Nanoscale Materials Characterization of Degradation in VCSELs

David T. Mathes∗ a, Robert Hullb, Kent Choquettec, Kent Geibd, Andy Allermand, Jim Guentera,Bobby Hawkinsa, Bobby Hawthornea

aVCSEL Optical Products, Honeywell Inc., 830 E. Arapaho Rd., Richardson, TX 75081;bDept. of Materials Science and Engineering, University of Virginia, 116 Engineer’s Way,

Charlottesville, VA 22904;cDept. of Electrical Engineering, University of Illinois at Urbana-Champaign, 208 North Wright St.,

Urbana, IL 61801;dCenter for Compound Semiconductor Science and Technology, Sandia National Labs, P.O. Box

5800 MS-0603 Albuquerque, NM 87185

ABSTRACT

Significant advancements have been made in the characterization and understanding of the degradation behavior of theIII-V semiconductor materials employed in Vertical Cavity Surface Emitting Laser (VCSEL) diodes. Briefly, for thefirst time a technique has been developed whereby it is possible to view the entire active region of a solid state laser in aTransmission Electron Microscope (TEM) using a novel Focussed Ion Beam (FIB) prepared plan-view samplegeometry. This technique, in conjunction with TEM cross-section imaging has enabled a three-dimensionalcharacterization of several of the degradation mechanisms that lead to laser failure. It is found that there may occur aninitial drop in laser power output due to the development of cracks in the upper mirror layers. In later stages ofdegradation, dislocations are punched out at stress-concentrating sites (e.g. oxide aperture tips) and these dislocationscan then extend over the active region in a manner consistent with recombination enhanced dislocation motion.Alternatively, complex three-dimensional dislocation arrays which exhibited dendritic-like growth and which cover theentire active region can nucleate on a single defect.Keywords: III-V material degradation, VCSEL, Vertical Cavity Surface Emitting Lasers, Nanoscale characterization,Failure analysis.

∗ david.mathes@honeywell.com; phone 1 972 470 4683

1. INTRODUCTION

Vertical Cavity Surface Emitting Lasers (VCSELs) offer numerous advantages over previous laser designs and have thusbecome a crucial component in the advancement of telecommunications and other laser-based technologies. However,to fully enable this technology, it is necessary to understand the fundamental materials science of the degradationprocesses that hinder the realization of long lifetimes for the arbitrary VCSEL design.

The necessity to study failure in VCSELs, rather than rely on previous failure analysis of edge emitting lasers, stemsfrom the differences in these laser designs and the overall greater complexity of the VCSEL design. Hence, it is possiblethat fundamentally different failure mechanisms may occur in VCSELs that will need to be addressed in order tofacilitate fabrication of commercially viable devices.

In general, well-made production-level VCSELs exhibit at least two failure modes. The VCSEL design’s “wear-out”failure mode controls the maximum potential lifetime and operates on the entire population. However, due to the factthat wear-out failure statistics are lognormally distributed, in most applications the failure rate due to this mechanism isnegligibly small during the first year(s) of operation. Instead, the wear-out rate is significant only after operationlifetimes on the order of ten to twenty years. Conversely, failures which occur after operation lifetimes on the order of

minutes to years are dominated by a second mode, which is actually a collection of defect-related failure modes,randomly distributed (or induced) in a small fraction of the population. While in a stable production process, failuresdue to this second mode typically represent only a few parts per million, they are, regrettably, the first failures acustomer will see. Furthermore, since electrical characterization of failures due either to pre-existing defects or tosubsequent exposure to damaging events may provide similar results, some other characterization technique is neededthat can unambiguously distinguish the possible failure mechanisms so that they can be eliminated. This paper describeswork done to develop and apply such a tool to discover and understand the prevalent material failure mechanisms thatare responsible for VCSEL failure.

2. EXPERIMENTAL

Several experimental methods have been developed in this work in order to elucidate the nature of VCSEL degradation.Of interest are, the locations of nucleation sites for defects, the final defect configuration within the VCSEL, the effect ofthe various components of the VCSEL (e.g. mirror layers or current confinement strategies) on defect generation, andthe nature of the defect growth with respect to current injection conditions.

The main body of this work is concerned with an extensive study of the failure mechanisms operative in one non-production oxide confined VCSEL device, referred to below as V1 (a 955nm device with GaAs/In0.14Ga0.86As quantumwells, grown by MOCVD). Many lasers of this configuration were operated for various durations at close to the rollover current (23mA, corresponding to 41kAcm-2) at room temperature. Device V1 was chosen due to it tendency torapidly fail (i.e. cease to lase on the time scale of seconds to minutes) under these operating conditions and so to providean extensive set of degraded structures. Even for these accelerated degradation conditions, it was necessary to choosedevice sets from within a single processed batch that exhibited uncharacteristically short lifetimes. The alternative wasto choose a commercially feasible device that rapidly failed only after a considerable stable lifetime, or to subject long-lived lasers to accelerated lifetime conditions (e.g. high temperature). Time constraints and lack of access to lifetimeaccelerating equipment precluded these latter two options. However, a few devices have been made available for thiswork which were failed under accelerated conditions (V2 A & B) or which failed after operation in the field forthousands of hours (V3 and V4). These devices help to add breadth to the results obtained from the main sample set V1,and aid us in understanding damage mechanisms in devices that degraded with more characteristic life times. Laser V2is an 850nm oxide-confined device with Al0.85Ga0.15As/Al0.15Ga0.85As quantum wells, degraded with 4.2kAcm-2 for 96hours. V3 is an 850nm proton-implant confined device using GaAs/Al0.15Ga0.85As quantum wells, which failed afteryears of use in the field, and V4 is an 850nm proton-implant confined device using GaAs/Al0.25Ga0.75As quantum wellsthat also failed after years of use in the field. Finally, catastrophic degradation in two oxide confined VCSELs (V5 andV6) was examined. V5 and V6 are 840nm and 850nm oxide confined devices respectively, employingGaAs/Al0.20Ga0.80As quantum wells.

To enable characterization of these devices, a sample preparation technique was devised for which the use of a FIB wascrucial. Of primary interest is the defect microstructure in the active region of the VCSEL. For this reason, a novel FIBtechnique is used, whereby, for the first time it is possible to reliably and consistently view the entire active region of asolid state laser in the TEM. Previous studies in this field have typically relied upon thinning small volumes of thestructure to electron transparency, raising extensive questions about the statistical significance of such small sampleregions, and the conclusions drawn therefrom. Using the present technique, the active region is isolated with the FIBi ina plan-view configuration as depicted in Figure 1 and Figure 2. It should be noted that a few mirror layers are left intacton either side of the membrane in order to prevent damage to the active region. Furthermore, as is seen in some of theplan-view data presented below, a slight unintentional angle between the Ga+ beam and the heteroepitaxial planes of theVCSEL causes the membrane surface to intersect several DBR layers as is shown in Figure 2. Plan-view samplesmanufactured under slight tilt will thus exhibit bands of contrast associated with the intersection of different DBR layerswith the membrane surface. The bands may be in the form of bars perpendicular to the beam direction, or if themembrane is slightly bent, there may instead be non-uniform regions of contrast variation.

i In this work, a FEI FIB200 employing a 30keV gallium ion source.

Figure 1. Schematic of the finished plan-view sample geometry. Figure 2. Schematic of the effect of Ga+ beam/heterointerfacemisorientation on the surface composition of the TEMmembrane.

Cross-section samples have also been important in characterizingdegraded structures. As in the planview samples, use of the FIBmakes possible the isolation of a membrane in the center of thedegraded VCSEL as depicted in Figure 3.

Thus, the use of the planview and cross section techniques together(e.g. one membrane of each type for two similarly degradedVCSELs) provides a clearer three-dimensional picture of anymaterial degradation present.

3. RESULTS AND DISCUSSION

3.1 Degradation in oxide confined VCSELsV1-A, degraded for 30 seconds at 23mA is shown in the plan-view TEM image in Figure 4 and the relevant powerversus time (P-t) curve is shown in Figure 5. Both oxide apertures (above and below the active region) are clearly seenin this image as squares surrounding the center of the device; i.e. the active region (the oxide aperture above the QWactive region extends further toward the center of the device than the oxide aperture below the QW region. Thus, thebottom oxide aperture is visible near the outer edge of the image). The large area of different contrast to the right of thefeature labeled ‘sample prep artifact’ is due to sample/Ga+ beam misorientation as described above. Further evidencethat this feature type is an artifact of sample preparation rather than a signature of material degradation induced duringdevice operation is the fact that similar features are also detected in plan-view samples of undegraded VCSELs.

At this stage of degradation, two types of structural degradation features appear, (although only to a relatively smallextent compared to data shown later for longer degradation times). A region of disruption to the DBR is visible in theupper right hand and left hand corners of the oxide aperture. Additional insight into this degradation feature will beshown via comparison of the cross-section and plan-view samples that follow. That the contrast features labeled “DBRdisruption” are structural and not artifacts of TEM imaging (e.g. not bend contours) is established by the fact that theyremain stationary during specimen tilting. Additionally, a small punched-out dislocation half-loop appears in the lowerleft hand corner of the oxide aperture. This loop is likely related to the small segment of DBR disruption visible in thatcorner of the device as is supported by additional evidence below. (The term “punched-out” dislocation refers todislocations that are apparently caused due to stresses associated with the oxide layers (i.e. in contrast to misfit orthreading dislocations). Hull and Bacon (reference 1, p. 165) use the term in describing dislocations punched out aroundan inclusion.)

Figure 3. Schematic of cross sectional samplegeometry. White lines depict lateral oxidelayers.

Figure 4. Plan-view TEM image of V1-A.

0.011

0.016

0.021

0.026

0.031

0.036

0.041

0.046

0.051

0 10 20 30 40

Time (sec)

Out

put P

ower

(mW

)

Figure 5. Power vs. time for operation at 23 mA for V1-A.

Finally, a long-range modulation feature with ~120nm periodicity is detected running horizontally in the active region(Figure 4). The fact that this feature runs perpendicular to the ion beam direction precludes its origin as a samplepreparation artifact. It is noted that in some ternary III-V alloys, including the InGaAs alloy system used in the activeregion of V1 (and more generally for the InGaAsP alloy system), compositional ordering or phase-separation is possibledepending on the growth conditions2-5 and that these compositional fluctuations may contribute midgap electrical statesthat contribute to non-radiative recombination6 or affect the optical properties of the material through spectralbroadening of the luminescence of the material4. However, stereomicroscopy indicates that the image modulation seenin Figure 4 occurs on the upper membrane surface (especially in the region labeled ‘sample preparation artifact’).Furthermore, Treacy et al.7 and others3,8 have shown how strain present in a TEM membrane can give rise to periodicbending of near-surface lattice planes, thus giving the impression of phase separation. These features are experimentallyand theoretically seen for sample thicknesses of up to ten times the extinction distance of the electron beam in thesample. Therefore, the long-range periodicity visible in the plan view TEM image may be more likely indicative ofstrain in the membrane and not a reflection of compositional modulation in the active region of the laser.

The presence of dislocations in the active regions of semiconductor lasers is particularly detrimental. Briefly,dislocations in the active region can nucleate Dark Line Defects (DLDs), which are comprised of dense dislocationarrays and which render the material region in which they exist (i.e. within the carrier diffusion length) incapable ofgenerating light due the tendency of carriers to non-radiatively recombine thereat. In addition, the growth of DLDs hasbeen attributed to this non-radiative recombination of electrical carriers. In a few words, if there exists an availableenergy state within the bandgap due to a defect, a charge carrier may be captured at the defect where it is then likely torelax via phonon production9 (i.e. non-radiatively) instead of by photon production. A collection of these phonons cancouple their energy into vibrational modes at the defect and thus enhance defect motion beyond thermally activateddefect migration rates10-12. Existing models predict that for a line dislocation, there is a greater chance that kinks willform and migrate under conditions of minority carrier injection via non-radiative recombination mechanisms (e.g.Radiation Enhanced Dislocation Glide (REDG)13). If there is a high density of point defects, the non-radiativerecombination of charge carriers assists in additional formation and migration of these defects, and consequentlyincreases the probability of non-conservative climb for segments of line dislocations (e.g. dipoles) 14,15. Thus, these non-radiative mechanisms are capable of increasing the defect density present, leading to a greater percentage of material that

cannot produce light; eventually rendering the laser inoperableii. It is thus extremely important to minimize theoccurrence of any defect in the as-grown laser. It is similarly important to minimize built-in epitaxial stress in the devicethat would magnify this process.

A greater extent of damage is observed in VCSEL V1-B, which has been degraded at 23mA for 240 seconds. As thecross section image shows in Figure 6 (the relevant P-t curve is given in Figure 7), there is cracking (i.e. two crackslabeled (a) and (b)) in the upper DBR. The nature of the crack is less certain for segments of the defect such as the areain Figure 6 marked (c) which do not exhibit void regions under the sample tilt used for the image shown. However, tiltscan be found for which all segments of the defect do exhibit a separation of the DBR crystal.

Figure 6. Cross-section bright field TEM image of V1-B.

0

0.01

0.02

0.03

0.04

0.05

0.06

0 50 100 150 200 250

Time (sec)

Pow

er O

utpu

t (m

W)

Figure 7. Power vs. time trace during degradation of V1-B at 23mA.

As seen in Figure 8, portions of the crack contain a nanocrystalline material (evidenced by the white speckle contrastwithin the crack in this dark field image). Electron diffraction analysis of this material has yielded a ring pattern withspacings consistent with the planar spacings of γ-Al2O3 as is shown in Figure 9. These oxide regions were also evidentin the FIB during TEM sample preparation. Thus the oxidation does not occur as a result of exposing the thin membraneto air (i.e. the FIB chamber is under a vacuum of 10-7 Torr). Undegraded devices did not exhibit cracks, indicating thatdevice processing (e.g. oxidation of current confinement layers or rapid thermal annealing) did not directly result inthese features. Thus, it appears that the cracking and development of the oxide bridges occurred during the degradationof the deviceiii. The source of oxygen for the oxide bridges to occur is possibly the oxide confinement layer, as all of theDBR cracks (for samples discussed later as well) are seen to terminate on the oxide layer. (It is possible that an excessH2O concentration is present in the oxide layer due to the fact that the oxidation process is terminated by removing thesample from the oxidation furnace. Thus, H2O molecules in transport16,17 through the oxide layer when the sample isremoved from the furnace may be quenched in). It is not likely that the development of the oxide bridges acts toseparate the surfaces of the crack as the oxide of AlxGa1-xAs will occupy a smaller volume than the crystal it consumes.Thus, it is likely that the cracks developed because of stresses associated with the oxide layer and that as the cracksformed, the high Al content layers of the DBR experienced conditions favorable to oxidation (e.g. heat from theoperating device and a source of water molecules. Regarding the former, active region temperatures in good devices canreach ~100ºC when operated near the roll-over current18. Poor heat sinking in this device may have contributed to yethigher temperatures).

ii In contrast, wear-out failure, which also renders a VCSEL useless, does not proceed via a dislocation growthmechanism.iii Similar crack-like features have been observed by researchers at SUNY Albany (unpublished results).

Figure 8. Dark field TEM image showing regions ofnanocrystalline oxide in a DBR crack.

Figure 9. Diffraction pattern from the oxide contained in thecrack, which has maxima consistent with γ-Al2O3.

VCSEL V1-C, degraded at 23mA for 800 seconds is shown in Figure 10 and the relevant P-t trace is given in Figure 11.This VCSEL exhibits punched-out dislocations and a dislocation array. The light stripe that transverses the imagehorizontally and the ribbon of speckle contrast are due to the FIB and are not features of degradation. Stereographicimaging shows that the punched out dislocations extend from the lower to upper membrane surface while the dislocationarray is confined to one plane of the membrane (the active region, as is shown later).

It should be noted in Figure 11 that a second serious reduction in output power occurs at ~700 seconds, whereas in thepreviously described sample, such a drop occurred near 200 seconds. It is shown below that there exists a correlationbetween the occurrence of a second power drop and the onset of the dislocation array growth. For example, in V1-C,given the relatively short amount of time the device was operated after the second power drop, the coverage of thedislocation dipole array is relatively small (< ½ of the active region), compared with samples degraded for longer timesafter the second power drop (that are discussed later).

Figure 10. Plan-view TEM image of degradation in V1-C.

00.010.020.030.040.050.060.070.080.090.1

0 200 400 600 800 1000

Time (sec)

Pow

er O

utpu

t (m

W)

Figure 11. Power vs. time during degradation of V1-C at 23 mA.

VCSEL V1-D, which has been degraded at 23mA for 1200 seconds, is shown in Figure 12 and the relevant P-t curve isshown in Figure 13. This sample exhibits disruption of the DBR layer, punched-out dislocations associated with theoxide aperture and coverage of the entire active region by a dislocation dipole/loop configuration. Accordingly, the P-tcurve exhibits a second drop at ~500 seconds and presumably, the remaining ~700 seconds of degradation weresufficient to generate the defect density seen throughout the active region.

Figure 12. Plan-view TEM image of degradation in V1-D.

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0 200 400 600 800 1000 1200

Time (sec)Po

wer

Out

put (

mW

)

Figure 13. Power vs. time during degradation of V1-D at 23 mA.

The corner of the active region containing the high density of punched-out dislocations is shown at higher magnificationin Figure 14. Segments of some of the punched-out dislocations undergo climb as evidenced by helical shapeddislocation segments1 (marked (b) in Figure 14). Furthermore, dislocation loops seen in Figure 12 and Figure 14 appearto be “self standing”; that is, without an apparent association to the punched-out dislocations. The fact that theobservable loops do not exhibit stacking fault contrast for low index diffraction conditions of the <110> zone axis (thecross section images) or the <001> zone axis (the plan-view images) indicates that they are perfect dislocation loops.

Figure 14. Loop/dipole array in active region of V1-D. Labeled are, (a) a screw segment as determined by dislocation contrast analysisand, (b) a helical dislocation segment.

The Burgers vector of the dislocations present in VCSEL V1-D was determined by standard contrast techniques. Of thelow index diffraction conditions checked (±400, ±040, ±220, ±-220 at the [001] zone axis and ±111, ±-31-1, and ±1-3-1at the [112] zone axis) only the ±400 and ±-31-1 two beam conditions give low (residual) contrast. Thus, the crossproduct of [-400] and [-31-1] gives the Burgers vector ±½a[0-11] which is inclined 45º from the plane of the activeregion.

A portion of the active region of VCSEL V1-E which was also degraded for 1200 seconds at 23mA is shown in Figure15 (the P-t trace is nearly identical to that shown in Figure 13 for V1-D). This sample exhibits the degradation featuresalready seen (e.g. oxide/crystal separation, DBR disruption in the form of cracking, punched-out dislocations extendinginto the upper and lower DBRs and a dislocation dipole/loop array in the active region). Again, the complete coverageof the active region by the dipole/loop array is consistent with continued degradation for a significant time (~700seconds) after the onset of a second power drop.

In V1-E, the vertical extent of the dislocation dipole/loops is seen to be limited to the active region. This spatialconfinement could be related to the fact that carrier recombination is largely limited to the active region of the device, ordue to the fact that the stress state of the DBR layers is not conducive to climb of the dislocations (which is consistentwith the higher compressive stress of the Al0.98Ga0.02As DBR layer above the GaAs active region if the climb is extrinsicin natureiv).

Figure 15. Cross-sectional bright field TEM image of V1-E.

Consideration of the results just presented (especially those for V1-D and V1-E) provides a clearer understanding of themechanisms operable in the failure of VCSEL V1. The power output and applied voltage are plotted as a function oftime for VCSEL V1-D in Figure 16. Due to the fact that the driving current is held constant (at 23 mA), an increase inapplied voltage represents an increase in device resistance (i.e. dynamic or series). Thus the initial increase in voltage isconsistent with a mechanism that increases the resistance of the VCSEL, possibly the current shunting mechanismwhereby current is forced to the periphery of the active region (due to the passivation of dopants by mobile pointdefects19 (e.g. hydrogen) in the p-n junction immediately adjacent to the active region20,21). The more severe fluctuationsin the voltage trace after ~100 seconds may be associated with the development of the cracks in the upper DBR. Finally,after ~400 seconds, the resistance of the device appears to decrease at the same time that the secondary drop inspontaneous emission occurs. Thus, it is likely that a low resistance current path or possibly the onset of a nkT current(due to non-radiative recombination on newly created dislocations) develops such that current flows again through theactive region (i.e. negating the current-shunting mechanism), which provides the energy source (i.e. by non-radiativerecombination) for the continued development of the dislocation array that grows across the active region.

iv At room temperature, the Al0.98Ga0.02As layer is under a compressive stress of ~0.14%.

That entire loops are confined to the thin active region (havinga [001] normal) and the fact that the Burgers vector is oriented45º from the plane of the active region indicates that thedislocations propagate by continually “reflecting” back to thequantum well region (for example, through a combination ofclimb, glide, and multiple cross slip events as is discussedshortly). Continuous glide or climb motion of the ±½a[0-11]dislocation (i.e. without cross-slip to motion on anotherallowed 111 plane) would not produce a dislocation patternhaving an average normal in the [001] direction (i.e.perpendicular to the plane of the active region). (For example,the pattern produced by continuous glide motion would havean average normal perpendicular to the ±½a[0-11] direction).This confining behavior suggests that the dislocationpropagation is dependent on the available energy due to non-radiative recombination occurring in the quantum well regionand that the dislocation segments preferentially propagatewithin a region where this source of activating energy exists.

The spatial arrangement of dislocations in VCSEL V1-D in the plane of the substrate is also significant. As shown inFigure 17, there appears to be a primary dislocation (traced in the image) that has moved through the active region,leaving dislocation loops in its wake. The apparent origin of the primary dislocation was a punched-out dislocation inthe upper left corner of the image of the active region presumably caused by stresses related to the oxide aperture (otherpunched-out dislocations have also begun to exhibit a jagged nature). There are other documented cases whereby aprimary dislocation gives rise to dislocation loops as it moves through material (e.g. the Orowan mechanism1). Thebasic mechanism is one in which a dislocation bows around an obstacle to the extent that the separated fronts meet andform a new dislocation front as well as a loop. While there are no obvious obstacles observed by TEM imaging in theactive region of V1-D, it appears that as fingers of the primary dislocation developed, some of them met, forming a newdislocation front and leaving “oxbow-lake” loops behind. Thus, a model is proposed, as explained below, whereby theonly obstacles leading to the development of the oxbow-lake loops are the heteroepitaxial interfaces bounding (and thosepossibly within) the active region of the device.

As was seen previously, the dislocations are generally contained within the inner edges of the DBRs. Thus the upperand lower bounds of the active region, acting as obstacles to the dislocation, may lead to the development of the oxbow-lake dislocations. A possible scenario by which this might occur is depicted in Figure 18. The schematic depicts fivesuccessive snapshots of a growing dislocation loop, although this proposed model works equally well for a longdislocation segment passing through the region depicted. The terrain shown is a random set of 111 planes, which arethe sets of planes on which dislocations in zinc-blende semiconductors glide. In the model, the dislocation is prohibitedfrom moving above or below the region depicted (the boundaries of which are marked by bold lines on the crests and inthe valleys) similar to the manner in which the dislocations under discussion are confined to the active region. As theloop expands, screw portions of the loop may cross slip onto other 111 planes containing the ½a<110> type Burgersvector (i.e. similar to the model in which multiple cross-slip occurs under conditions of recombination assisteddislocation motion as proposed by Matsui et. al.22). Upon reaching the upper or lower boundary of the region, edge ormixed character segments may be induced to switch from glide motion to climb motion (on a 111 plane not containingthe Burgers vector) or vice versa. Thus, as depicted, a series of these actions may produce a dislocation front with ajagged nature, some fingers of which can meet, annihilate and form oxbow-lake dislocation loops. While the residualstress state present in the VCSEL is initially complex and further complicated by the introduction of the dislocation arrayinto the active region, it must be assumed in applying this model to VCSEL V1-D that the average stress state is onewhich favors the occurrence of repeated cross-slip and glide-to-climb orientations such that the net dislocation motion isin the direction shown in Figure 17.

00.010.020.030.040.050.060.07

0 500 1000Time (sec)

Pow

er o

utpu

t (m

W)

3.323.333.343.353.363.373.383.39

Bia

s (V

)

Figure 16. Power output and applied voltage as afunction of time

Figure 17. 220 bright field TEM image of active region of V1-Dwith primary dislocation front traced. The arrow indicates theapproximate direction of dislocation motion.

Figure 18. Schematic of five successive iterations in thedislocation behavior leading to development of “oxbow lake”loops. In the schematic, a dislocation (b = ½a<110>) exists on a111 plane and is confined vertically (within the bold lines atthe crests and valleys shown). Through a series of glide, g,climb, c, and cross-slip motions, s, segments of opposite sign canmeet, forming the oxbow-lake loop.

Finally, Figure 19 shows a plot of the fraction of defect freeactive region of the V1 samples as a function of time after thesecondary power drop (e.g. as occurred at ~400 seconds forV1-D as shown in Figure 13). Here, the defect free activeregion is the area of the active region through which nodislocations have traversed. In areas where dislocations doexist the dislocation spacing is typically less than 1um(approximately the minority carrier diffusion length for III-Vmaterials23-26). Thus, luminescence from these areas will besignificantly decreased as carriers migrate to and non-radiatively recombine at the dislocations. This graph lendssupport to the argument that the secondary drop in powerindicates the onset of dislocation propagation throughout theactive region. Specifically, the two data points with a

dislocation free active region area of ~37µm2 are fromsamples V1-B and V1-C operated for a total time of 240seconds and 800 seconds respectively, thus illustrating a muchstronger correlation between the dislocation free active regionand the time after the second power drop than there is betweenthe dislocation free active region and the total operating time.As further evidence of this relationship, for V1 samplesshowing no secondary power drop, the active region remainsfree of dislocations.

3.1.1 Commercial oxide confined devices

As is seen in the cross-section and plan-view images of severely degraded oxide-confined VCSEL, V2 (Figure 20 andFigure 21 respectively), a dislocation dipole array extends well outside (~7um) the edge of the oxide aperture tip

Figure 19. Defect free active region area as a functionof time after the second power drop. The insetrepresents one of the P-t traces (from Figure 13)showing a secondary drop in optical power and thearrow indicates the relevant data point.

indicating that a non-radiative recombination assisteddislocation growth mechanism is operative away from theprincipal current path. It is generally accepted that bothoptical and electrical fluxes exist outside the oxide aperturewithin a few microns due to lateral carrier diffusion(depending on the carrier concentration gradient, carrierpopulation, carrier diffusion length and device architecture27)or due to the active region having a resistance greater than themirror layers (i.e. leading to a lower resistance current pathparallel to and above the active region instead of through it28-

30). However, as is suggested by VCSEL V2, current may beforced to the peripheries of the active region, leading torecombination-assisted dislocation growth there, if theresistance through the dislocation network is relatively high.This may occur as the average dislocation spacing falls belowtwice the minority carrier diffusion length at which pointfurther increases in dislocation density do not increase therecombination rate but instead serve only to remove area fromthe current path, leading to a higher resistance.

3.2 Degradation in Proton Implanted devices

An example of degradation in a proton-implant confinedVCSEL is shown in Figure 22. A large dipole array hasapparently nucleated from a short segment (~6µm) of a lineardislocation that runs through the active region. The majorityof the arms of the dipole array can be traced back to thenucleation of a single dipole as is demonstrated by the jaggedtraced line in Figure 23. Other segments of the dipole arraycross the linear dislocation but do not appear to haveinteracted with it, which suggests that the dislocations exist atdifferent depths in the active region (further evidence of this isgiven shortly). The circular confinement of the dense array ofdislocations is apparently a reflection of the circular protonimplant aperture used. Segments of the dipoles areapproximately aligned along <100> directions in the plane ofthe active region which is consistent with the preferredorientation of the climb based DLDs as seen in edge emittinglasers.

A portion of the dipole array (the upper ~5um of the regionenclosed by the line in Figure 23 and shown at highermagnification in Figure 24) exhibits an array of stand-alonedislocation loops in an arrangement similar to thecharacteristic pattern of dislocation arms in the remainder ofthe array. While it is thus likely that these loops in device V3 were once connected to form dipoles in a pattern similarto that seen in the remainder of the loop, it appears possible that the formation of the loops represents a trend towards alower energy state (implying that the dipole configuration represents a non-equilibrium condition) and/or that some ofthe point defects involved in the formation of the original dipoles may have been drawn to the tips of the other, morerecently grown dipoles leading to the formation of these loops. Similar behavior was reported in the literature in a TEMsample containing a DLD type dislocation array whereupon by heating the sample to 440ºC for an extended period,some loops and dipoles grew at the expense of others31. Furthermore, while not demonstrated here, heating a sample that

Figure 21. TEM plan-view image of V2-B.

Figure 20. TEM cross-section image of V2-A.

exhibited a similar dislocation pattern to >600ºC for 20 minutes caused the long dipole arms to coalesce into dislocationloops indicating that the loops represent a trend toward an equilibrium distribution of dislocations.

Figure 22. Bright field TEM image of the dislocationdipole array in VCSEL V3.

Figure 23. Plan-view TEM image of degradation inVCSEL V3. The enclosed array is from an earlier growth.The jagged traced line demonstrates that the end of anydislocation arm can be traced back to a single nucleationsite.

Stereomicroscopy showed that in contrast to the dislocationarray seen in VCSEL V1, the arms of dislocations in VCSELV3 appear to have a component in the vertical direction (withrespect to the plane of the active region). Consideration of theBurgers vector of the dislocation array in this sample(determined to be ±½a[0-11] using standard diffraction contrasttechniques) and the three-dimensional data fromstereomicroscopy indicated that the extension of the dislocationarms was by climb.

A second possible mechanism by which dislocation arrays formapparently via a recombination-assisted mechanism is depictedin Figure 25. In this sample a dense cluster of lineardislocations extends from the lower to upper DBR (and likelyfrom the substrate to the surface of the VCSEL although this isnot verifiable in light of the fact that this portion of the device isremoved in sample preparation). The vertical cluster ofdislocation loops appears to have generated a set of glidev

dislocation half-loops that extend perpendicular to the cluster. On several of these glide dislocations, dipoles haveformed that extend approximately in the [010] direction. A fortunate consequence of the growth of the defects in this

v The assignment of glide nature to the dislocations seen here is due to the fact that glide dislocations are typicallyrelatively straight. Conversely, climb dislocations typically exhibit extensions with a relatively high radius of curvature(e.g. dipoles) or a wavy nature (e.g. from the two-dimensional projection of a helical dislocation).

Figure 24. Nucleation of dipole array in VCSEL V3.

sample is that it elucidates, in a series of snapshots, a mechanism of formation of dipoles and subsequent loops fromglide dislocations. Specifically, on the smallest glide dislocation, labeled (a), there is a small asperity in the upper leftleading edge of the dislocation, which presumably is the start of a dipole. On the second and third glide dislocations, (b)and (c), the dipole is increasingly extended while the base is increasingly pinched off. Finally, on the fourth glidedislocation, (d), the dipole is completely pinched off leaving a dislocation loop behind. It is also likely that the glidedislocations continued to move after the loop pinched off as is seen in the fact that while dipoles (a)-(c) occur on theleading upper edge of the glide dislocation, in dislocation (d), the upper leading edge has progressed past the bottom ofthe loop. Thus, dislocation loops may be formed in the active region by a gliding dislocation. Subsequently, the glidedislocation may continue to move leaving freestanding dislocation loops. The source of the point defects necessary tocreate the dislocation loops may be generated at the glidingdislocation from non-radiative recombination32 or from possiblecompositional irregularities at the numerous heterointerfaces inthe active region, both of which are point defect sources thathave been reported in the literature33-37.

Finally, there are no obvious crystalline defects that wouldindicate a clear reason for dipole growth to occur in theorientation seen (e.g. akin to a drag point or other obstacle thatone would expect to influence a gliding dislocation). However,dipole extension is favored in <100> directions33. Thus, thecombination of, (i) a source of point defects, (ii) minoritycarriers from the operation of the laser, (iii) a dislocation uponwhich to nucleate (and where non-radiative recombinationlikely occurs), (iv) the preferred crystallographic growthorientation and, (v) a stress field capable of driving the growth,are believed to be sufficient to cause the growth of the dipoles inthe manner seen in this sample.

3.3 Catastrophic failure

Devices subjected to unintentional high current pulses such thatoutput powers reach 106W/cm2 typically degradecatastrophically due to the extreme optical density and/orheating. Alternatively, if a device is operated under constantpower conditions (i.e. such that current varies to keep the outputpower constant), current densities near the end of the VCSELlifetime increase sharply as the efficiency decreases, and theVCSEL can fail due to electrical overstress38,39.

A case of catastrophic failure is shown in cross section forVCSEL V5 in Figure 26. Whereas catastrophic failure oftenoccurs at the mirror surface in an edge-emitting laser, the focalpoint of damage in V5 appears to be the active region. Featuresof the catastrophic damage include melting and void formation at the active region and a high density of dislocationtangles and stacking faults that appear to have grown semi-radially from the active region.

Catastrophic damage in V6 is shown in plan-view in Figure 27. Voiding has occurred which is apparently associatedwith melted portions of the active region. Arms of darker contrast are apparently radial extensions of molten zones fromthe center of the active region. Finally, a network of dislocation tangles also exists around the catastrophic damage inthe center of the active region presumably due to stresses generated during melting and recrystallization.

Thus, catastrophic failure in VCSELs appears to proceed differently than is typical for edge emitting lasers, presumablydue to the very different device geometries. Specifically, catastrophic failure in edge emitting lasers usually begins onthe mirror surface or at an inclusion in the active region of the laser. For the two VCSELs just shown, it is evident that

Figure 25. Mechanism of loop nucleation in VCSELV4.

Figure 26. Cross section TEM image of catastrophicdamage in VCSEL V5.

failure has begun in the center of the device and propagatedoutward. Furthermore, catastrophic failure in part relies onthermal runaway from intense non-radiative recombination (i.e. aself-feeding mechanism in which the defect created generatesmore heat via non-radiative recombination which propagates thedefect further). That the failure begins at the center of theVCSEL is consistent with the confinement of carrierrecombination to the active region (in the pre-degraded VCSEL).Furthermore, the catastrophic failure mechanism may befacilitated by the relatively higher series resistance of the typicalVCSEL design (~20Ω compared to ~2Ω for an edge emittinglaser) and thus greater degree of joule heating. Lastly, the factthat the optical and electrical field intensities are greatest at thecenter of the device (due to the high reflectivity of the DBRmirrors and to current confinement techniques respectively) arefurther reasons for catastrophic failure to begin at the center of thedevice. Therefore, following the models given for catastrophicfailure in edge emitting lasers, the catastrophic failure of theseVCSELs likely proceeds as follows. A preexisting defect or adefect such as a dislocation array occurs in the active region.Intense non-radiative recombination (due to greater than normaldriving conditions such as a current spike or due to a rise in thedrive current as the laser degrades under constant power operation) occurs at this defect leading to the melting of thenearby material. Intense non-radiative recombination continues on the molten material further propagating the moltenarm. A tangle of dislocations is formed to compensate for the stresses involved in the melting and solidification of thesearms.

4. CONCLUSIONSSeveral new and important results have been discussed in this paper. For the first time a technique has been developedwhereby it is possible to view the entire active region of a solid state laser in a TEM and thereby conclusively identifythe degradation mechanisms that lead to laser failure. These techniques have been systematically applied to thecharacterization of a single, rapidly failing VCSEL device structure thus showing the evolution of material failure in thedevice and ultimately linking the material failure to the output characteristics of the device. Additional examples ofdegraded VCSELs, drawn from a wider set of failure conditions, were examined to give breadth to our understanding ofVCSEL failure mechanisms.

In non-production VCSEL V1, the degradation proceeds as follows. There is an initial drop in laser output due to thedevelopment of cracks in the upper DBR layers (leading to a loss of the resonance cavity) and also, possibly due to amechanism such as current shunting. Although this latter mechanism is not readily detected by TEM, it appears to be alikely cause of degradation in the absence of damage to the DBR or a dislocation network in the active region. The forcedriving crack propagation is not fully understood at this time. However, the oxide layers do seem to be involved in thecrack formation (all of the observed cracks ended on the oxide layers).

The damage to the DBR is typically connected with separation of the upper DBR from the oxide layers and in someinstances extends over the entire active region. In some cases, dislocations are punched out near the oxide aperturepresumably due to stresses associated with damage to the DBR layers. That these dislocations are a result of the damageto the DBR layers (as opposed to the cause of it) is seen by the fact that several of the V1 devices (not shown in thispaper) exhibited cracking but no dislocations while the opposite scenario (i.e. dislocations without cracking) was notobserved. Some of these punched-out dislocations then extend over the active region, presumably by a combinedprocess of climb and glide that keeps them largely confined to the active region. The controlling parameter is likely non-radiative recombination in the active region that supplies energy for the evolution of the dislocation array. The Burgersvector of the dislocations in VCSEL V1 is ½a[0-11], which is inclined 45º to the plane of the laser substrate and the type

Figure 27. Catastrophic damage in V6.

of which is consistent with the Burgers vector for dislocations in III-V semiconductors. Freestanding dislocation loopswhich are left in the active region are explained by a process in which segments of the jagged dislocation front meet andform a new front leaving a loop behind. The jagged nature of the dislocation front can be explained by a combination ofcross-slip and fluctuation between climb and glide due to confinement of the dislocation motion to the active region.Furthermore, it has been shown how the beginning and extent of active region coverage by the dislocation array can bedetermined by the onset and duration of time after a secondary drop in the power versus time curve.

Two additional mechanisms for the formation of stand-alone dislocation loops seen in the active regions of failed lasersare also described. In the first, the individual extended arms of a dislocation array pinch off to form loops. In thesecond, a gliding dislocation grows a dislocation dipole that eventually pinches off to form a free standing loop. Theloops formed from these two mechanisms, as well as those seen in VCSEL V1 do not exhibit stacking faults asevidenced by the lack of stacking fault image contrast for all of the several diffraction conditions employed.

Some of the VCSELs studied (especially V3 and V4) demonstrate the role that preexisting defects play in laser failure.Specifically, in these devices, complex dislocation arrays were shown to nucleate on a single grown-in defect. Thus,while the dislocation arrays lead to the loss of laser output, the initial preexisting defect was ultimately responsible forthe laser failure.

Finally, the study of catastrophically failed VCSELs shown in this paper demonstrates that several of the degradationmechanisms that occur during the catastrophic failure of edge emitting lasers are also operative in VCSELs. Namely,laser material is melted and resolidified causing the formation of a tangle of dislocations. However, catastrophic failurein VCSELs is seen to originate at the active region in the center of the device while it was reported to occur on mirrorfaces in edge emitting lasers. Presumably the cause for this difference is the carrier recombination that occurs at theactive region in the center of the VCSEL.

5. ACKNOWLEDGEMENTS

This work has been supported by Sandia National Laboratory and by the National Science Foundation, Division ofMaterials Research, under grant numbers 9612283 and 0080630.

REFERENCES 1 D. Hull, and D.J. Bacon, Introduction to Dislocations, 3rd ed., Butterworth Heinemann, UK, 1984.2 G. Stringfellow, “Miscibility gaps in quaternary III/V alloys”, J. Crystal Growth, 58, 194-202, 1982.3 T.L. Mcdevitt, S. Mahajan, D.E. Laughlin, W.A. Bonner and V.G. Keramidas, “Two-dimensional phase separation inInxGa1-xAsyP1-y epitaxial layers”, Phys. Rev. B, 45(12), 6614-6622, 1992.4 D. Gonzalez, G. Aragon, D. Araujo, M.J. de Castro, and R. Garcia, “Growth rate and critical temperatures to avoid themodulation of composition of InGaAs epitaxial layers”, Appl. Phys. Lett., 74(18), 2649-2651, 1999.5 D. Gonzalez, G. Aragon, D. Araujo, and R. Garcia, “Control of phase modulation in InGaAs epilayers”, Appl. Phys.Lett., 76(22), 3236-3238, 2000.6 P. Ernst, C. Geng, F. Scholz, and H. Schweizer, “Ordering in GaInP2 studied by optical spectroscopy”, Phys. Stat. Sol.(B), 193, 213-228, 1996.7 M. Treacy, J.M. Gibson, and A. Howie, “On elastic relaxation and long wavelength microstructures in spinodallydecomposed InxGa1-xAsyP1-y epitaxial layers”, Phil. Mag. A, 51(3), 389-417, 1985.8 S. Mahajan, M. Shahid, and D. Laughlin, “Current status of atomic ordering and phase separation in ternary andquaternary III-V compound semiconductors”, Microscopy of Semiconducting Materials, Inst. Phys. Conf. Ser. No. 100sect. 3, 143-153, IOP Publishing Ltd. 1989.9 V. Narayanamurti, R.A. Logan, and M.A. Chin, “Direct observation of phonons generated during nonradiative capturein GaAs p-n junctions”, Phys. Rev. Lett., 40 63-66, 1978.10 R.D. Gold and L.R. Weisberg, “Permanent degradation of GaAs tunnel diodes”, Solid State Electron, 7, 811-821,1964.11 D.V. Lang and L.C. Kimerling, “Observation of recombination-enhanced defect reactions in semiconductors”, Phys.Rev. Lett., 33(8), 489-492, 1974.

12 D.V. Lang and C.H. Henry, “Nonradiative recombination at deep levels in GaAs and GaP by Lattice-RelaxationMultiphonon Emission”, Phys. Rev. Lett., 35(22) 1525-1528, 1975.13 P. Haasen, “Kink formation and migration as dependent on the Fermi level”, J. De Physique, 40(C6) 111-116, 1979.14 L.C. Kimerling, P.Petroff, and H.J. Leamy, “Injection-stimulated dislocation motion in semiconductors”, Appl. Phys.Lett., 28(6), 297-300, 1976.15 P. Petroff and R.L. Hartman, “Defect structure introduced during operation of heterojunction GaAs lasers”, Appl.Phys. Lett., 23(8), 469-471, 1973.16 C.H. Ashby, M.M. bridges, A.A. Allerman, B.E. Hammons, H.Q Hou, “Origin of the time dependence of wetoxidation of AlGaAs”, Appl. Phys. Lett,. 75(1), 73-75, 1999.17 S. Guha, F. Agahi, B. Pezeshki, J. A. Kish, D.W. Kisker, and N.A. Bojarczuk, “Microstructure of AlGaAs-oxideheterolayers formed by wet oxidation”, Appl. Phys. Lett., 68(7), 906-908, 1996.18 C. Wilmsen, H. Temkin, and L.A. Coldren, Vertical-Cavity Surface-Emitting Lasers, Cambridge University Press,Cambridge, UK, 1999.19 R.B. Martins, O. Henoc, B. Akamatsu, and J.F. Palmier, “A model for the degradation of Ga(Al)As single-quantum-well lasers”, J. Appl. Phys., 70(2), 554-561, 1991.20 R.W. Herrick, Degradation in Vertical Cavity Lasers, an Electrical Engineering Ph.D. Thesis, University of Californiaat Santa Barbara, 1997.21 R.W. Herrick, and P.M. Petroff, “Gradual degradation in 850-nm vertical-cavity surface-emitting lasers”, IEEE J.Quantum Electronics, 34(10), 1963-1969, 1998.22 J. Matsui, K. Ishida, and Y. Nannichi, “Rapid degradation in double-heterostructure lasers. I. Proposal of a new modelfor the directional growth of dislocation networks”, Jap. J Appl. Phys., 14(10), 1555-1560, 1975.23 P. Berger, N. Chand, and N. Dutta, “An AlGaAs double-heterojunction bipolar transistor grown by molecular-beamepitaxy”, Appl. Phys. Lett., 59(9), 1099-1101, 1991.24 G. Grummt, G. Oelgart, and N. Puhlmann, “Transport length variation due to dopant gradients in AlGaAs”,Semiconductor Science & Technology, 8(7), 1191-1195, 1993.25 D. Araujo, G. Oelgart, J. Ganiere, and F. Reinhart, “Comparison of the lateral carrier transport between a GaAs singlequantum well and the AlGaAs barrier during cathodoluminescence excitation”, J. Appl. Phys., 76(1), 342-346, 1994.26 H. Zarem, J. Lebens, K. Nordstrom, P. Sercel, S. Sanders, L. Eng, A. Yariv, and K. Vahala, “Effect of Al molefraction on carrier diffusion lengths and lifetimes in AlxGA1-xs”, Appl. Phys. Lett., 55(25), 2622-2624, 1989.27 W. Tsang, “The effects of lateral current spreading, carrier out-diffusion, and optical mode losses on the thresholdcurrent density of GaAs-AlxGa1-xAs stripe-geometry DH lasers”, J. Appl. Phys., 49(3), 1031-1044, 1978.28 Personal communication with Kent Choquette, University of Illinois, Urbana Champaign.29 B. Thibeault, E. Hegblom, Y. Akulova, J. Ko, R. Naone, L. Coldren, and P. Floyd, “Electrical and optical losses indielectrically apertured vertical cavity lasers”, Vertical Cavity Surface Emitting Lasers, 3003, 86-99, 1997.30 W. Joyce, “Current-crowded carrier confinement in double-heterostructure lasers”, J. Appl. Phys., 51(5), 2394-2401,1980.31 P.W. Hutchinson, P.S. Dobson, S. O’Hara, and G.H. Newman, “Defect structure of degraded heterojunction GaAlAs-GaAs lasers”, Appl. Phys. Lett., 26(5), 250-252, 1975.32 O’Hara, P.W. Hutchinson, and P.S. Dobson, “The origin on dislocation climb during laser operation”, Appl. Phys.Lett., 30(8), 368-371, 1977.33 P. Petroff and R.L. Hartman, “Rapid degradation phenomenon in heterojuction GaAlAs-GaAs lasers”, J. Appl. Phys.,45(9), 3899-3903, 1974.34 P. Petroff, D. Johnston, Jr., and R.L. Hartman, “Nature of optically induced defects in Ga1-xAlxAs-GaAs double-heterojunction laser structures”, Appl. Phys. Lett., 25(4), 226-228, 1974.35 E.S. Yang, “Formation of interface states and defects in GaAs-Al,Ga1-xAs DH lasers under room-temperature cwoperation”, J. Appl. Phys., 45, 3801-3805, 1974.36 G.M. Blom, “Native defects and stoichiometry in GaAlAs”, J. Cryst. Growth, 36, 125-137, 1976.37 M.M. Sobolev, A.V. Gittsovich, M.I. Papenstev, I.V. Kochnev, and B.S. Yavich, “Degradation mechanism ofGaAs/AlGaAs quantum well laser”, Sov. Phys. Semicond., 26(10), 985-989, 1992.38 Personal communication with Kent Choquette, Department of Electrical Engineering, University of Illinois.39 Personal communication with James Guenter, Honeywell VCSEL Products Division, Richardson TX.