiencies for these graded-indexfibers are about 30 percent of thoseforstepindex fibers of the same NA and core diameter. We conclude that for NA < 0.3 edge emitters coupleslightly better than surface emitters and that the roles are reversed for NA > 0.3. Thedifferences, however,are not very dramatic. It will be shown in the next subsection that the real difference between thetwoemitters lies intheirfrequency response. We note that our experimentally found crossover point for edge and surface emitters (i.e., NA = 0.3) is ingood agreement with theoretically calculated values by Gloge  (i.e., NA = 0.25) and Marcuse  (i.e., NA = 0.23). C. Frequency Response Thefrequency response of the two types of emitters was measured at abias level of 150 mA and the results are presented in Fig. 10. The measurements were made with a 30-mA peak-to-peak signal. The electrical bandwidth (i.e., the 1.5-dB rolloff point in the modulated light output) is 17 MHz for the surface emitter and 100MHz for the edge emitter. To understand this difference, we must explore the factors which controlthebandwidth.Tofirstorderthe electrical bandwidth (f) is a measure of the injected carrier lifetime (7) ~261
5 7 IO 20 MODULATION FREQUENCY
Fig. 10. The frequency response for the LED's of Fig. 6, at the 150mA drive current level. A 30-mA peak-to-peak signal was used.
The nonradiative component for relatively lightly doped active regions and active regions thicknesses (d) less than or equal to a carrier diffusion length is dominated by interfacial recombination velocity (s) at the Alo.22Gao.,8As/Alo,05Gao.9sAs interface
The radiative component for a p-type active region is 
- = B ( p o + An) 1 r
where B is therecombinationcoefficient, p o is the active region hole concentration, and A n is the injected carrier concentration. At injection levels such that the injected carrier density is much lower than the doping density ( A n << p o ) and/or the active region thickness is larger than the carrier diffusion length thelifetime is solely determinedbythe doping level [81 , [211
- = Bpo,An<<po
At the other extreme, we have the case An >> p o when bimolecular recombination (i.e., direct recombination of injected electrons and holes) occurs and then , , 
which, in turn, is composedof
( , and 7)
For heavily doped active layers the lifetime is determined (at most current levels) by the doping density level (i.e., (7)). For lightly doped active layers the situation can be quite complicated. Thus as shown by Lee and Dentai for surface emitters , as the drive level is increased the lifetime is successively determined by the doping level; by a combination of doping and bimolecular recombination  ,  ;and by bimolecular recombination (see (8)). For our surface emitters, which are optimizedfor high externalquantum efficiency, thebandwidth is relatively small due to two factors: on one hand we have thick active layers (2-2.5 pm) which considerably reduce
IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. ED-26, NO. 8 , AUGUST 1 9 7 9
‘ IO ‘
EDGE E M I T T E R SURFACE EMITTEF Fig. 11. Optical pulse response of the LED’s compared in Figs. 6-,10, when subjected to current pulse trains of various data rates.
the effects of bimolecular recombination (i.e,, (8)), and on t.le other hand, the doping level is moderately low(5 X 1017 cm-”,I. The edge emitters studied here (i.e., of undoped and vlery thin active layers) operateina regime where the lifetime is controlled almost completely by injected carrier density  , Due to strong bimolecular recombination,the edge emitter operates with a short carrier lifetime, which results in a hi1;b modulation frequency. Theeffect of modulationbandwidthonoptical pulse r:., sponse is illustrated in Fig. 11 for lo-, 20-, 3 5 , and 50-Mbit/s data rate pulse trains applied to our edge and surface emitters. While the edge emitter can handleeach of these datarates comfortably, the surface emitter produces intersymbol interference at rates muchin excess of 20 Mbitsls. For surface emitters, the most common way to increase the bandwidth is to heavily dope the active layer , , [l;] since, as shown in (7), this decreases the carrier lifetime. Hal\+ ever, abandwidth increase by heavy dopingoccurs atthe expense of external quantum efficiency  , , [ 121, [30 1 . The decrease in diode external efficiency with increased doping is commonly attributed to an increase in the density of nonradiative recombination centers as the active layer d0pir.g density is increased  , , . The tradeoff between J9, the output optical power, and f, the electrical bandwidth is best illustrated by the fact that the power-bandwidth product (at the same drive level) is found to be almost constant for all surface emitters of active layer dopingconcentrations p o 10l8cm-3-.For lowerdoping levels the P X ,f
product actually decreases with respect toits heavy-doping almost constant value [ 121 , [ 131 . We show in Fig. 12 output power versus bandwidth results of Lee and Dentai  for DH surface emitters at the drive level I = 300 mA, and with active layer doping levels in the range 2 X 1017 ~ m - ~ - 210’’ ~ m - ~ . X Their data are seen to be bordered by two parallel lines corresponding to P f - o . 8 . Here, it is apparent that the light output of the surface emitter cannot be increased without a corresponding sacrifice in frequency response. In Fig. 12, we also include our data points for surface and edge emitters versus the results ofLee and Dentai  , For the reader’s benefit we show a scale of the equivalent optical bandwidth, i.e., the frequency at which the diode light output drops to half its dc value. Oursurface emitters (round dot) fall in the range defined by those authors. The edge emitters (square dot) of 120-MHz electrical bandwidth, provide 30-60 percent more power (into air) then surface emitters of same bandwidth. If we now take into account thediodes’ respective coupling efficiency to optical fibers (Fig. 5) it becomes obvious thatfor low NA fibers (i,e,, NA < 0.2) the edge emitters couple 5-6 timesmore power than surface emitters of same bandwidth. Furthermore, thisoccurs at three times less current density for the edge-emitting LED’s. The question that arises is: can surface emitters be improved to have the same radiance-bandwidth product as that of edge emitters? One approach for increasing the surface LED bandwidth is tightcurrentconfinement (e.g., protonbombardment)  . Although the bandwidth is increased, the diode external efficiency is reduced and thus the power-bandwidth product is lowered or remains the same as before the current confining process [ 131 . As in the edge-emitter case, the bandwidth can also be improved by reducing the active layer thickness, since 7, fi(see (8)). However, as d is decreased from 2.5 to 0.5 pm, the nonradiative interfacial surface recombination becomes a factor (see (S)), and reduces the diode external quantum efficiency. For instance, the authors of  show a reduction in power by as much as a factor of 2 as they decrease
BOTEZ AND ETTENBERG: COMPARISON OF SURFACE-
AND EDGE-EMITTING LED’S
the active thickness layer from 2.5 to 0.5 pm.For layers thinner than 0.5 pm, they find some relative improvement in thediodeefficiency, which they attribute to a reduction of the interfacial recombination velocity? good agreement with in the previously reported studies by Ettenberg and Kressel  . Thus it is conceivable that surface emitters of very thin active layers (d N- 500 A) could achieve higher radiance-bandwidth products than the best surface emitters reported to date.
State-of-the-art surface-emitting LED’s were fabricated and compared to previously reported state-of-the-art  edgeemitting LED’s. For typically good devices, it is found that at the same drive level, the surface emitter provides 2.5-3 times more power into air than the edge emitter. For evaluation of diode coupling to optical fibers, we used surface emitters with 50-pm-diameter dot contacts and edge emitters with contact stripes 65 pm wide and 100 pm long. Thus at the same drive currents, our edge emitters were operated at a current density approximately 3 times smaller than that of our surface emitters. For fiber NA less than 0.3, edge emitters couple sli&tly more power than surface emitters (a maximum ratio of 1.4 is obtainedfor NA = 0.14). For NA > 0.3,thesituation is reversed. At the same drive level, thebandwidth of the edge emitter is found to be more than 5 times larger than the surface emitter’s (i.e., 100 versus 17 MHz). The power versus bandwidth results are compared to extensive data by previous workers . From that comparison, it is inferred that edge emitters can couple 5-6 times more power in low NA (NA 5 0.2)fibers than surface emitters of comparable bandwidth. Based on theseobservations, we conclude that edge emitters have a clearly superior performance over surface emitters when considering optical communications data rates between20 and 100 Mbits/s, where low NA fibers are usually required. Surface emitters, in turn, are preferable t o edge emitters for short optical data links with high NA fiber ,  and data rates less than 20 Mbits/s. Finally, we find surface emitters to be more difficult to fabricate, and thus potentially more costly than edge emitters.
ACKNOWLEDGMENT The authors gratefully acknowledge helpful discussions with H. Kressel, C. J. Nuese, and I. Ladany, and the excellent technical expertise of I. J. Hegyi. They also thank M. Harvey, D. B. Gilbert, D. Marinelli, T. Purman, F. Hawrylo, and H. Kowger for technical assistance.
[ 1 ] C. A. Burrus,“Efficient small-area GaAs-Gal,Al,As heterostructure electroluminescentdiodescoupled tooptical fibers,” Proc. IEEE,vol. pp. 1263-1264, Aug. 1971. 59, [ 2 ] M. Ettenberg, K.C. Hudson, and H. F.Lockwood, “High radiance light-emitting diodes,” IEEE J. QuantumElectron., vol. QE-9, no. 10, pp. 987-991, Oct. 1973. [ 3 ] M. Ettenberg, H. Kressel, and J. P. Wittke,“Very high radiance edgeemitting LED,” IEEE J. QuantumElectron., vol. QE-12, p. 360, 1976. [ 4 ] H. F. Lockwood and M . Ettenberg, “Thin solution multiple layer epitaxy,” J. Cryst. Growth, vol. 15,pp. 81-83, 1972. [ 5 ] I. Crossley andM. B. Small, “The applicationof numerical methods to simulate the liquid phase epitaxial growth of Gal,Al,As from
anunstirred solution,” J. Cryst. Growth, vol. 15, pp.’ 268-214, 1972. M. Abe, I. Umebu, 0. Hasegawa, S. Yamakoshi, T. Yamaoka, T. Kotani, H. Okada, and M. Takanashi,“High-efficiencylongIEEE lived GaAlAs LED’s for fiber-optical communications,” Trans. Electron. Devices, vol. ED-24, pp. 990-994, July 1971. S. Isozumi, Y. Komatsu, M. Abe, T. Yamaoka, and T. Kotani, “Liquid phase epitaxial growth of Ga+,Al,As for LEDs with a 5 0 W r n thickwindow layer,” Paper C.1. presented at the 3rd IEEE Specialist Conf. on the Technology of Electroluminescent Diodes, San Francisco, CA, Nov. 1978. T. P. Lee and A. G. Dentai, “Power and modulation bandwidth of GaAs-AlGaAs high radiance LED’s for optical communication systems,” IEEE J. QuantumElectron., vol. QE-14, no. 3, pp. 150-159, Mar. 1978. D. Marcuse, “LED fundamentals: Comparison of front- andedgeemitting diodes,” IEEE J. QuantumElectron., vol. QE-13, no. 10, pp. 819-827, Oct. 1977. J. Zucker and R. B. Lauer, “Optimization and characterization of high radiance (Al, Ga)As double-heterostructure LED’s for optical communication systems,” IEEE Trans. Electron. Devices, vol. ED-25, pp. 193-199, Feb. 1978. A. J. SpringThorpe, F. D. King, and A. Becke, “Teand GeJ. Electron. Mater., vol. 4 , pp. Doping studies in Gal-,Al,As,” 101-118, Feb. 1975. F. D. King, A. J. SpringThorpe, and 0. I. Szentesi, “High-power long-lived double heterostructure LED’s for optical communications,” in Int. Electron Devices Meeting, Tech. Dig., Washington, DC, 1975, paper 21.2, p. 480. J. C. Dyment, A. J. SpringThorpe, F. D. King, and 5. Straus, “Proton bombarded double heterostructure LED’s,’’ J. Electron. Mater., vol. 6, no. 2, pp. 113-193, 1977. J. C. Dyment, F. P. Kapron, and A. J. SpringThorpe, “Top and side emission from double heterostructure LED’s and lasers,” IEEETrans.Electron Devices, vol. ED-24, pp. 995-1000, July 1977. N. Shimano, T. Yamamoto, T. Kawasaki, H. Takano,and Y. Morimoto, “Observation of internal interference in emission spectrum of the Burrus type GaAs-GaAlAs DH LED’s with thin active layer,” Japan. J. Appl. Phys., vol. 17, no. 1, pp. 237-238, 1918. D. Botez,“Analytical approximation of the radiationconfinement factor for the TEo mode of a double heterojunction laser,” IEEE J, Quantum Electron.,vol. QE-14, pp. 230-232, API. 1978. D. Botez, P. Zory, and M. J. Brady, “Simple expression for the light output of edgeemitting DHLED’s,’’ Electron. Lett., vol. 14, no. 22, pp. 716-718, Oct. 1978. Y. Horikoshi, Y. Takanashi, and G. Iwane, “High radiance lightemitting diodes,” Japan J. Appl.Phys., vol. 15, pp. 485-492, Mar. 1976. Y. Seki, “Light extraction efficiency of the LED with guide layers,” Japan J. Appl. Phys.,vol. 15, pp. 327-338, Feb. 1976. J. P. Wittke, “Spontaneousemission-rate alteration by dielectric and other waveguiding structures,” RCA Rev., vol. 36, pp. 655666, Dec. 1975. J. P. Wittke, M. Ettenberg, and H. Kressel, “High radiance LED for single-fiber optical links,” RCA Rev., vol. 37, pp. 159-183, June 1976. K. C. Kao and M. E. Collier, “Fiber optical system in future telecommunications networks,” Telecommun., p. 25,1977. D. Kato, “Light coupling from a stripe-geometry GaAs diode laser into an optical fiber with spherical end,” J. Appl. Phys., vol. 44, pp. 2756-2758, June 1973. D. Gloge, “LED design for fibre systems,” Electron. Lett. vol. 13, no. 4, pp. 399-400, July 1977. Y. S. Liu and D. A. Smith, “The frequency response of an amplitude-Modulated GaAs luminescence diode,” Proc. IEEE (Lett.), vol. 63, pp. 542-544, Mar. 1975. H. F. Lockwood, J. P. Wittke, and M. Ettenberg, “LED for high datarate,optical communications,” OpticsCommun., vol. 16, no. 1, pp. 193-196, Jan. 1976. M. Ettenberg and H. Kressel, “Interfacial recombination at AlGaAs/GaAs heterojunction structures,” J. Appl. Phys., vol. 47, pp. 1538-1544, Apr. 1976. H. Namizaki, H. Kan, M. Ishii, and A. Ito, “Current dependence of spontaneous lifetimes in GaAs-AlGaAs double heterostructure lasers,” Appl. Phys. Lett., vol. 24, pp. 486-487, May 1974. J. Zucker, R. B. Lauer, and J. Schlafer, “Response time of Ge-
XEEE TWANSCiCTIONS ON ELECTRON DEVICES, VOL. ED-26, NO. 8, AUGUST 1979
[31j  
doped (Al, Ga) As-GaAs double-heterostructure LED’s,” J. Ay>,7l. Phys., vol. 47, no. 5, pp. 2082-2084, May 1976. K. Ikeda, S. Horiuchi, T. Tanaka, and W. Susaki, “Design paran0eters of frequency response of GaAs-(Ga, A1)As double hete~ structure LED’s for optical communications,” IEEE Trans. Elc’etron Devices, vol. ED-24, pp. 1001-1005, July 1977. T. Yamaoka, M. Abe, and 0. Hasegawa, “GaAlAs LED’s for fibtzopticalcommunication systems,” Fujitsu Sci. Tech. J., vol. 14, no. 1, pp. 133-146, Mar. 1978. M. Honda, T. Kihara, and M. Endo, “Design and performance of an optical fiber digital link for widespread applications,”Fujitw Sei. Tech. J., vol. 14, no. 3, pp. 1-24,Sept. 1978. F. D. King, J. Straus, 0. I.Szentesi, and A. J. SpringThorpe, “High-radiance long-lived LED’s for analogue signalling,” Prc c.
Inst. Elec. Eng., vol. 123, no. 6, pp. 619-622, June 1976.  R. C. Goodfellow and A. W. Mabbitt, “Wide-bandwidth highradiance gallium-arsenide light-emitting diodes for fibre-optic communication,” Electron. Lett., vol. 12, no. 2, pp. 50-51, Jan. 1976.  J. Heinen, W. Huber, and W. Harth, “Light emitting diodes with a modulation bandwidth of more than 1 GHz,” Electron. Lett., vol. 12,110. 21,p. 553, 1976.  P. B. O’Connor, J. B. MacChesney, and C. M. Melliar-Smith, “Large-core high-N.A. fibres for data-link applications,” Electron. Lett., vol. 13, no. 7, pp. 170-171, Mar. 1977.  U. Daido, “The estimated coupling efficiency of a light emitting diode with large-core high N.A. fibers,” Fujitsu Sci. Tech. J., vol. 14, no. 3, pp. 25-36, Sept. 1978.
LED Pulse Response Analysis Considering the Distributed CR Constant in the Peripheral Junction
Absstruct-An improved equivalent circuit of a current crowding type LED is proposed to explain the LED pulse response behavior, takblg intoaccountthejunction capacitance andthe spreading resista1n’:e distributed over the entire junction area. Calculated results bawd on thismodel can consistentlyexplain the experimentally observ1:d LED behavior, includingdc-bias effects and patterneffects, which occllr in high-speed pulse modulation.
I. INTRODUCTION ECENT developementsinhigh-radiance and high-speed light-emitting diodes (LED’s) has made possible their 9)plication to high-bit-rate PCM opticalcommunications.For such applications, proper understanding of transient behavhr of an LED is needed. LED transient behavior is mainly go.(erned by junction space-charge capacitance and injectedcarrb r lifetime. A simple equivalent circuit, in which an LED is r(:placed by discrete circuit elements of a single junction including diffusion capacitance, space-charge capacitance, and norllinear resistance, was previously proposed to analyze the LEllIl transient behavior. By solving this equivalent circuit, ca1cu’l;Itions of the LED transient behavior were performed numer.cally  , [ 2 ] and also inclosed-form expressions . Tlhe
Manuscript received November 13, 1978. The authors are with Central Research Laboratories, Nippon Electr .c Company, Ltd., Takatsuku, Kawasaki, Japan.
pulse response of a large emitting area LED, whose emitting area is as large as an entire p-n junction, can beessentially understood by this model  . An LED for optical communications has a small emitting area in the center of a large p-n junction to obtain sufficient fiber coupling. In this structure, it does not seem reasonable to treat emittingarea junction and peripheral junction surrounding the emitting area as a singlebodyjunction. In fact,thetransient behavior of thistype LED, such as the turn-on delay and emission transient, cannot be fully explained by the calculations based on the simple discrete circuit model. In this paper, an improved equivalentcircuit of a current crowding type LED is proposed, taking into account the junctiondistributed capacitance connected by the spreadingresistance spread over the entire junction area. Simulation of the LED pulse response, based on this model, can consistently explain theexperimental results,including dc-bias effectsand pattern effects.
Fig. 1 shows the etched-well structure LED ,  to be investigated in this paper. double The heterostructure, in which a p-type active Al,,,sGao.95As layer is sandwiched by n-Al,,,,Ga,.,,As and p-A10,3sGao.6sAs cladding layers, is grown on an n-GaAs substrate. The small circular emitting re-
0018-9383/79/0800-1238$00.75 0 1979 IEEE
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.