Vertical Light-Emitting Diode Fabrication by Controlled Spalling

Stephen W. Bedell�, Can Bayram, Keith Fogel, Paul Lauro, Jonathan Kiser, John Ott, Yu Zhu, and Devendra Sadana

IBM T. J. Watson Research Center, Yorktown Heights, NY 10598, U.S.A.

E-mail: bedells@us.ibm.com

Received July 30, 2013; accepted October 1, 2013; published online October 18, 2013

A fracture-based layer transfer technique referred to as controlled spalling was used to separate a conventional InGaN/GaN multiple quantum

well light-emitting diode (LED) structure from a 50mm sapphire wafer enabling the formation of vertical spalled LEDs (SLEDs). A 25-�m-thick

tensile Ni layer was electrodeposited on the surface of the wafer, followed by the application of a polyimide tape layer. By mechanically guiding the

tape layer, a 3-�m-thick layer of the LED epitaxy was removed. Transmission electron microscopy imaging indicated that spalling preserved the

quality of the epitaxial layers, and electroluminescence verifies the operation of the SLED. # 2013 The Japan Society of Applied Physics

Dramatic advances have been made in solid statelighting within the past few decades.1) In additionto improvements in the starting material quality and

light extraction techniques, thermal management has playedan important role specifically in the development of high-power light emitting diode (LED) lighting.2) Sapphire orpatterned sapphire substrates (PSS) are used extensively asepitaxial growth templates for GaN/InGaN based materials.Although relatively cost-effective, the low thermal con-ductivity of sapphire restricts heat flow from the devicethereby limiting the maximum operating power. The super-ior heat conductivity of silicon carbide has made it thesubstrate of choice for high-power applications; however,the cost is significantly higher as well.

The concept of a vertical LED (VLED) structure improvesmany aspects of these devices including current spreadingand thermal management. In order to fabricate a VLEDdevice, the epitaxial layers must be removed from thesapphire substrate. There are essentially three different waysthis has been accomplished: laser liftoff (LLO), chemicalliftoff (CLO), and backside grinding of the sapphire. In laserliftoff,3–5) an ultraviolet excimer laser is projected throughthe backside of the sapphire wafer and results in decom-position of the GaN at the GaN/Al2O3 interface. In chemicalliftoff, a specific layer within the epitaxial stack is pre-ferentially etched and subsequently releases the overlyingdevice layers. This has been demonstrated using selectiveetching of layers composed of CrN,6) GaN:Si,7) ZnO8) orusing photoelectrochemical (PEC) etching techniques.9) Thedrawback to laser liftoff is that damage to the GaN layerscan occur during irradiation resulting in reduced deviceperformance and yield.10) The chemical liftoff processesthat incorporate new materials into the epitaxial stack arefaced with the challenge of maintaining high-quality GaNgrowth, and large-area CLO processes are often difficult toimplement in practice.

We have reported previously on a layer transfer techniquereferred to as controlled spalling.11,12) This approach usesa stressed layer deposited on the surface of a substrate toinduce lateral fracture within the underlying substrate and‘‘peel off’’ a controlled thickness from the surface. We haveapplied controlled spalling to the fabrication of flexible,high-efficiency GaAs-based photovoltaics13) as well as high-performance nanoscale Si circuits14) and in all cases thedefectivity of the material is unchanged by this process.In this work, we use controlled spalling to remove multi-quantum well (MQW)-based InGaN/GaN epitaxial LEDlayers grown on PSS substrates and demonstrate successful

optical emission from the resulting vertical spalled LED(SLED) layers.

The physical origin of spalling mode fracture comes fromthe combination of opening stress (mode I) and shear stress(mode II) due to the presence of a surface stressor layeracting on a subsurface crack tip.15,16) For a tensile surfacestressor layer, a crack will propagate downward into thesubstrate to a depth where the shear component is zero. If thecrack continues below this point, it will be deflected backtowards the surface by the change in sign of the shear stresscomponent. Therefore, the fracture depth is stable within thesubstrate as the crack tip trajectory oscillates about the zeroshear position due to the corrective action of the shear field.For fracture to be energetically possible within the substrate,the magnitude of the opening mode stress field (given by thestress intensity factor KI) must be equal to, or greater than,the material fracture toughness KIC at the zero shear position(KII ¼ 0). For a particular choice of stressor material (e.g.,Ni) there will be a combination of stress and thickness abovewhich fracture is possible for a given substrate type (usingKIC from literature17)). In order to apply the controlledspalling process to the epitaxial GaN material system usingNi as the stressor layer, the required loading conditions(stress and thickness) were determined experimentally.If the loading conditions are too high, spalling can occurspontaneously during Ni deposition and generally leads tounusable layers. Therefore, it is important to determinethe combinations of stress and thickness that energeticallypermit fracture, but do not lead to spontaneous spalling.

PSS substrates (c-plane), 50mm in diameter, were usedas growth templates for InxGað1�xÞN/GaN MQW LEDstructures using a commercial metal–organic chemical vapordeposition (MOCVD) reactor. The epitaxial structure con-sisted of a thin low-temperature GaN buffer layer, followedby a 2-�m-thick undoped GaN layer (u-GaN), a 3-�m-thickSi doped n-GaN layer, InGaN/GaN MQW layers, and Mg-doped p-GaN capping layers. After growth, the wafers wereannealed in a N2 ambient to activate the p-GaN layer.

Figure 1 shows a schematic of the controlled spallingprocess used in this work. A Ni stressor layer was depositedonto the surface of the wafers by first depositing a thin(75 nm) adhesion layer of Ti followed by 500 nm of Niby DC magnetron sputtering. Electroplating using a NiCl2-based solution at room temperature resulted in approxi-mately 400MPa tensile stress as measured by wafer bowing.We then determined that approximately 25 �m of Ni depo-sited onto the sputtered Ni was required to permit controlledspalling of the GaN. After deposition of the thick Ni layer,

Applied Physics Express 6 (2013) 112301

112301-1 # 2013 The Japan Society of Applied Physics

http://dx.doi.org/10.7567/APEX.6.112301

the wafers were rinsed and dried and held flat using acustom-made vacuum chuck table with an integrated tapedispenser. A layer of polyimide tape (25 �m) was roll-applied to the surface of the Ni, and fracture occurred upongently pulling the tape away from the wafer surface, asillustrated in Fig. 1(b). To prevent excessive bending ofthe thin film after spalling, the outer portions of the tapeare used to secure the sample to a stainless steel frame. Theentire process, from metal deposition to layer removal, wasperformed at room temperature. Figure 2 shows an entire50-mm-diameter InGaN/GaN epitaxial structure removedfrom the sapphire wafer by controlled spalling and mountedin a handling frame.

The surface artifacts observed on the sample shownin Fig. 2 are due to roughness caused by fracture depthoscillations around the equilibrium spalling depth and arenot cracks in the film. Surface profilometry was performedand the results are shown in Fig. 3. Although the resultingsurface roughness after spalling varies with length scale, themaximum amplitude occurs with a spatial period of about700 �m and an amplitude of about 70 nm. Fracture prop-agation speed directly affects this spatial period, with fasterfracture resulting in a longer oscillation period. In this work,fracture propagation across the 50mm wafer surface tookon the order of one second leading to a fracture velocity ofabout 50mm/s. In the case of LED structure spalling, thisroughness may offer a natural means of surface texturing forimproved light extraction.

Figure 4 shows a cross-sectional scanning electron mi-croscopy (XSEM) image of the InGaN/GaN LED structurebefore [Fig. 4(a)] and after [Fig. 4(b)] controlled spalling.Coloration has been added to the figure to indicate the variousregions of the epitaxial structure. By comparison of the two

images it is evident that fracture propagated approximately3 �m below the surface within the n-GaN region. This has thetwofold benefit of having immediate access to the n-typecontact area without etching, and the possibility of reusingnot only the original sapphire wafer, but also the u-GaNbuffer region as well. Spalling depth is generally controlledby the Ni thickness. We have observed a spalling depthchange of approximately 500 nm per micron of Ni thick-ness change for the GaN on PSS system. Also, due to theextremely high fracture toughness of sapphire, GaN epitaxylayers could be removed entirely from non-PSS (planar)sapphire substrates. Figure 5 shows cross-sectional trans-mission electron microscopy (XTEM) images of the SLEDstructure. The low magnification image [Fig. 5(a)] shows thefull thickness of the SLED structure indicating a total layerthickness of about 3.1 �m and no spalling-related defectivitywas observed in the sample. Figures 5(b) and 5(c) showhigh-resolution imaging of the InGaN/GaN MQW structureand the 10-period active layer of the device, respectively.

Figure 6 shows the electroluminescence spectrum takenfrom direct probing (no metallization) of the exposed n-GaN

Sapphire

GaN EpitaxyNi (tensile)

Tape

Controlled Spalling

(a)

(b)

Fig. 1. Schematic illustration of the controlled spalling process used to

remove InGaN/GaN MQW LED layers from a sapphire substrate. The

GaN/Al2O3 substrate was prepared for spalling by (a) first depositing a thin

Ti (75 nm)/Ni (500 nm) seed layer by sputtering, followed by

electrodeposition of a 25-�m-thick Ni layer (400MPa tensile), then

application of a tape layer. Removal of the epitaxial layers from the sapphire

substrate was performed by (b) peeling the tape layer away from the surface.

Fig. 2. Image of a spalled 50-mm-diameter InGaN/GaN MQW LED

structure (SLED) mounted onto a stainless steel frame using the handling

tape.

-200

-150

-100

-50

0

50

100

150

200

0 500 1000 1500 2000 2500 3000Distance (μm)

Hei

gh

t (n

m)

Fig. 3. Surface profilometry data showing the characteristic roughness

associated with the as-fractured surface after spalling. The largest amplitude

is seen at a spatial period of about 700 �m with a magnitude of about 70 nm.

S. W. Bedell et al.Appl. Phys. Express 6 (2013) 112301

112301-2 # 2013 The Japan Society of Applied Physics

surface of the vertical SLED (inset), as well as a con-ventional device formed on a small non-spalled portion fromthe edge of the same wafer (Bulk). The optical emissionfrom the SLED films was inhomogeneous, as seen in theimage (Fig. 6 inset). This is because the Ti adhesion layerformed a rectifying contact with the p-GaN surface and con-

duction during forward bias occurred primarily at localizedbreakdown regions. The peak emission �p was measured tobe 529.2 nm from the SLED device and 533.9 for the bulkdevice (with annealed ohmic NiAu contacts). This corre-sponds to an energy shift �Eg of 21meV. Although changesin the stress state of the SLED device occur due to thepresence of the tensile Ni and are known to result in Eg

shift,18) the energy shift in green InGaN/GaN LED struc-tures such as these is also very sensitive to biasing condi-tions.19,20) Owing to the inhomogeneous conduction ob-served in the SLED devices, the biasing conditions betweenthe two devices cannot be properly compared and the originof the observed �Eg is likely a combination of both effects.

In addition to the direct probing of the as-spalled surfacedescribed above, a portion of the SLED layer was bondedto a Si wafer using In. After bonding the as-spalled n-GaNsurface to Si, the polyimide tape was removed and the Nistressor and Ti adhesion layers were chemically etchedrevealing the original p-GaN surface. Ni/Au contacts wereevaporated; however, no annealing could be performed dueto the low melting point of the In bonding material. Figure 7shows the resulting current density versus bias of the bulkInGaN/GaN device with annealed Ni/Au contacts comparedto the In-bonded SLED device with as-deposited Ni/Aucontacts. The bonded SLED device showed an expectedincrease in parasitic resistance as well as a �200mVincrease in turn-on voltage due to the non-annealed contactlayers compared to the bulk device.

One of the most important differences between controlledspalling and other liftoff technologies is the generalityassociated with spalling. In other words, because there is noneed for specialized layers, transfer can be performed on anyoptimized epitaxial structure. Therefore, layer transfer of avariety of GaN-based structures (LED, laser, HEMT, etc.)grown on either sapphire, SiC, free-standing GaN, or Si isstraightforward. The other attractive aspect of layer transferby spalling is that it can be accomplished using conventionallaboratory equipment; no lasers or other specialized toolsare required.

Sapphire

u-GaN

n-GaN

MQWp-GaN

u-GaN

n-GaN

~ 3μm

(a) (b)

Fig. 4. Cross-sectional SEM image showing the InGaN/GaN device

structure (a) before and (b) after controlled spalling. The image has been

colorized to indicate the various regions of the structure. Approximately

3�m of the surface was removed, with fracture occurring near the center of

the n-GaN layer.

Ni

Ti

Spalled n-GaN

0.5 μm 100 nm

10 nm

(a)

(c)

(b)

Fig. 5. Cross-sectional TEM image showing (a) entire spalled layer

(b) high-resolution image of the MQW structure, and (c) image of the

10-period active device region. The few defects observed in the sample were

related to GaN growth suggesting that the material quality is preserved

during controlled spalling.

0

0.2

0.4

0.6

0.8

1

1.2

475 525 575 625 675Wavelength (nm)

No

rmal

ized

Inte

nsi

ty

Bulk

SLED

λp=529.2 nm (SLED)λp=533.9 nm (Bulk)

Fig. 6. Electroluminescence curves for the SLED structure and the

conventional (bulk) structure. The inset shows the green optical emission by

directly probing the exposed n-GaN surface of the SLED device. The

inhomogeneous emission is likely caused by localized conduction regions

due to the use of a non-ohmic Ti contact to p-GaN.

S. W. Bedell et al.Appl. Phys. Express 6 (2013) 112301

112301-3 # 2013 The Japan Society of Applied Physics

In summary, a previously described layer transfer methodcalled controlled spalling was used to separate a conven-tional InGaN/GaN MQW LED structure from a 50-mm-diameter sapphire wafer enabling the formation of a verticalspalled LED (SLED). A 25-�m-thick Ni layer possessing�400MPa tensile stress was electrodeposited on the surfaceof the wafer, followed by the application of a polyimidetape layer. By securing the wafer using a vacuum chuck,and pulling on the tape layer, a continuous 3-�m-thick layerof the LED epitaxy was removed. XTEM imaging indicatedthat spalling preserved the quality of the epitaxial layers,and electroluminescence verifies the operation of the SLED.An electroluminescence energy shift �Eg of 21meV wasobserved as compared to a conventional device and is likely

due to a combination of mechanical stress and dissimilarbiasing conditions.

Acknowledgments The authors would like to thank Drs. G. Shahidi and

T. C. Chen for their encouragement and support of this work.

1) S. Nakamura, M. Senoh, and T. Mukai: Appl. Phys. Lett. 62 (1993) 2390.

2) M. R. Krames, O. B. Shchekin, R. Mueller-Mach, G. O. Mueller, L. Zhou,

G. Harbers, and M. G. Craford: J. Disp. Technol. 3 (2007) 160.

3) W. S. Wong, T. Sands, and N. W. Cheung: Appl. Phys. Lett. 72 (1998)

599.

4) T. Fujii, Y. Gao, R. Sharma, E. L. Hu, S. P. DenBaars, and S. Nakamura:

Appl. Phys. Lett. 84 (2004) 855.

5) C. A. Tran, C.-F. Chu, C.-C. Cheng, W.-H. Liu, J.-Y. Chu, H.-C. Cheng,

F.-H. Fan, J.-K. Yen, and T. Doan: J. Cryst. Growth 298 (2007) 722.

6) K. Fujii, S. Lee, J.-S. Ha, H.-J. Lee, H.-J. Lee, S.-H. Lee, T. Kato, M.-W.

Cho, and T. Yao: Appl. Phys. Lett. 94 (2009) 242108.

7) J. Park, K. M. Song, S.-R. Jeon, J. H. Baek, and S.-W. Ryu: Appl. Phys.

Lett. 94 (2009) 221907.

8) D. J. Rogers, F. H. Teherani, A. Ougazzaden, S. Gautier, L. Divay, A.

Lusson, O. Durand, F. Wyczisk, G. Garry, T. Monteiro, M. R. Correira, M.

Peres, A. Neves, D. McGrouther, J. N. Chapman, and M. Razeghi: Appl.

Phys. Lett. 91 (2007) 071120.

9) C. Hsieh, H.-S. Chen, C.-H. Liao, C.-Y. Chen, C.-H. Lin, C.-H. Lin, S.-Y.

Ting, Y.-F. Yao, H.-T. Chen, Y.-W. Kiang, and C.-C. Yang: IEEE

Photonics Technol. Lett. 24 (2012) 1775.

10) Y. S. Wu, J.-H. Cheng, W. C. Peng, and H. Ouyang: Appl. Phys. Lett. 90

(2007) 251110.

11) S. W. Bedell, D. Shahrjerdi, B. Hekmatshoar, K. Fogel, P. Lauro, J. Ott, N.

Sosa, and D. Sadana: IEEE J. Photovoltaics 2 (2012) 141.

12) S. W. Bedell, K. Fogel, P. Lauro, D. Shahrjerdi, J. A. Ott, and D. Sadana:

J. Phys. D 46 (2013) 152002.

13) D. Shahrjerdi, S. W. Bedell, C. Bayram, C. C. Lubguban, K. Fogel, P.

Lauro, J. A. Ott, M. Hopstaken, M. Gayness, and D. Sadana: Adv. Energy

Mater. 3 (2013) 566.

14) D. Shahrjerdi and S. W. Bedell: Nano Lett. 13 (2013) 315.

15) M. D. Thouless, A. G. Evans, M. F. Ashby, and J. W. Hutchinson: Acta

Metall. 35 (1987) 1333.

16) Z. Suo and J. W. Hutchinson: Int. J. Solids Struct. 25 (1989) 1337.

17) M. D. Drory, J. W. Ager, T. Suski, I. Grzegory, and S. Porowski: Appl.

Phys. Lett. 69 (1996) 4044.

18) C. Kisielowski, J. Kruger, S. Ruvimov, T. Suski, J. W. Ager III, E. Jones,

Z. Liliental-Weber, M. Rubin, E. R. Weber, M. D. Bremser, and R. F.

Davis: Phys. Rev. B 54 (1996) 17745.

19) C. Bayram, J. L. Pau, R. McClintock, and M. Razeghi: Appl. Phys. B 95

(2009) 307.

20) S. Nakamura, T. Mukai, and M. Senoh: J. Appl. Phys. 76 (1994) 8189.

0

0.1

0.2

0.3

0.4

0 1 2 3 4 5Bias (V)

Cu

rren

t D

ensi

ty (

A/c

m2 )

PSS Si

GaN

Ni/Au (annealed)

InIn

Ni/Au (no anneal)

Bulk SLED (bonded)

SLED

Area = 1 mm2

Fig. 7. Current density versus bias for the bulk InGaN/GaN LED device

compared to a SLED device which has been bonded to a Si wafer. The as-

spalled n-GaN surface was bonded using In, and the tape and Ni layers

removed revealing the original p-GaN surface. Ni/Au contacts were

deposited but not annealed due to the low melting point of the In bond. The

data show an expected increase in parasitic resistance and turn-on voltage

due to the non-annealed contacts.

S. W. Bedell et al.Appl. Phys. Express 6 (2013) 112301

112301-4 # 2013 The Japan Society of Applied Physics